top of page

The Cost of Slow Diagnosis

Henry Rivero
Aug 12
8 min read

When the revenue answer arrives after the opportunity has gone


Consider a composite scenario familiar to many video publishers.


On Monday morning, four reports appear to tell four different stories:


  • Player analytics reports viewing time up 9%.

  • The ad server reports delivered impressions up 4%.

  • The principal SSP reports average gross CPM up 6%.

  • Finance reports recognised net advertising revenue down 8%.


Sales sees weaker agency demand. Ad operations suspects unfilled connected-TV inventory. Product points to audience growth in less heavily monetised content. Finance asks whether the difference sits somewhere between gross delivery, partner fees, campaign credits and revenue recognition.


Every explanation is plausible.


None is yet strong enough to act on.


The figures are not necessarily wrong. They describe different parts of the business, using different scopes, definitions and reporting periods.


That is the diagnostic problem.


It is not the time between a revenue movement and its appearance on a dashboard. It is the time between seeing the movement and constructing a sufficiently complete explanation to make a commercial decision.


Slow diagnosis did not cause the original decline. Its cost is the additional value that could have been protected or recovered if the business had understood the movement sooner.


The critical question is:


How much value remains recoverable—and for how long?


One business, several versions of revenue

Six connected publisher systems spanning player analytics, video and SSAI, ad serving, demand, sales and finance.
Each system records a different stage of the same commercial event. The diagnostic challenge lies in joining those partial views into one defensible explanation.

A publisher does not have one video monetisation system.


It has a collection of systems that record different stages of the same commercial event:


  • Player analytics measures starts, sessions, viewing time and content behaviour.

  • Video or SSAI platforms record break creation, pod construction, stitching, errors and latency.

  • The ad server records eligible requests, decisions, delivery and campaign pacing.

  • Demand partners report auctions, bids, clearing prices and gross revenue.

  • Sales and order-management systems hold commitments, negotiated prices and underdelivery.

  • Finance reports revenue after fees, revenue shares, credits and reconciliation.


Some of these systems may be operated directly by the publisher; others are supplied by external technology partners. Together, they form the publisher’s operating stack.



The joins between them are where diagnosis slows down.


A content identifier in the player may not match the placement taxonomy in the ad server. Device categories may differ between the SSAI platform and the SSP. Systems may use different time zones, attribution windows and definitions of gross and net revenue.


Even the word “impression” may refer to an ad-server decision, a player event, an SSP-recorded transaction or a billable event.


Each team can therefore be correct within its own system while the organisation’s commercial explanation remains incomplete.


Following the diagnosis through the stack


Return to the composite scenario.


The initial assumption is that higher viewing, more impressions and stronger CPM should produce more revenue. But that assumes the measures cover the same inventory and describe the same stage of monetisation.


They do not.


When evidence is joined across the stack, the investigation finds several contributing movements.


First, player analytics shows that much of the viewing growth came from shorter sessions and content with fewer completed ad breaks. Viewing increased faster than monetisable opportunities. This was primarily a mix effect, not an operational failure.


Second, SSAI telemetry identifies more failures and timeouts affecting ad pods on two connected-TV app versions. Some eligible breaks therefore produced fewer completed ad deliveries than expected. This was potentially recoverable.


Third, the SSP’s higher CPM applied only to the programmatic impressions it sold. It did not represent the yield of the publisher’s entire video business. The remaining programmatic inventory had also become more concentrated in higher-value markets and formats, raising the average price while total sellable volume weakened.


Finally, sales forecasting and finance reporting reflected direct-campaign underdelivery, credits and makegood obligations that were absent from the SSP’s gross-revenue view.


There was no single root cause.


The decline combined:


  • An accepted content-mix effect

  • A preventable delivery problem

  • A misleading price-mix movement

  • Subsequent commercial consequences


Even after the data was joined, some uncertainty remained. Incomplete mapping between content and ad-server identifiers meant that the precise contribution of the content-mix change could only be estimated.


That is a realistic diagnostic result: sufficient confidence to act on the technical fault, without pretending that every component has been measured perfectly.


Not every decline should be reversed


In this case, the SSAI problem should be addressed. The content-mix movement may not need to be.


A streaming business might deliberately reduce ad load to protect viewer retention. It might enter a new market knowing that demand will initially lag viewing. It might prioritise a guaranteed campaign over an immediately higher-priced opportunity.


These are commercial trade-offs, not automatically leakage.


The diagnostic question is therefore not simply:


Where did revenue fall?


It is:


How much of the movement was intended, how much was an accepted consequence of another decision, and how much was preventable?


Without that distinction, a publisher can lower floors when the problem is delivery, increase ad load when the problem is eligibility, or redirect sales effort when the issue is technical.


Diagnosis connects the financial outcome to the decisions and constraints that produced it.


Video opportunities have an expiry time


Perishable inventory is not unique to video. But the problem can be particularly acute in premium video, where valuable opportunities are concentrated in short viewing and demand windows.


A live event ends. A programme premiere passes. A seasonal demand spike recedes. A guaranteed campaign enters its final delivery days.


The individual opportunity cannot be recreated.


Some campaign value may be recovered through later delivery or makegoods. But this can consume other inventory, displace another buyer or deliver the audience at a less valuable time.


A technical delivery fault identified after a live event has ended is no longer an optimisation opportunity. It is an explanation of revenue that cannot now be earned in the same circumstances.


How quickly does a fill gap become material?

Table showing daily revenue exposure at different fill shortfalls and marginal net eCPMs.
Illustrative daily revenue exposure at 10 million eligible video ad opportunities. Exposure is not the same as avoidable loss; recoverability depends on whether an earlier, correct intervention was possible.

