Cost Per Hour: Using a Time-Based Currency for Digital Advertising
•
25 likes•24,506 views
- Cost Per Hour (CPH) is a new proposed currency for digital advertising that values ad exposure time instead of just impressions. This allows publishers to distinguish high-value impressions that drive branding objectives like awareness. - Using predictive analytics, publishers can target impressions that provide more exposure time and sell them to brand advertisers as "hours" to be delivered in fewer impressions than typical campaigns. - Measurement tools like Chartbeat can track the active exposure time of ads on pages to support a CPH currency, accounting only for time when ads are visible and users are actively engaged.
Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (18)
Similar to Cost Per Hour: Using a Time-Based Currency for Digital Advertising
Similar to Cost Per Hour: Using a Time-Based Currency for Digital Advertising (20)
Recently uploaded
Recently uploaded (20)
Cost Per Hour: Using a Time-Based Currency for Digital Advertising
- 1. Cost Per Hour Using a Time-Based Currency for Digital Advertising Nikul Sanghvi, Analytics Consultant (London, UK) – May 2015 ABSTRACT An impression-based economy, for display advertising, values all ad impressions equally. Cost Per Hour (CPH) is a new trading currency that values ad exposure time as a dimension and provides a metric for trading purposes. Using CPH, publishers can distinguish and monetise those impressions that drive brand objectives, such as ad recall or brand awareness. Through the use of predictive analytics, publishers can specifically target these high-time-value impressions and sell them to brand advertisers as Hours. This targeting method enables a set amount of hours to be delivered in fewer impressions than an equivalent, non-targeted campaign. Deploying CPH requires organisational and operational change, as well as investment into ad technology and measurement systems. However, by using advertising inventory more efficiently across multiple trading currencies, a publisher creates potential for higher revenue and yield. In return, the brand advertiser receives a higher return on advertising spend, when compared to traditional measurements such as Cost Per Mille (CPM). INTRODUCTION The ‘Time-Based Selling’ Project started in March 2014, with two questions: For an ad that is in view, does exposure time have a positive relationship with branding outcomes? If the answer is yes, how should a publisher productise, price and sell time instead of impressions? This document is a summary of thoughts and findings in the process of creating a time-based currency at the Financial Times. ‘Cost Per Hour’ was designed to enable the digital advertising industry to adopt an attention-driven economy in addition to the current volume based model. As a time-based trading method, it values the dimension of time as an effective advertising metric. To date, a beta-version of the product has been sold to over 10 major clients, including Microsoft and BP1 , generating over $1m of incremental revenue for the Financial Times. Within the following pages, key questions will be addressed on subjects such as product design, pricing methodology and operational workflow. The focus is on providing practical solutions to catalyse a movement for publishers (and media owners) towards attention based advertising metrics. The included content is provided under Creative Commons (Attribution 4.0 International License) with the hope that other marketing practitioners will share, adapt and improve the discussed ideas. ACKNOWLEDGEMENTS FROM THE AUTHOR I would like to thank the Time-Based selling project team at the Financial Times - Tom Ward, Alistair Smith, Jon Slade, Anthony Hitchings, Danny Aldred, Isa Wills and Victoria Morgan-Smith. In addition, this body of work would not have been possible without the support and helpful conversations from Dominic Good (Financial Times), Henry Rowe (FaR Research Partners), Tony Haile (Chartbeat), Alex Carusillo (Chartbeat), Josh Schwartz (Chartbeat), David van Dokkum (Chartbeat), Daniel G. Goldstein (Yahoo Research) and Siddharth Suri (Yahoo Research). Cost Per Hour (CPH) was designed and developed by: Nikul Sanghvi, Analytics Consultant – nikul@adyields.com Tom Ward, Financial Times – tom.ward@ft.com Alistair Smith, Financial Times – alistair.smith@ft.com 1 http://adage.com/article/digital/microsoft-bp-bought-digital-ads-charge-hour/296363/
- 2. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 2 BACKGROUND In 2014, global spend on digital display advertising was estimated to be $52 billion 2 . Cost Per Mille (Cost Per Thousand) is the primary trading currency used for buying and selling within this economy. CPM is based on the measurement and pricing of digital ‘impressions’, a metric that has been around for more than two decades. Along with the CPC (Cost Per Click), CPM has become one of the defining KPIs of the digital advertising industry. More recently, questions have been raised concerning both the validity and value of these metrics especially as a proxy for something that brand marketers crave – attention. A standard impression is historically a machine metric and does not have a strong relationship to human consumption or exposure. The Click Through Rate (CTR) is also commonly used to measure ad performance but this metric has been considered erratic. Amongst others, Hewlett-Packard researchers, Suzanne Leighton and Sibel Satiroglu, have found that click through rates are a highly random, unreliable measure of effectiveness. In April 2015, DoubleClick (the display advertising arm of Google) calculated that overall display CTRs across all ad formats and placements was 0.06% - less than one click out of 1000 impressions3 . As the display advertising industry evolved, the website design and layout developed around a monetisation method that valued quantity over quality. The pursuit of better financial yields via CPM and CPC, has led to practices that influence business strategy, user experience, site design and content. Some tactics originate from the early days of the internet and many of these practices persist to date. Below are several well known examples: -‐ Pop up ads -‐ Dividing a single text-based article over multiple pages -‐ Ad units designed as content in order deceive and generate clicks -‐ Ad units that are designed to be clicked on or tapped by accident -‐ Image slideshows that call a new page URL and thus generate additional impressions -‐ Interstitial pages where isolated ads are displayed before a user can proceed -‐ Content that is purposely written to minimise length/attention -‐ Articles and content created in order to function as ‘click-bait’ -‐ Excessive page surface-area dedicated to pushing users to additional pages -‐ Websites where the combined surface-area of ad units exceeds that of consumable content The position of CPM as the primary trading currency means that all publishers must play a ‘volume game’ in order to maximise revenue. Media owners that have huge traffic numbers, and/or those that artificially boost page impressions to serve more ads will always dominate an impression-based economy. This provides a competitive dis-advantage for publishers who prioritise high quality, in-depth journalism and/or focus on niche audiences. The CPM model does not take into account the engagement or participation that loyal readers produce. This has a knock-on effect for brand advertisers, who often pay for their ads to be displayed in sub-optimal circumstances, generating low attention scores per pound or dollar spent. In this scenario, all impressions within a selected environment are valued equally – and there is no emphasis placed on impressions that are seen for 20 seconds, as opposed to two seconds (or those not seen at all). To counteract many of these challenges, the proposed goals of a Digital Attention Economy can be summarised as a: -‐ pricing system that factors in a dimension which has a human context (such as time) -‐ level playing field for publishers, that does not favour volume alone 4 -‐ better approach for ROI on ad spend for campaigns where key objectives are brand related -‐ monetisation strategy that aligns to editorial goals and leads to a better user experience for readers 2 http://www.zenithoptimedia.com/global-adspend-set-to-return-to-pre-financial-crisis-growth-rates/ 3 http://www.smartinsights.com/internet-advertising/internet-advertising-analytics/display-advertising- clickthrough-rates/ 4 http://digiday.com/publishers/publishers-want-attention-replace-impression/
