The outcomes opportunity
What 140 advertising and media companies promise on their own websites, and what the gap is worth to whoever closes it first.
138 companies read page by page · September 2026 · Calidescope LLC
The context · What advertisers are asking to buy
A sale
A purchase the advertiser can see in its own books.
A signup
A trial started, an account opened, a lead handed over.
A store visit
A trip to the shelf, matched back to the ad.
A lift in revenue
More than the business would have done anyway.
137 of 138
companies make at least one claim like this
740 of 2,213
claims we read are about a result, not a feature
30 of 140
build their whole front-door pitch on the word
The gap · What those 740 claims actually promise
8
of 2,213 claims put anything at risk if the number is missed
What the eight cover. Every one of them happens before the sale.
Fraud
Google DV360, Madhive
A refund on fraudulent impressions.
Attention
Adelaide
An attention score, for one named advertiser.
Delivery
VideoAmp, Samba TV
Delivery against an agreed currency.
Completion
Channel Factory
A cost per view, against a benchmark.
The precedent · Four times this industry changed what it sells
Audience1952–1967
→
Nielsen ratings
→
Make-good time
Impressions1995
→
Ad server logs, the ABC
→
Make-good impressions
Viewability2013–2015
→
The MRC
→
No payment
Attention2026
→
Adelaide
→
Announced · no remedy stated
A business resultNow
→
None yet accredited
→
Open
It has been done anyway, five times, each one narrow: Meredith 2011 · the MPA and 16 publishers 2015 · Simulmedia 2015 · Westwood One 2017 · LG Ads 2022. Each named a remedy. None became a standard.
What is new · Agents, sold as the way results get delivered
35 of the 64
companies we re-read now make a claim about AI agents planning, buying or optimising media on their own
On the home page
“The first operating system that lets AI agents plan, execute, and optimize media.”
pubmatic.com
On the company blog
“There should be a meaningful separation between the execution of a workflow and the approval of that workflow.”
pubmatic.com/blog
In the contract, §3.2
“PubMatic does not guarantee the outcome of any given campaign and does not guarantee any specific results.”
pubmatic.com/legal
*MAVEN Claim and Evidence Analysis by Michael Risucci
The study · What we did, September 2026
627 pages
Home page first, then product, pricing, press. Every company the same way.
2,213 claims
Each one copied word for word, with the link to the page it sat on.
740 about a result
Sorted by one question: what can a buyer check without leaving the page?
A second read
An outside reviewer checked every conclusion and wrote down what it could not support.
120 companies across twelve parts of the supply chain, plus 20 whose product is the result itself.
*MAVEN Claim and Evidence Analysis by Michael Risucci
What they say · Six results claims, exactly as published
Sprout Social
“Go from setup to inevitable ROI”
sproutsocial.comNothing sits beside it
Circana
“Deliver up to 9x ROI”
circana.comA number, no source
Taboola
“Hyundai partnered with Taboola to boost their campaign performance, resulting in a 30% lower Cost per Session.”
taboola.comThe client is named
Vizio · Walmart
“how much incremental reach they’re gaining with VIZIO vs. their linear campaigns”
vizio.comThe test is described
Paramount
“6X longer ad views than social (Source: … x MediaScience, Q423)”
paramount.comAn outside firm is named
AppLovin
“Amazon delivered the stronger return with a 14.61% revenue lift, generating $117K in incremental revenue at a $1.24 incremental ROAS.”
The client named, the test described, and the test run by the advertiser rather than the seller — a geo-holdout by Ridge.
The pattern · All 740 results claims, sorted by what a buyer can check
Says it
The claim, and nothing beside it.
221
Take it on trust670 claims
Shows a number
A figure, with no source named.
159
Names the client
The advertiser it happened to.
290
Shows how it was measured
The test design, on the page.
42
Check it yourself70 claims
Names the independent measurement
An outside measurement firm, or a test the seller did not run.
28
Each claim counts once, on the highest thing a buyer can check on the page. The 449 that already print a number or name a client are closest to the line: one sentence about how it was measured moves them across it.
The newer claims · 80 agent claims, read exactly the same way
Says it
81%
Shows a number
9%
Names the client
8%
Shows how it was measured
2%
Names the independent measurement
0%
Tatari
“In 17 of 18 valid cases, the unmodified AI plan performed as well or better than the human-adjusted version.”
The only agentic claim we found with a test behind it. Tatari’s own four-week randomised A/B test, on its own clients.
64 companies re-read on 30 September · 35 carry an agentic claim · 80 claims coded. Not the same kind of sentence as a results claim, so read the comparison as directional.
The opportunity · Four positions, and what each one takes
Test what the agent did
“Our agents run the buy, and we measure what they changed.”
What it takesA holdout on agent-run spend against human-run spend, published.
Tatari is closest today
Bring in an outside firm
“Someone who does not sell this measured the result.”
What it takesAn outside measurement firm named beside the number, or a test you did not run.
Paramount and AppLovin do this
Publish the test design
“Here is exactly how the number was produced.”
What it takesThe design printed on the page, next to the result.
Vizio and Tatari do this
Say what happens if you miss
“If we miss the number, here is what you get.”
What it takesA remedy in writing, tied to a business result.
LG Ads and Simulmedia have done it
The first move · The claim you already lead with, rewritten
01 The proofWhat already happened
At the client, we delivered the result, using the method, verified by an independent measurer.
The client
Who it happened to
The result
The number, and what it is a number of
The method
How the number was produced
An independent measurer
The firm that checked it, and does not sell what you sell
02 The promiseWhat happens next time
If we miss the number, you get the remedy.
The number
The threshold you will stand behind
The remedy
What the advertiser gets if you miss
What good looks like · The same two sentences, filled from one real page
AppLovin · applovin.com/en/resources/ridgeAll four filled
At Ridge, we delivered 128% above the iROAS goal, using a three-week geo-holdout incrementality test, verified by Haus, hired by Ridge.
NobodyNot one of the 2,213 claims we read writes this sentence
If we miss the number, you get the remedy.
Verbatim from the page: “Ridge beats iROAS goal by 128%” and “Ridge partnered with Haus to run a three-week geo-holdout incrementality test measuring AppLovin’s impact on sales of Ridge’s power banks.”
The limits · What an outside reviewer would not let us say
01
The top rungs are thinly occupied.
That describes the pages we read. An empty page is not proof of demand.
02
Companies selling results show their test more often.
That is publishing practice. We did not measure honesty.
03
Plain language does not predict strong proof.
Walled gardens use the plainest words and show a test on none of their claims.
04
Sellers have put money at risk before.
Five times since 2011, each one narrow and conditioned.
05
The agentic comparison is directional.
80 claims against 740, and not the same kind of sentence.
*MAVEN Claim and Evidence Analysis by Michael Risucci
The record · Everything behind these thirteen slides
The full read
22 sections · 25 minutes
Part of the supply chain by part, with every figure in place.
The category notes
13 short reads
One per segment. Every number carries its link.
The second read
5 files · 193 rows
Every conclusion checked, with the limit written beside it.
The raw record
2,213 claims · 627 pages
Every claim word for word, with the page it came from.
Company records, segment definitions and news history came from the Marketecture Advertising Database. Their API is what made a 140-company sample buildable in two days.
How to read this · 140 companies · 627 pages · 2,213 claims, each with its URL · 740 about a business result · 138 of the 140 read 16–17 September 2026 (two sites blocked capture) · 64 companies re-read for agent claims on 30 September · figures describe the pages we read, not the whole market · Calidescope LLC.
*MAVEN Claim and Evidence Analysis by Michael Risucci