Assume the affected part of the composite operation generates 10 million eligible video ad opportunities per day.


For this illustration, an eligible opportunity means an ad slot that has passed the relevant rights, policy and technical checks and is available for demand. Fill is the proportion that becomes a delivered, revenue-generating impression.


At this scale, a one-percentage-point fill shortfall represents 100,000 fewer delivered impressions.


The daily revenue exposure is:


Eligible opportunities × fill shortfall × estimated marginal net eCPM ÷ 1,000


The relevant price is the expected value of the marginal opportunities that failed to fill. This may be lower than the operation’s reported average eCPM.



At an estimated marginal net eCPM of $18:


  • A one-point fill shortfall represents $1,800 of daily exposure.

  • A three-point shortfall represents $5,400.

  • A five-point shortfall represents $9,000.


If the five-point shortfall persists during a six-day investigation, cumulative revenue exposure reaches $54,000.


That does not mean diagnostic delay caused a $54,000 loss.


Some of the movement may reflect genuine demand conditions. Some campaign value may be recovered later. An intervention may also have costs or side effects.


The avoidable component requires a counterfactual:


Estimated avoidable loss = daily revenue exposure × delay attributable to diagnosis × recoverable proportion


To illustrate the sensitivity, assumed recoverability of 30% to 60% would produce an avoidable-loss range of approximately $16,000 to $32,000.


This is a scenario range, not an industry benchmark. In practice, the assumption should be based on the affected demand routes, remaining campaign windows and evidence from previous incidents.


The purpose is not to produce a perfectly precise leakage number. It is to make the economic importance of diagnostic delay visible enough to manage.


A premature answer can be more expensive than waiting


Commercial pressure often forces teams to act before the evidence has been connected.


Suppose the publisher responds to the headline decline by lowering programmatic floors. The SSP may sell more impressions, but the change does nothing to correct the SSAI failure or direct-campaign underdelivery. It reduces price on the inventory that was already working.


Alternatively, the publisher might increase ad load because revenue per viewing hour has fallen. But if connected-TV pods are failing before completed delivery, creating more nominal opportunities may simply add complexity and damage the viewer experience.


Sales might be told that agency demand has weakened when the evidence shows a product-mix change and a technical delivery problem. That explanation can then influence forecasts, pricing and client conversations.


The standard should not be perfect certainty. Video markets move too quickly for that.


It should be structured uncertainty:


  • What is confirmed?

  • Which systems support it?

  • What is the leading explanation?

  • What credible alternatives remain?

  • How much revenue is exposed?

  • Which actions are reversible?

  • When does the decision window close?


This allows the organisation to act proportionately without pretending that every variable is known.


Evidence that reconciliation remains difficult


A related problem can be seen at industry scale in the ISBA/AOP programmatic supply-chain studies conducted by PwC.


The original study analysed 267 million impressions involving 15 advertisers and 12 publishers. Only 31 million impressions—12%—could be matched successfully from advertiser to publisher.


The obstacles included inconsistent formats, different levels of granularity, missing fields, contractual permissions and unequal data capture across platforms. PwC also identified a 15% “unknown delta”, but explicitly warned that it could not determine with certainty what the delta represented. It should not be interpreted as proven fraud or revenue leakage. 2020 ISBA/PwC study


In the follow-up study, improved access to log-level data and greater standardisation increased the match rate from 12% to 58%. On a comparable restated basis, the unknown delta fell from 17% to 3%. The audit still took nine months. 2022 ISBA/PwC study


These studies do not measure diagnostic delay inside a publisher. They demonstrate something narrower: even sophisticated participants struggle to connect advertising transactions when identifiers, definitions, permissions and data structures are inconsistent.


It is reasonable to infer that the same class of reconciliation problem can arise inside a publisher’s stack—on a smaller but much faster operating timescale.


An industry financial audit may take months. A video-revenue decision may need to be made within hours.


Measure the speed of explanation separately


A publisher can deliver every scheduled report on time and still take too long to explain an unexpected movement.


A practical diagnostic framework could track four intervals:


Time to detection

How quickly was the material movement identified?


Time to first credible explanation

When did the organisation move from possible causes to an evidence-based working account?


Time to decision confidence

When was the evidence sufficient to select an intervention and accept its risks?


Time to verified value

When was the intervention confirmed to have protected or recovered value?


These intervals should be assessed alongside the urgency and commercial scale of the event.


A two-day investigation may be excellent performance for a structural problem and commercially useless for inventory that expires tonight.


The objective is not to eliminate investigation. It is to prevent evidence gathering and reconciliation from consuming the period in which action still has value.


The deeper cost is loss of operating confidence


Repeatedly inconclusive investigations can create wider organisational risks.


Sales, operations, product and finance can develop competing accounts of performance. Analysts may build shadow reconciliations. Forecasts can acquire larger judgemental overlays because teams cannot determine whether a movement is temporary, structural or erroneous.


Executives then face an uncomfortable choice: act quickly on incomplete evidence or wait for confidence while the recovery window closes.


The problem is not that the publisher lacks dashboards.


It is that no individual dashboard can connect what happened in the player, what became an ad opportunity, what reached the ad server, what demand bought, what sales had committed and what finance ultimately reported.


The commercial question is therefore not:


How quickly can we report yesterday?


It is:


How much value remains recoverable by the time we understand yesterday?


That is the real cost of slow diagnosis.


The next article in this series will examine where AI can—and cannot—help close this gap: not by inventing explanations, but by accelerating the evidence gathering, reconciliation and hypothesis testing needed to connect a revenue movement across the publisher’s stack.


Editorial note: The operating scenario and financial figures in this article are illustrative composites designed to demonstrate the mechanics of diagnostic delay. They are not presented as results from a single publisher.

bottom of page