- 3. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 3 TIME-BASED MEASUREMENT TOOLS Traditional web analytics tools measure ‘time-on-page’, calculating the gaps between timestamps on page loads. This can lead to problems such as instances where only one page is viewed – and time cannot be calculated. These tools also consider the page as a whole rather than a particular ad unit. Chartbeat5 and Moat6 are tools that provide analytics on time performance of individual ad units. Conversations with Chartbeat started with examining data from its Editorial product. One such insight was that Financial Times had six times higher average time-on-site than Chartbeat’s industry benchmark for news sites. The image above shows the Chartbeat debugging-tool in action on the Financial Times homepage, measuring an ad in view Chartbeat (for Ads) has a unique feature is named Active Exposure Time. This allows the tool to measure an ad, only whilst it is in view and the user is active. It does this using a JavaScript tag that is placed on the page, along with HTML5 code on individual ad containers. Chartbeat will then continuously (every hundred milliseconds) check for three things… - ‘Is more than 50% of the ad in view (displayed in the browser window)?’ - ‘Over the last five seconds, has the user shown any signs of activity? – mouse movement, scroll, key press etc. - ‘What are the display properties of the browser / device such as window dimensions?’ The timer doesn’t start until the ad is in view. It pauses when the browser tab is changed or the ad goes out of the viewable screen. After five seconds of inactivity, it also pauses measurement. Chartbeat have openly shared the methodology behind this function and are encouraging other data companies to adopt it7 . The use of Active Exposure forms a methodology where advertisers only pay for time when the ad is at least 50% in view and the user is actively at their screen. This solves cost-pollution such as an ad passively sitting on screen for seven minutes whilst a user has gone to make a cup of tea. In brings ad exposure as close to attention as possible, in the absence of knowing precisely whether or not a user is looking at the screen. The successful implementation of a measurement tool (with Media Rating Council accredited time- metrics) is the starting point for any publisher that is looking towards an attention-based currency8 . This may be in addition to or in replacement of selling viewable impressions. It is worth noting that the methodology within this paper can also be applied to Native Content, such as Advertorials, along with container-based display ads. 5 https://chartbeat.com 6 http://www.moat.com/moat_analytics 7 https://chartbeat.com/public-release-methodology/ 8 http://www.mediaratingcouncil.org/Accredited%20Services.htm
- 4. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 4 VIEWABILITY & SELLING ‘MINIMUM TIME’ IMPRESSIONS Viewability is incredibly important – as a concept and as a standard9 . Viewability benchmarks have shown that up to 54% of ads are not seen by humans10 . The major causes are cited as ads being loaded out of view (below the fold), ads not loading in time and ad fraud. Just because an ad has been served, it does not mean that it has been viewed. The MRC and Internet Advertising Bureau (iAB) define a viewable impression at least 50% of the ad surface area having been visible for at least one second11 . The below image shows the basic calculations used in measuring Viewability rates. Viewability is a great starting point but doesn’t go far enough to place value on premium advertising inventory. Although it promotes efficiency for buyers, who previously used half their ad spend on unseen impressions, it still values a one second impression the same as a five second impression. After deciding on a tool to use for measuring Viewability, publishers who choose to sell viewable impressions as a minimum standard will be faced with three immediate questions: - What is our overall Viewability rate? - What impact does a contractual minimum-viewability rate (e.g. 70%) have on delivery? - Knowing our Viewability scores, how should we price viewable impressions? 9 http://www.thedrum.com/news/2014/03/12/advertisers-are-wasting-spend-unseen-ads-viewability-problems- need-be-tackled-report 10 http://www.comscore.com/Insights/Blog/Viewability-Benchmarks-Show-Many-Ads-Are-Not-In-View-but- Rates-Vary-by-Publisher 11 http://www.iab.net/media/file/MRC-Viewable-Ad-Impression-Measurement-Guideline.pdf
- 5. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 5 The simplest and most common answer leads to a symmetric solution for the last question - a pricing equilibrium that is often hidden from the buyer12 . To price viewable impressions and retain the same net revenue, the seller must: 1) Measure and calculate the viewability rate of the targeted inventory e.g. 50% 2) Multiply the CPM of the same inventory by reversing the ratio e.g. double the price Here is an example deal where viewable impressions are sold by a publisher: - 1000 impressions cost £20 and 50% of these impressions are viewable. - The vCPM (Viewable CPM) is now increased to £40 and 2000 impressions are served to reach the target. - Thus the net cost to the advertiser for the viewable impressions has not changed. Time is a natural attribute of user engagement – advertisers already receive it even if they purchase the ads via CPM. Increasing the minimum exposure time for these viewable impressions (2, 3, 4 or 5 seconds…) has the same outcome. An identical scenario is to be expected for these longer impressions but with differing ratios. It isn’t possible to exclusively show selected campaigns to users who view an ad slot for five or more seconds (unless an ad-server comes with an integrated time machine). In order to reach a higher volume of users above five seconds, there will also be a volume of users who view the advert for less than five seconds. It initially seems that it is not possible to sell a ‘minimum-time’ currency such as a ‘Five Second Minimum Guarantee’ without either making less money as a publisher or placing the buyer at a price/efficiency disadvantage. The diagram below shows that (for a similar inventory group) both purchases share the same ratio for impressions that are greater than (or equal to) five seconds and those that are less than five seconds. 12 http://adexchanger.com/data-driven-thinking/introducing-vcpm-the-right-way-to-think-about-viewability/
- 6. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 6 CREATING A TIME-BASED CURRENCY - COST PER HOUR To build a product such as a new trading currency, it is useful to lay down a micro-manifesto to serve as a philosophical radar in all activities that follow. These four ‘cornerstones’ help to shape the product and in times of dispute or uncertainty, can become a point of reference: 1) In data we trust: the product direction must be driven by insights from data and analytics 2) A naked product is honest: the product needs to be as transparent as possible 3) Content is king: nothing matters more than a publisher’s audience engaging with its content 4) Pound for pound: a client must not be in ‘attention’ deficit as a result of buying time over CPM Once the preferred analytics tool has been deployed and suitable data collected, it is possible to extract a table showing impressions broken down at second by second increments. The data can then be grouped into segments for intervals such as ‘More than five seconds’ and ‘Equal to or less than 10 seconds’. Furthermore, this can be extended to include time-segmented data for dimensions such as page name or site section. Simple analysis (of the FT.com desktop site) showed that 20% of impressions contained 80% of the total available time. True to industry averages13 , just under half of all impressions had no time value at all (these are sub-second impressions). Using this data to increase the CPM for time-based impressions results in a product that resembles Viewable CPM (vCPM). To value time itself, rather than impressions that contain time, it is necessary to calculate pricing for a block of time with a desired quality level. These blocks (or buckets) are composed of one hour as the default size (60 minutes or 3600 seconds) - resulting as ‘Cost Per Hour’, which is abbreviated as CPH. The challenge is to understand how to price an hour of this time… using the current CPM rates as a reference point. 13 https://think.storage.googleapis.com/docs/the-importance-of-being-seen_study.pdf
- 7. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 7 CPH PRICING FORMULA To understand how to price time, the starting point was to identify it’s relationship with the impression. The CPH formula is a simple calculation that determines how much time 1000 impressions create. This ‘speed’ is then used to create a multiplier, which can be applied to the price of those 1000 impressions – in order to understand the price of each hour generated by them. Below is the method for pricing Time (in relation to CPM), using a one hour bucket, composed only of impressions that are a specified number of seconds or longer. Use the a multiplier to generate revenue consistent with CPM = 𝒚𝑪𝑷𝑴 Formula: 𝒚 = 𝟏 / (𝒗 / 𝒕) Therefore: 𝒚 = 𝟏 / ((𝟏𝟎𝟎𝟎 ∗ 𝒏) / (𝟑𝟔𝟎𝟎/𝒓)) The symbols are explained in more detail… 𝒂 = number of minimum seconds per exposure required for qualification into time block 𝒏 = % of impressions that are 𝒂 seconds or longer 𝒓 = Average (mean) time (in seconds) of impressions that are 𝒂 seconds or longer 𝒕 = How many 𝒓 values can be fitted into a 1 hour bucket = 𝟑𝟔𝟎𝟎 / 𝒓 𝒗 = Out of 1000 impressions served, how many will be viewable impressions longer than 𝒂 seconds = 𝟏𝟎𝟎𝟎 ∗ 𝒏 𝒙 = Using only impressions which are longer than 𝒂 seconds, how many hours are available from 1000 impressions? = 𝒗 / 𝒕 𝒚 = The CPH multiplier = 𝟏 / 𝒙 - - - - - If the goal is to retain consistent revenue for a volume of impressions, the multiplier will always increase as the minimum qualification time is raised. a = number of seconds (per exposure) required for qualification 𝒚 = The CPH multiplier = 𝟏 / 𝒙 This means that to sell using a particular minimum qualifying Time - and aim to make similar revenue to selling the same inventory on CPM, the business must multiply the CPM rate-card using a predetermined multiplier. This now functions as an exchange rate between CPM and CPH. As the 𝒂 value increases, so does 𝒚 It is optional for the business to apply a discount onto 𝒚 in order to incentivise the buy-side. However, this produces lower yields, when compared to selling via CPM.
