Agree on success and record-matching rules before counting outcomes.
Separate acceptance, receipt and meetings. Remove retries and test data. Keep unmatched activity unknown.
A&A's proposed measurement contract: shared counting rules, with a worked example and a blank worksheet.
Automate & Augment | 2026-11-01 | English edition
Processing more tasks does not settle whether to keep an initiative. Agree on success, population, window and evidence first.
Count a first meeting held with a target account, confirmed in the sales record. Sending a message is a different event.
Choose accounts, people or deals. Keep that unit and the follow-up window consistent. Exclude retries and test records.
Report unmatched meetings separately. Compare spending and review time, then set a review date and decision owner.
Source: A&A, Content measurement operations (design record), checked 2026-11-01. Search rankings, AI citations and meetings remain separate measures.
An event records what happened. Evidence for an earlier stage does not prove a later outcome.
| Stage | Evidence that counts | What this does not prove |
|---|---|---|
| Provider acceptance | A valid acceptance ID from the sending service | Recipient received or read it |
| Receipt verified | The specific message checked at the receiving end | Recipient understood it |
| Meeting held | Date, account and attendance in the sales record | A deal closed or revenue was earned |
| Revenue recorded | Entry under your revenue recognition policy | AI caused an increase in revenue |
For prospect email, inbox receipt is usually not directly observable. Count observable responses, such as replies, as their own events.
Source: A&A, Content measurement operations, checked 2026-11-01. Its contact_accepted event means notification-provider acceptance, not a qualified inquiry, meeting or sale.
Example: generating first meetings for a B2B service. These are illustrative settings, not A&A results or benchmarks.
| Agreement | Definition for this initiative |
|---|---|
| Denominator | 100 target accounts with an accepted first send; deduplicate by account ID |
| Success numerator | Accounts in that population with a first meeting held within the window; exclude repeat meetings |
| Window and time zone | Within 14 days of each account's first acceptance; use event time and display in UTC |
| Exclusions and finalization | Remove test data. Finalize after every account's follow-up window and late-record reconciliation |
Source: A&A's hypothetical example, created 2026-11-01. Freeze the target list before launch. Report accounts with no accepted send separately; do not silently drop them.
An ID identifies a record. Similar names or timestamps are not enough to establish a match.
Unverified mappings stay in the unknown category.
Source: A&A, Content measurement operations, checked 2026-11-01. Its content-attribution cookie has a 30-day design lifetime. That is separate from Umami's anonymous IDs and analytics retention.
A retry with the same ID and payload adds no new event. A conflicting payload is held for investigation, not overwritten.
| Received event | Classification | Customer total |
|---|---|---|
| E-A: accepted / Account A | First record | Include |
| E-A: identical retry | Duplicate | Do not add |
| E-B: accepted / Account B | Different record | Include |
| E-Q: test acceptance | QA: quality-assurance test | Keep separate |
Source: A&A, Content measurement operations, checked 2026-11-01. Fictional log: 4 received records minus 1 retry and 1 QA record leaves 2 valid events.
Of 100 target accounts, 8 accounts held a meeting. Another 3 meetings have no verified source. Keep them out of the numerator and report them separately.
| Meeting rate in the population | Unmatched meetings | What remains unproven |
|---|---|---|
| 8 accounts ÷ 100 accounts = 8% of accounts | Unmatched: 3 meetings | Whether AI increased meetings |
Source: A&A's hypothetical calculation, created 2026-11-01. Event totals alone are not an ordered funnel: check the same subject ID, sequence and time window. Generated image is conceptual.
Track the resources spent on the 100-account initiative with 8 meeting accounts. Lower fees can still leave more work for people.
Source: A&A's hypothetical calculation, created 2026-11-01. USD values are illustrative, not converted Japanese costs. These ratios measure resource use, not profit or AI's revenue contribution.
The business owner, sales lead and measurement owner fill this in together. Do not judge outcomes using an unfinished metric definition.
| Agree on | Your definition |
|---|---|
| Success, evidence source and counting unit | |
| Population, start event, window and time zone | |
| ID matching, retention and unknowns | |
| Retries, QA, cancellations and late records | |
| Cost scope, review time and comparison group | |
| Continue / pause rule, review date and owner |
Source: A&A's proposed worksheet, created 2026-11-01. Use the worked example above. A “contract” here is an agreement on measurement rules, not a legal contract.
In this example, review after every 14-day window and late-record check. The sales lead decides: continue, revise the design or defer judgment.
Source: A&A's proposal, based on Content measurement operations; checked 2026-11-01. Without an appropriate comparison design and uncertainty assessment, report observations rather than causal impact.
Sources & editorial note: AI researched and drafted this deck; a separate AI reviewer and machine checks reviewed it. No human reviewed the content.
Design records, product specifications and fictional examples are different evidence. Past records do not establish current operation.
| Source | What it supports, and its limits |
|---|---|
| A&A: Content measurement operations | Acceptance, retries, QA exclusion and unknown attribution. A design record, not current deployment proof |
| Umami: Metric definitions and FAQ | Default anonymous IDs do not auto-link across rotation periods. Self-hosted data remains until manual deletion. Verify your version and configuration |
| Google Search Console: Dimensions and data groupings | Page grouping and query omissions can make table values differ from chart totals. Search visibility is not meeting evidence |
Source: The English primary sources above, checked 2026-11-01. A&A proposes logging AI citations separately with the question, date and cited URL.
Choose the owner and review date. Complete the worksheet with sales and measurement leads before the next initiative starts.
Read the original English deck
https://www.automateaugment.com/en/decks/ai-gtm-measurement-contract
Read next: Verify AI work at the receiving end · Test sales follow-ups with measurable checks
Explore A&A's AI-native GTM support or contact A&A.