- 8. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 8 Processing for variants of 𝒂 will result in varying outcomes dependant on the characteristics of the publisher, its audience and their consumption of its content. At the Financial Times, there were multiple reasons that ‘five seconds’ was selected as the minimum number of seconds required for qualification into the time bucket… -‐ When 𝒂 is five seconds, the value for 𝒓 surpassed the 20 seconds mark. This is important as many luxury and financial ads are designed with 20 second rotations. (Some do not show the company logo until the last few seconds of the rotation) -‐ Research against test campaigns showed satisfactory uplifts in ad recall and brand awareness for impressions above 5s vs. below 5s 14 -‐ Using a divisor of 60 (such as 2, 3, 4, 5, 6) allowed division into 3600 seconds without going into decimals or having remainders, making calculations simpler. -‐ Beyond 10 seconds, the CPH multiplier started to scale up exponentially as the available volume of qualifying impressions decreased. -‐ 5 seconds represented a significant increase over the iAB one-second minimum. -‐ As a number, it is optically aesthetic for marketing purposes. Returning to the formula, it is possible to demonstrate values for 𝒏 and 𝒓 where the value for 𝒂 is 5 seconds. 𝒚 = 𝟏 / (𝒗 / 𝒕) Therefore: 𝒚 = 𝟏 / ((𝟏𝟎𝟎𝟎 ∗ 𝒏) / (𝟑𝟔𝟎𝟎/𝒓)) Using Chartbeat data, it was calculated that when 𝒂 is 5 seconds… 𝒏 = 0.31 (or 31% of ads are seen for more than 5 seconds) 𝒓 = 21.9 seconds (is the average time a 5+ second ad will be seen for) Formula: 𝒚 = 𝟏 / ((𝟏𝟎𝟎𝟎 ∗ 𝒏) / (𝟑𝟔𝟎𝟎/𝒓)) Therefore 𝒚 = 𝟏 / ((𝟏𝟎𝟎𝟎 ∗ 𝟎. 𝟑𝟏) / (𝟑𝟔𝟎𝟎/𝟐𝟏. 𝟗)) Therefore 𝒚 = 𝟏 / (𝟑𝟏𝟎 / 𝟏𝟔𝟒. 𝟑𝟖) Therefore 𝒚 = 𝟏 / 𝟏. 𝟖𝟖𝟓 Therefore 𝒚 = 𝟎. 𝟓𝟑 We will now apply the CPH multiplier (𝒚) where, as an example CPM = £50 £𝟓𝟎 ∗ 𝟎. 𝟓𝟑 = £𝟐𝟔. 𝟓𝟎 To sell a one-hour block using CPH (and make the same amount of net revenue as CPM), inventory must be sold at £26.50 per hour instead of £50 per CPM This formula will enable a publisher to understand the comparative time value of inventory. By integrating the new currency into yield analysis, a business can determine the segments of its inventory that can be monetised more efficiently against CPH than CPM, vCPM or CPC. The complexity of this task will be determined by the range of products sold by the publisher and the availability of data that reflects the Effective CPM and Sell Through Rates. 14 http://blog.chartbeat.com/2013/05/20/how-engaged-time-affects-brand-recall/
- 9. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 9 THE CHALLENGES OF A PREDICTIVE CURRENCY Current ad serving tools such as Google’s DoubleClick For Publishers (DFP)15 use impressions as the base metric. This means that all bookings are made and delivered using an ‘impression’ as the lowest denominator of delivery. This can be labelled as a Conclusive system. A thousand booked impressions can be sold, measured through to delivery and the campaign can be subsequently concluded. The greatest challenge with creating and deploying a time-based currency is that the current tools and systems do not permit the booking and capping of ad delivery to be measured on seconds rather than impressions. For example, it is not possible for DFP to automatically stop serving an advert after 1000 hours are viewed. This poses a risk to the net margin. Instead, using the formulas provided in the above slides, Ad Ops teams must initially calculate and predict the number of impressions that will the booked in order to serve the volume of sold Time. Selling on time is therefore a ‘Predictive’ currency (in a similar way, so is Viewability). The client will purchase ‘Time’ as a unit – rather than impressions. The table below summarises the key difference between the types of trading currency… Currency Type Conclusive [CPM] Predictive [CPH] Delivery Systems Supported Not available Measurement Systems Ad Server + Third Party Third Party only Key Metrics Impressions and Clicks Seconds (and Hours) Current Process Sold, booked and served as impressions Sold as hours. Booked and served as impressions. This predictive process has a variety of problems as calculations rely heavily on two main variables… 𝒏 = Percentage of impressions that are 𝒂 seconds or longer 𝒓 = Average (mean) time of impressions that are 𝒂 seconds or longer These two variables are used to drive the predictions and rely on historical data. So what happens when these variables do not have the predicted values during campaign execution? 15 http://www.google.com/doubleclick/publishers/
- 10. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 10 FAST AND SLOW TIME BUCKETS There is a common relationship between 𝒏 and 𝒓 - they can both increase and decrease based on the user behaviour of the delivery page or the chosen container position. The example below shows how these variables might impact a bucket and its fill speed. - - - - - - The below time bucket represents one-hour (3600 seconds) of Time sold. - The height of the bucket represents how many impressions are required to fill it. - The smaller the bucket, the higher the eCPM for the client The following events will impact both 𝒏 and 𝒓 - Type of page – homepage, hub page, article page - Media on page – written content, videos, images, comments section - Content – article length, author, subject area - Traffic sources – which also impacts type of visitors on site (registered/subscribers vs anonymous) - User engagement – bounce rates, dwell time - Ad creative size – dimension, orientation and pixel volume (surface area) - Ad position - container location within page - Backend and client technology – page load speeds, ad load speeds, device connection speed - Client-side platform – device type, product type (desktop or mobile), browser - Target audience – geography, cohort type, age, gender If time is sold against specifically targeted segments – the absence of an automated system would makes these detailed predictions a resource intensive process. An ideal system would use historical data to calculate, weight and predict 𝒏 and 𝒓 – but against a range of these dimensions and their characteristics.
- 11. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 11 Predicting these values requires the use of a bucket speed metric that measures the flow of time in relation to the volume of impressions served. The element x from the original formula reflects the hours per thousand impressions (HPM). The HPM is a crucial metric for projections relating to targeting and delivery speed. HPM also known as (𝒙 = 𝒗 / 𝒕) allows us to calculate: ‘How many hours are available from 1000 impressions, using only impressions which are longer than 𝒂 seconds?’ On a site-wide ad serving method, the balancing act of delivering a fixed HPM-speed makes it difficult to ensure that a client is receiving true value for money. The pricing algorithm creates a price that is initially on par with CPM. If the bucket speed is too fast, it gets filled up by fewer impressions and the client may have been better off purchasing via CPM to get the same outcome. If the bucket fills up too slow, the publisher is at a disadvantage, as they would have made a higher yield selling the larger volume of impressions via CPM instead of CPH. With this pricing formula, the value of time can be compared against CPM. However, its application within a day-to-day business context would still result in reaching a revenue equilibrium. Alone, it offers no commercial advantage over a pure CPM strategy. It would be inefficient to perform extensive calculations and operational adjustments to periodically rebalance the HPM of a campaign to achieve parity with its CPM-purchased equivalent. Furthermore, the publisher is at constant risk of providing a sub-optimal return to the buyer, conflicting with Cornerstone#4. In order to provide value to all parties in every transaction, the role of optimisation must be pivoted – changing it from an act of balancing the HPM into one of maximising HPM.
- 12. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 12 BUYER ADVANTAGE – Better brand outcomes via eCPH Returning to the original goal of selling time, media owners must plan ahead before adopting such a currency. Without understanding why time is important for certain campaigns, it is difficult or potentially misleading to convince a buyer to purchase through this method. The deciding factor resides in the obvious statement that ‘not all marketing campaigns have the same objectives’. Some campaigns may concentrate on lead-generation, others reach or volume. Although additional research is required to quantify the correlation between time and these traditional metrics, there is evidence to show that, for campaigns that have a branding objective, time is a measure of quality whereas impressions reflect on a quantity served. If eCPH is the core KPI of a campaign, the emphasis of eCPM and eCPC must be deprioritised16 . Once this is accepted, a publisher can focus on delivering a set ROI through eCPH without the limits that conflicting targets impose. Time-centric optimisation (via predicative targeting) can then evolve to become a process used to deliver the maximum value per pound spent on generating attention. If a publisher can accurately identify and precisely target high-HPM inventory, it can then attribute a portion of this increased yield back to the advertisers. An example is used below… A run-of-site branding campaign is purchased at £10 for 1000 impressions. These 1000 impressions generate an average of 30 minutes worth of time. Therefore, the ‘Effective Cost Per Hour’ is £20. In this example, it is the time which matters. These 30 minutes are vital to the campaigns success and its related performance on metrics such as Awareness and Brand Recall. The eCPH is the scoring KPI that is important here, not the fact that the publisher has delivered this campaign through 1000 impressions. When the publisher can predict and target impressions that are likely to be longer than the minimum qualifying time, HPM optimisation offers a pivotal advantage. If the publisher can then specifically target these impressions (e.g. those 5 seconds or longer), it would be able to deliver a much higher HPM and thus a lower eCPH. Returning to the example… What if the same publisher could now generate 40 minutes with 1000 impressions (still priced at £10), dropping the eCPH to £15? In either instance, the client still spends £10 – their budget does not change. However, through the optimised CPH delivery, they receive a higher density of time in comparison to spending that same £10 on a CPM campaign. In our example, the client can potentially receive 33% more time for the same amount of money. This lowers the eCPH by 25%. 16 http://adage.com/article/media/digital-advertising-ready-ditch-click/295143/
- 13. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 13 The role of educating clients and opening up discussion is not to be underestimated here. Publishers and commercial teams must focus on which KPIs really matter to their clients, especially those that engage in direct trading with long term clients. The goal has always to help them drive better business outcomes. The initial task may be convincing Brand Marketers that impressions have become a proxy to measure success but don’t really reflect value. Concentrating on delivering a higher volume of time against each impression means that a publisher’s primary focus isn’t on how many impressions are served. Extensive debate and discussion is required, with a view to shifting thinking within the advertising industry. Only then will eCPM cease to be the target or KPI that reflects the success of a brand campaign. In Q4 2014, the FT ran five test campaigns using technology from an independent research company. 17 Each of the five campaigns was treated as an advertising effectiveness study. Using Chartbeat, the business was able to identify if a user had been exposed to a campaign, and if they had been exposed, how long had they been actively engaged with the digital ad. The surveys were designed to measure the impact that active exposure time has on ad recall, brand awareness, brand association and brand consideration. The results of the surveys were combined to provide a sample size of 1688 respondents. These respondents were split into three time intervals – not seen, seen for less than five seconds and seen for five seconds or more. When comparing ads seen for five or more seconds with ads seen for less than five seconds, all metrics showed significant improvement: Ad Recall (+79%), Brand Familiarity (+55%), Brand Association (+51%) and Brand Consideration (+58%). The data also showed that within the first 60 seconds, these results continued to increase the longer that the ad was seen.18 17 http://www.thedrum.com/news/2014/05/22/financial-times-kicks-trials-sell-advertisers-blocks-time-tackle- industry-s 18 Alistair Smith, Financial Times Digital Attention Economy Study 2014
- 14. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 14 PUBLISHER ADVANTAGE – Inventory Efficiency and Market Share Now that the advantage to the advertiser has been established, there is the question of how a publisher may also benefit from adopting a time-based trading methodology. Although the volume of impressions needs to be managed, the goal of CPH is to use premium time inventory more efficiently. By having an optimisation method to increase HPM, it is possible to serve out an hour with less impressions than a standard CPH campaign. Naturally, this results in a higher eCPM for the client. Campaign planning centres on delivering a higher calibre of impressions rather than an exact quantity of random-quality impressions. A targeting method would allow the business to give the client greater time-value on CPH buys whilst also reducing the number of impressions required. The below table shows the differences in product offering between the two buy types: If a publisher could optimise to +30% extra time, it could split this bonus with the client. The client can take 25% extra time (in comparison to CPM) and in exchange, they allow the publisher to deliver the campaign in fewer impressions than the CPM buy. eCPM is no longer treated as the champion efficiency-metric for the brand campaign, as the priority is on driving brand awareness or recall.
- 15. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 15 The buyer should always be made aware that a CPH-trading publisher will not optimise CPH campaigns against non-time-based metrics. The methodology of processing time into 1 hour buckets also needs to ensure the exclusion of excessive outliers and erroneous exposure values. This is due diligence for removing anything that is unnatural or potentially fraudulent. The CPH product is designed to fit in with current client budgets without inflating total spend through the excessive use of a multiplier. With the same budget, a client focused on maximising Attention can now receive higher eCPH value per pound spent. The client does not get invoiced for any of the impressions that are lower than the qualifying time of 5 seconds, so there is an incentive on reducing these. Within reason, it is also possible to deliver the total time in fewer qualifying impressions. The publisher should decide on boundaries that it wishes to operate within, to allow for improved yields whilst not abusing the eCPM rate through low volume delivery. Although eCPM is not the primary KPI, it is still widely accepted as the industry standard for trading. Any deal completed through eCPH will still have buyers that also monitor how the eCPM has performed. Although they may accept a higher eCPM rate in exchange for a lower eCPH rate, there will be limits as to how high the equivalent eCPM can be pushed. These limits will vary depending on the buyer and it would be prudent to discuss this and agree on the acceptable discrepancies in advance. Throughout beta-testing, the Financial Times has been able to deliver a minimum of 20% extra time in an average of 12% fewer impressions. Although there is additional revenue potential through incremental sales, there is not any direct monetary gain from the delivery alone. To deliver maximum yield through a combination of CPM and CPH, a publisher must be aiming for a higher sell through rate. The remaining impressions should be monetised via CPM or CPH in order for the business to increase its total revenue; otherwise they simply become remnant stock at no additional revenue gain to the business. On top of inventory efficiency, a publisher may also be able to gain a greater share of the advertisers budget. By selling a transparent initiative that improves return on investment for brand-spend, the publisher can aim for higher order values. If the value of a time-based currency can be shown to the client, they may willingly increase their total spend with a publisher. This is beneficial to both parties but allows the publisher to improve market share. This revenue-focused publisher objective can be openly stated to the advertiser as part of the sales process. As impressions with high HPM value are isolated or retained to be sold as CPH inventory, the average time available through CPM (via the remaining impressions) will be reduced. The scale of reduction will be dependent on the proficiency of targeting and overall order volume split between the two currencies. Premium publishers with highly engaged audiences will still be able to offer above-industry-average eCPH performance on CPM campaigns. And for clients who value reach over brand recall, CPM will still be an effective purchase. The relationship between time and lead-generation, in the form of CPC and CTA, will require continued research. There may be correlation between impression exposure time and click through rates, but this still needs to be framed against CPM within the scope of ROAS (return on advertising spend). Research also needs to be continued into the diminishing returns that time may have beyond a certain point, such as 60 seconds. This opens up a potential solution for a time-based, refreshing slot hybrid that rotates the creative for optimal outcome efficiency. CPH must be positioned so as not to cannibalise or compete with CPM deals, but to supplement them against the appropriate campaign objectives.
- 16. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 16 TARGETING AND PREDICTION In its simplest form, the CPH prediction system scores a multitude of targeting variables for their propensity to produce impressions above and below the desired qualification value. Using an example of 𝒂 being equal to five seconds, we can put this into the analogy of a deck of cards. The deck of cards is shuffled and randomly split into five fixed piles. Black cards (spades and clubs) represent impressions that are five seconds or longer. Red cards (hearts and diamonds) represent impressions that are shorter than five seconds. A prediction system would look at all of the cards within each of these piles. It would then calculate the probability, (for a random card, selected from a given pile) to be a black card. To serve CPH with greater efficiency, the prediction engine would need to be able to tell which pile has the highest probability of surfacing as many black cards as possible (five seconds or longer impressions). In this scenario, there are only five different piles for the system to read. In practice, this extends into the thousands.
- 17. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 17 A potential set of targeting dimensions for a website is shown below with an example number of values that each dimension contains. Dimension Number Of Values Example Values Day of Week 7 Mon, Tues, Wed, etc. Hour of Day 24 00:00, 01:00, 02:00… 23:00 Site Section 20+ homepage, news, technology, world, business… Geographic Region 7+ (continents but could also be countries or cities) Ad Creative Size 5+ 300x250, 300x600, 728x90… Traffic Source 10+ Direct, Search Engine, Twitter, Facebook, Email Marketing, Drudge… Using these 6 targeting dimensions alone, the prediction system needs to forecast for 1,176,000 combinations (7 x 24 x 20 x 7 x 5 x 10) – or using the example, 1.18m random piles of playing cards. Furthermore, audience demographic could offer between a handful and several hundred combinations, depending on a combination of first-party data and segments from a Data Management Platform (DMP). Every dimensional layer or additional level of detail added to the Optimisation Schedule increases the processing required. Running the set above for the previous 90 days involves processing up to 106m rows of data per calculation. Buyer-driven targeting reduces the number of attributes that can be used to optimise for time efficiency. For example, if a client wants to buy time only for “728x90 in Europe”, it is not possible to use efficiency gains within the Ad Size or the Region to deliver the time at a higher HPM. This reduces the maximum potential efficiency gain. Out of the list of dimensions, some have a higher tendency to be preselected by the buy-side. The most common buy-side targeting options for premium publishers tend to be Region, Audience Demographics, Site Section (via Sponsorships) and DMP segments. The remaining dimensions (such as Time of Day, Day of Week, Traffic Source) remain relatively open to optimisation. In combination with site section and DMP segments, these non-buyer-driven dimensions tend to provide the simplest possible gains for CPH yield optimisation. In return for purchasing via an optimised CPH delivery, they advertiser will receive: 1) A higher total viewability rate (across all impressions) as a default property of a optimised CPH delivery 2) A higher average number of seconds per impression compared to the site average 3) A billing system that has only invoiced for 100% (iAB standard) viewable impressions that are also equal to or above the qualification time. 4) A currency that only values impressions where the ad is actively engaged
- 18. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 18 The targeting accuracy and optimisation capabilities of a publisher will facilitate its market proposition by determining how much discount it can potentially offer to a buyer. For example, if a publisher can target with a +25% HPM efficiency, it will gain 15 minutes on every hour that it delivers. It is now the decision of the publisher as to how much of this value to give back to the client. This bonus time works as a form of price discounting via added value. The publisher may choose to: A) Give the client all 25% bonus time in favour of building a longer term relationship or increasing overall order volumes from the client B) Retain part of the growth as impressions, giving the client the remaining part as bonus time C) Retain all of the gains The publisher may also be able to use monetisation methods such as tiered spending, giving away greater bonus time for larger orders. This would incentivise clients to increase their booking order size. The incentive could also be dynamically mapped to supply and demand, giving away greater bonus time during low yield periods or sections of undersold inventory. As with CPM, many factors can impact the ability for a publisher to successfully deliver a CPH campaign. Examples include the fluctuation of site traffic such as an unexpected spike or a sudden drop. The absence of a fully automated, sophisticated prediction systems as market-ready software, means that early prediction systems may need be built in house and to match bespoke requirements. The Financial Times has currently created its prediction engine using data extracted from Chartbeat and an off-the-shelf BI tool. When using recent data (representing current site design), it can forecast HPM speeds for combined dimensions with a +/- 2% accuracy (an example target being a 300x250 creative served to Senior Business Decision Makers in United States). As the data extraction and processing becomes closer to real-time, the accuracy of overall forecasting should also improve. As with any prediction that relies on historical data, the greater the range of data, the easier it will also be to factor in seasonal eventualities and fluctuations. The historical data used for targeting can also be used to quantify the equivalent number of hours that the client would receive if they purchased via CPM. The exposure data can be independent of the trading currency and provides HPM values for all historical circumstances. It can also be segmented by Campaign to identify overall performance variations for CPH vs. CPM. The basic process for forecasting before a campaign is… A) Understand the client/advertiser variable such as budget, desired outcomes, KPIs and targeting requirements (demo, region, etc.) B) Identify campaign structure – sponsorship, wrapped, ad units, delivery duration, etc.? C) Use the prediction engine to quantify the volume of hours (and HPM) that client currently gets from all impressions – based on historical average and budget. What is the eCPH? D) Calculate the maximum potential HPM that can be delivered with the same set of targeting options. If this exceeds the original HPM, work out the bonus that can be shared with the client (if any). E) Add bonus onto the total delivery time. What is the new target eCPH? F) Identify the exact optimisation schedule that will meet the bonus-inclusive target from impressions that qualify. How many impressions will be required and where will they need to be served? How will they be targeted to increase the overall HPM? G) Check inventory availability against ad server and DMP for the optimised schedule and plan for delivery (contracts, invoicing and booking)
- 19. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 19 USING AN EXCHANGE RATE The formula contained within this whitepaper needs to be run forwards and backwards at multiple points during the end-to-end campaign. It serves to function as an ‘Exchange Rate’ between the two different currencies. The table below shows the different directions of the exchange rate in relation to the known data. There are four main conversion types: Each of these four conversion types is used at a different stage of the operational workflow. Below are the key steps involved within the end-to-end process for a team that books direct orders. They may not always exist in an exact linear order and can also occur simultaneously. The relevant dataset is shown on the far left and exchange on the right. The numbers indicate the conversion type listed on the previous table. A tick mark denotes a step where a ‘true’ or desired value exists. ‘Calc’ refers to a stage where a metric must be calculated from the other.
- 20. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 20 From an operational perspective, these exchange points are necessary in order to answer the key questions that arise along the way. Using the example where 𝑎 = 5, the team will have to sense-check: During Prediction: - How many impressions are required to serve the target volume of hours? - Out of 1000 impressions – how many will be more than or equal to 5 seconds? - What is the average (mean) number of seconds per qualifying impression? - What is the overall Impressions per Hour? - What is the CPM price for this volume of impressions? - Based on above, how much should the business charge for the target volume of hours? - Can this volume of impressions and time be served during the required delivery timeframe? During Run-Time (in-flight): - How many impressions were served? - How many hours at >=5s have been served from these impressions? - How many hours are remaining before the goal target is met? - What is the rate of qualifying (>=5s) impressions per hour? - What is the rate of total impressions per hour? - If continuing at this rate, will delivery meet the target number of hours? - Are there enough remaining impressions to meet the target number of hours? - Is it necessary to adjust the volume of booked impressions to meet this target? - Is it necessary to change any of the targeting options to optimise HPM? - When can the campaign be terminated? And finally, post-campaign: - How did we perform? - Was the multiplier and projection accurate? If not, how did it differ? - If not, what happened and how can we improve for the future? - How can we use the multiplier and CPM data to improve future forecasting? - Was there any gap between server impressions booked and impressions measured? Asking these questions will allow the team to improve its workflow and highlight any issues, which could result in inaccurate forecasting, over/under-deliver and mispricing.
- 21. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 21 For Run-Time (when campaigns are live), the Financial Times worked closely with Chartbeat to build the formula and methodology into a ‘Clock’. The clock is a delivery dashboard that measures live campaigns and provides data in real-time for metrics such as number of campaign hours delivered, impression efficiency (impressions per hour of exposure time – the inverse of HPM) and percentage pace against a predicted ‘Hour Volume Goal’ and delivery timeframe. The delivery dashboard also provides suggestions for correcting campaigns where pace is forecasted for over or under-delivery. An example suggestion may be to increase the total booked impressions – and the system will only make suggestions having checked against the DFP ad-server for impression availability. Through this analysis of ad-server ‘availability’ data, the system can maximise HPM across all available inventory. This is in addition to only optimising out of a set of pre-reserved inventory. The usage of a real-time solution in the delivery dashboard (shown below) means that the exchange rate can be processed ‘on the fly’ for multiple simultaneously delivered campaigns. The delivery dashboard takes away the strain of processing the prediction engine for inflight campaigns, allowing it to focus on providing pricing and first phase targeting for the initial ad-server booking. The image above shows the Chartbeat delivery dashboard. It functions as a real-time clock, counting the hours served against a particular campaign.
- 22. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 22 ADVICE ON GETTING STARTED If building a prediction engine for pricing, it is advisable to start simple and scale up as the organisation gets more comfortable with working in this new way. Each dimension adds to forecasting complexity and the volume of data required for forecasting can scale up exponentially. For businesses with first party data, it’s easier to analyse fewer audience segments to begin with and learn about how each segment behaves. From a direct sales perspective, publishers may also want to consider minimum spend deals on CPH. This will ensure that any CPH resource does not get utilised against low margin agreements. Retaining as much contractual flexibility as possible on the delivery of the campaign will help to ensure that there is more manoeuvrability for targeting. Bookings such as site-section specific or page-specific sponsorships will reduce the ability to optimise. If the client is open to using a variety of ad formats (and creative sizes), it will be easier to experiment and learn. For direct trading desks, there is a strong emphasis on maintaining continuing dialogue, transparent pricing and insightful reporting with clients. This is especially important whilst the clients test out the new currency and familiarise themselves with a different process. Before trialling CPH, publishers may want to think about and discuss the following internally: - What is the estimated resource required to predict and manage CPH campaigns? - Considering any limits of time-based technology, what operational impact will this have on the business? - Are there any logistical challenges in the delivery process? - How will the roles of current delivery and measurement systems evolve? - What are the strategic implications of selling CPH? - Are there any existing optimisation tactics that may impact or be impacted by CPH? - How can billing, invoicing and contracts be structured? - How can this be added into financial forecasting and reporting? - What type of delivery reporting will be sent to the client? - Are there any anticipated site-design changes that may negatively affect a time-based strategy?19 Finally, as an external task, it is valuable to research into client reactions on the concept of receiving lower eCPMs to favour eCPH. 19 http://www.thedrum.com/news/2015/01/28/guardian-overhauls-site-anticipation-publishers-selling-ads-based- time
- 23. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 23 THE ROLE OF AUTOMATED SYSTEMS – 2014 vs. 2015 (and beyond) The below diagram reflects the position of CPH (and other sales products) in relation to potential revenue and resource cost. To increase the profit margin of CPH, targeting efficiency must improve alongside the capabilities of delivery systems. The resource cost is linked to the level of human resource that is required from an operational perspective. Any process that is human-resource intensive will reduce a publisher’s opportunity to make direct net profit gains. The potential to reduce resource and increase internal efficiency requires the usage of automated systems that include time-based data alongside impressions. Throughout beta testing at the Financial Times, forecasting for inventory availability was a manual and awkward process. As the amount of variables and dimensions increased to create bigger permutations, the difficulty of forecasting available time also increased. Whilst complexity is necessary to improve targeting, it adds to the analysis workload. An automated system is essential to calculate time availability for custom segments in real-time – driving future time prediction based on the historical performance of the dimensional attributes. Booking systems also need to be integrated to allow for smooth workflow between sales and delivery, producing automated adjustments to line items on time-based bookings. Excessive targeting options may require fracturing a single line item within a booking system into multiple line items. Unless a booking system allows for weighted distribution of a campaign against a dimension, each weighting needs to be pre-calculated and then booked as a separate line item on the same order. An example would be a campaign that is optimised for delivery across a week, with an emphasised weighting on Monday, Tuesday and Thursday. Current ad servers may distribute the campaign evenly across the week, rather than allowing a prioritised weighting against the Day of Week dimension.
- 24. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 24 Whilst delivery systems do not currently feature time-based booking, an integration with the measurement platform could allow for automated campaign cancellation on 100% delivery of time goal, regardless of any outstanding impressions from the original forecasted booking. As measurement systems evolve, publishers can book sponsorship campaigns that cumulatively count time across multiple ads, simultaneously running on a page. This would mean that a publisher could charge the client for the combined minimum time on a page – for example, 6 seconds from a 728x90 and 4 seconds from the 300x600 – giving 10s in total. By removing the resource cost that erodes the profit margin, CPH becomes a highly efficient delivery mechanism to monetise the dimension of Time. [Update – 16th May 2015] A revised planning tool within Chartbeat has simplified some of the planning and operational processes by automating pre-booking prediction. The system can examine the historical HPM performance of a given set of inventory and check against DFP for the availability of this inventory. It works across a range of variables including DMP data, site sections, ad positions, creative sizes and geographic regions – as well as custom criteria used by an ad-server. Using an API integration with DFP, the planning tool features the ability to automatically generate targeted line items. These bespoke line item objects are based on the targeted impressions that are required to deliver a given amount of active exposure time. Combined, these functions provide HPM-maximised targeting without the human effort of manually constructing complex line items. Overall, these recent developments make a positive contribution towards pushing CPH into the upper-right quadrant of the trading matrix shown above.
- 25. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 25 THE FUTURE OF CPH CPH has been created at a time when publishers are scaling back on the fixed costs of housing large direct sales and operational teams. Without complete automation, it becomes a resource intensive task. If the supporting technology infrastructure does not follow, it will remain a currency of large, premium publishers who can dedicate human resource towards its current methodology. The end goal will be to deliver time through programmatic trading and across a multitude of platforms. In order to do this, there needs to be free-flowing information between different tools. It is possible that an end-to-end technology stack could take away the data transfer challenges often seen in multi-vendor implementations. These systems will be required before CPH campaigns can be fully executed across devices or against pre-roll video inventory. Once information can pass openly between a booking system, a DMP and a measurement platform - targeting can evolve to include web analytics data and well as more complex target segments. For example, passing the propensity value for time-generation for an individual user or device would allow a DMP to filter out or include only specific cohorts. Target segments such as ‘Users who have a high probability of viewing ads more than 10 seconds’ could be applied onto targeting efforts to boost yield. Inversely, ‘Users who have a propensity to have sub-5 second exposures’ could be applied as an exclude filter to achieve another HPM increase. CPH is designed to function as a cross-platform currency but needs to adapt and evolve to deliver value to publishers, many of whom now have 50% of traffic accessing their websites via mobile devices. Publishers with first-party, registration data will then be able to create cross-device metrics for values such as Total Campaign Exposure or even Lifetime exposure metrics. This will enable CPH performance to be inputted into Life-Time Value (LTV) calculations. The ability to designate time delivery against Unique Users rather than impressions will also open up opportunities for data-savvy marketers. For example, publishers with registration data and subscriber data will be able to target particular segments based on ‘Share of Voice’. An example sales-ptch might be ‘Dear Advertiser, we have 1,000 CEOs that spend 5,000 hours a month with our site. What percentage of this scarce inventory would you like to buy and at what price?’ This supply-and-demand scenario is driven by the ability to have the lowest denominator as Unique Users (UU) rather than impressions. It puts the context of advertising back onto the conversation between an advertiser and its audience. Progressive layering of time-based ads will allow for target audiences to be pushed down the brand-funnel, utilising set time intervals to complete each stage. In this layering method, a target UU may need 30 seconds of exposure to become aware of the brand. Then the same user can be targeted for a total of two minutes (over multiple impressions) to push them through Brand Association and finally, retargeted for a short amount with the focus on Brand Consideration. At each step of the way, the ad creative can be tailored for the desired effect. Progressive and sequenced ad-layering can also operate across currencies depending on the most effective way to complete each stage. For example, Reach may be cheaper through CPM, Brand campaigns through CPH, and finally a lead-generation event may come from a semantically-matched CPC campaign. An ecosystem such as this would demonstrate how each currency has its own unique role to play in maximising return on ad spend. Creating links between multiple exposures in this way, enables businesses to maximise value against Reach and Frequency. CPH aligns itself with the Editorial goals of a publisher. For the first time, both commercial teams and editorial can share a desired outcome – Attention. That is not to suggest that Editorial should be driven by commercial direction – and journalists should not be coerced to falsely inflate the word count by padding articles with ‘B-grade’ content. There are however, opportunities to tie the two parts of the business together. New Editorial metrics could play a significant role in predicting time-based inventory. The variables 𝑛 and 𝑟 have a dominant relationship with time on a page or article. In order to correctly predict their values on certain site sections, it would be very useful to think of an editorial metric that provides a predictive weight – such as an ‘Attention Unit’. An Attention unit would aim to calculate the average time on page and/or article in advance of the article being published. This could start with a single ‘Total Estimated Reading Time’ data value for articles awaiting to be published. It would take into account
- 26. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 26 predicted reading time for an article – a metric that could be entered by journalists themselves or a text- scraping bot. Such a system may also give updated weighting to elements of engagement which prolong page dwell time, such as reading / leaving comments. In theory, Attention Units (AU) and Total Estimated Reading Time (TERT) could then be rolled up and calculated at wider level for the following dimensions: - Hour of Day, Day of Week, Day, Week and Month - Demographics, Position, Industry and Responsibility - Site sections, article pages and authors This would result in a weighting system that could feed into the calculation formula – and give greater accuracy in predicting values that can be skewed by user behaviour (in relation to Time). E.g. If selling by Time within the Finance section for Accountants, an analyst can use the historical AU and TERT values for this segment in particular to predict future 𝑛 and 𝑟 values with greater accuracy. The same type of predictive weighting system could also be applied to video inventory in a similar method ‘Total Estimated Viewing Time’. As a by-product, there is opportunity for time-based metrics, along with interaction metrics, to be fed back into the design and creative process. Externally, sharing this data with partner agencies and advertisers will enable them to factor past performance into future ad design. This is especially true for rich media ads and interactive ads that are designed to engage target audiences. A simple win for a branding creative would involve sharing the average view time of an optimised delivery, allowing the creative agency to adjust both the rotation of the creative and the timed appearance of the brand logo. Internally, to maintain net revenue or increase HPM performance, website design and UX teams will also be advised to factor in HPM performance when creating new designs or running A/B tests. Design changes tested this way can be attributed with monetary values to show negative or positive impact on CPH yield along with CPM. Well positioned ads will have positive outcomes for HPM. Container tweaks such as lazy-loading and sticky ad units will also increase HPM performance and thus CPH yield. Finally, the current design of CPM relies heavily on individual publishers self-regulating processes and internally auditing bespoke in-house systems for vulnerabilities or miscalculations. This could (and should) eventually evolve to be accessible by external auditors. Delivery verification will also be useful via third- party monitoring tools. There is a need for these tools to be MRC-accredited for the same trading metric that is used within the original transaction. Without incorporating Active-Exposure methodology, third- party verification tools are likely to over-measure the total time delivered for an Active-Exposure-based campaign.
- 27. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 27 CONCLUSION A time-based currency enables a publisher to monetise it’s content through the identification and targeting of viewable ad impressions. The exposure time generated by these particular impressions has a positive impact on brand objectives and is valuable to brand advertisers. Whilst publishers and researchers have previously understood this correlation, the industry had not successfully productised this in the form of a trading currency. In October 2014, Digital Content Next (DCN) published a special report on Time-Based Measurement.20 Out of the digital publishers surveyed, 80% indicated an interest in using a time-based metric to price and sell ads. The publishers also recognised that a lack of standard methodology was one of the main obstacles, with 68% of respondents including it in their top three hurdles. Amid the growing interest in an attention-based economy, the Financial Times has developed, trialled and launched, a practical solution - Cost Per Hour (CPH). This paper has detailed the methodology used to create CPH, in order to encourage other publishers to test and adopt the time-based currency. During and after beta testing at the Financial Times, CPH has increased the overall inventory-based yield for the publisher. It has now evolved into a market-ready proposition – much to the interest of other media owners. Whilst initial results are positive, there is still wider discussion and analysis required to determine if CPH has the potential to become a standardised trading currency. Along with a standard methodology, the mainstream adoption of the currency will also be determined by the advertising technology available to publishers. There is currently a very advanced, but limited, choice in off-the-shelf prediction and targeting systems. Existing industry players, innovative publishers or emerging start-ups could take on the range of opportunities described within these pages. The primary benefit to the publisher or media owner has been demonstrated via the use of a predictive system, which reduces the number of impressions (in comparison to CPM) required to deliver a set volume of time. This efficiency for time-delivery results in a yield gain if the saved volume of impressions can also be monetised. The secondary benefit, which also requires additional research, is that of an increased market share. By providing a higher value to the brand advertiser, per pound or dollar spent, a publisher can try to strategically increase its share of the client budget. To advocate fair-trading and avoid price-discrimination, the design of CPH has connected its price to CPM via a financial multiplier. As the targeting and data management evolves, there is potential to disconnect the pricing between the two currencies by forming distinct inventory pools that exist within a singular environment. Each inventory pool can be structured to serve a particular purpose, by identifying the probability that a specific group of impressions will be better used against a particular advertiser objective. This will lead to a yield management strategy that forecasts inventory value based on multiple characteristics including ad exposure. The adoption of a time-based strategy is not without organisational change. Operational processes will need to be adjusted and direct sales teams will also have to ensure that client-facing conversations begin with a focus on campaign objectives. A time-based economy is a tangible and worthwhile endeavour. Education and on-going conversation will be pivotal in driving change within the media industry. As a trading currency, Cost Per Hour (CPH) exemplifies the necessary evolution towards a sustainable industry; where quality content and attention, matter more than ad impressions and clicks. 20 http://digitalcontentnext.org/wp-content/uploads/2014/10/DCN-Report_Time- BasedMeasurement_10.22.14.pdf
- 28. Cost Per Hour: Using a Time-‐Based Currency for Digital Advertising – Nikul Sanghvi, May 2015 28 RELATED RESEARCH Within ‘The Effects of Exposure Time on Memory of Display Advertisements’21 , the authors conclude that there is a strong, causal influence of exposure time on ad recognition and recall. (It is referenced in the footnote and is highly recommended as further reading on time-based advertising). Their pioneering experiments and documented findings highlight a positive relationship between the time that and ad is seen and branding outcomes. Another company of interest is WebSpectator22 . As a highly innovative solution, they have established and championed a time-based economy since 2011. WebSpectator’s end-to-end product provides both ad serving and analytics against a Media Rating Council (MRC) accredited metric23 – Guaranteed Time Slot (GTS). It creates additional monetization opportunities by refreshing ad slots after 20 seconds of exposure time.24 Following additional investigation into refreshing ad units, the FT conducted related research with key clients and internal business stakeholders, including Editorial. There were some concerns against refreshing ads – partly due to the complexity of rules involved in keeping sequencing space between competing luxury brands. Many clients also prioritised exclusivity of the page itself and frequently paid for this via sponsorship of the homepage or site sections. Finally, rich- media creative formats often had animated rotation lengths beyond the 20 seconds offered by GTS. It was also deemed necessary as a proof-of-concept, to build a pricing model for selling time, which did not involve the inclusion of additional ad-networks or refresh-based ad-serving tools. The role of refreshing ads still has many advantages such as releasing additional inventory – and the benefits of applying this method have not been ruled out. It is a subject matter worthy of its own research25 . Although exposure time may correlate with brand objectives, a refreshing ad container has been proven to produce a higher combined recall value across multiple ads that share the exposure time. By calculating the variable exposure time at which diminishing returns begin for the first creative, it is possible to refresh to a second creative and provide greater economic value to the ad slot. [UPDATE 16th May 2015] Since, this paper was originally written, Chartbeat have announced the ability to refresh ads using the client's existing ad library. This allows publishers to directly monetise the additional inventory created, in a similar way to another impression on the website. Refresh intervals are customisable at the site, page, slot, campaign, line item, and creative level. Publishers may choose to have the flexibility to adjust their refresh times cases where a creative message requires a longer exposure. The methodology used to refresh ads still remains based on Active Exposure, increasing the probability of subsequent refresh-impressions being viewable. ADDITIONAL LINKS FOR FURTHER READING 1) http://blog.upworthy.com/post/75795679502/what-uniques-and-pageviews-leave-out-and-why 2) http://contently.com/strategist/2014/06/16/the-financial-times-on-why-they-ditched-cpms-for-a-new-ad- currency-time-spent/ 3) https://gigaom.com/2014/09/29/chartbeat-gets-certified-to-measure-attention-tries-to-move-advertising- away-from-clicks-and-pageviews/ 4) http://guruinabottle.brandrepublic.com/2014/06/22/traditional-cpm-is-rapidly-passing-its-sell-by-date/ 5) https://medium.com/data-lab/mediums-metric-that-matters-total-time-reading-86c4970837d5 6) http://towcenter.org/research/traffic-factories/ 21 Goldstein, McAfee and Suri (2011), ‘The Effects of Exposure Time on Memory of Display Advertisements’ http://www.dangoldstein.com/papers/Goldstein_McAfee_Suri_Ad_Exposure_EC11.pdf 22 http://webspectator.com/ 23 http://www.mediaratingcouncil.org/ 24 https://webspectator.com/downloads/WS_ResearchLabs_Jan2014.pdf 25 http://blog.chartbeat.com/2014/11/05/economics-ad-refreshing/