The Probative Standard: how we count signed cases

The Probative Standard is how Probative counts. It defines a lead, a qualified lead and a signed case, sets how cost per signed case is calculated and how a signed case is credited to a channel, and says what evidence is needed before anyone claims marketing caused a case. Every number in a Probative report follows it, and so does every claim we make about our own results.

One rule sits above the rest: attributed is not caused. Attribution says which dollars touched a signed case. Only a controlled test says which dollars created it. We label the difference every time.

Probative is a marketing firm, not a law firm or lawyer referral service, and nothing in the Standard is legal advice. Each firm’s responsible attorney decides what information about its clients and prospective clients may leave the firm.

Version 1.0. Last reviewed October 24, 2026, by the Probative editorial team.

The definitions

Scope

The Standard covers every Probative deliverable (weekly signed-case reports, intake scorecards, monthly and quarterly reviews, test readouts, forecasts and AI-answer reports), every engagement, and every practice area and language we serve. It also covers proposals and any public case study, data study or benchmark, ours included. It doesn’t replace a firm’s own case-management statuses: at kickoff we map them to the Standard’s stages, and the mapping is printed in the firm’s report.

Versions and changes

• Numbering. Versions are numbered MAJOR.MINOR. A minor version clarifies wording or adds a rule without changing any reported number. A major version changes a definition or formula in a way that changes reported numbers.
• Notice. A major change is published 30 days before it takes effect and is open for public comment during that time; the comments are published with it. For one full quarter afterward, reports show the affected numbers under both versions, and history is restated under the new version and labeled.
• Labels. Every report names the version it follows (“Standard v1.0”).
• Review. We review the Standard every quarter, after any platform policy change that touches measurement, and in full once a year. Every change goes in the changelog below.
• Contract definitions. If a client’s contract defines a term differently, the contract governs for that client, and the number is labeled “contract definition”.

License, citation and conformance claims

License. This text is published under CC BY-ND 4.0: you may copy and share it unchanged, with attribution. The 40-point checklist and the templates are under CC BY 4.0, so you may adapt them. “The Probative Standard” is Probative’s name; the licenses cover the text and templates, not the name.

Citation. Cite by version and section, for example “The Probative Standard v1.0, Holdout test”. You may quote it with attribution and a link.

The conformance-claim policy, in short:

1. Anyone may claim that a report or published result follows the Standard, with no license, fee or permission. We certify, approve and endorse no one, so a claim is the claimant’s own statement. Internal-only use needs no statement.
2. A claim says “Reported under The Probative Standard v1.0”, always with the version. It never uses our logo, a seal or a badge, never says “certified”, “approved” or “endorsed” by Probative, and never implies we reviewed the work.
3. A public conformance statement comes first. On a page anyone can open without signing in, the claimant gives the version; the reporting windows and the basis and attribution view of each published number; the case-management system the signed cases come from; the base counts behind each published cost per signed case; every contract definition in use; and a link to this page. Every claim links to that statement.
4. One departure is allowed: a contract definition, stated in the conformance statement and labeled on every number it affects. Any other departure means the report may describe its own method and cite the Standard for comparison, but not claim to report under it. An unstated departure voids the claim.
5. Versions. A claim holds for the version it names.
6. Corrections. Anyone can report a claim that seems not to meet these terms, ours included, to team@probativegrowth.com. Within 30 days of learning of it, we write to the claimant, name each term it misses and ask it to fix its statement or withdraw the claim within 30 days; if it does neither, our trademark counsel sends a letter. A correction request concerns the claim only; we don’t comment publicly on others’ results.

Reporting rules

• Outcomes first. Page one of every report leads with signed cases, cost per signed case and what isn’t working. Impressions, clicks, click-through rate, cost per click, sessions, rankings, raw call and lead counts, platform-reported conversions and AI “visibility scores” without intervals appear only in labeled diagnostic sections.
• Every number shows its version, period, basis, source system and maturity.
• Small numbers. No cost per signed case is shown for any cut with fewer than 5 attributable signed cases; the cut shows its spend and counts instead.
• Fair comparisons. Equal-length windows on the same basis, the year-earlier window where it exists, shared axes, bars that start at zero, and an ILLUSTRATIVE stamp on any invented data.
• Raw data. The firm owns its accounts and data. It can also read our working records row by row (one row per lead, keyed by its record ID, with no names), and we trace any reported number to its rows within 2 business days.
• Errors. An error is material if it changes a signed-case count, moves a cost per signed case by 5% or more, changes a test result’s category or touches a decision we recommended. We tell the firm in writing within 2 business days, issue a corrected report and log it. Errors in our favor are handled exactly like errors against us, and published material is corrected on the same page, with the date and what changed.
• Public results come from a period that passed every Signal Health check, name the version, period, basis and attribution view, show their test status or say “attributed, not tested”, disclose any material connection, carry the client’s written approval and are rounded down. Results from work our founder led outside Probative are labeled that way and were not measured under this Standard unless the label says so.

Forecasts

Signed cases are forecast at P10, P50 and P90. Each quarter the previous quarter is backtested, and we report the mean absolute percentage error of the P50 and the share of months whose actual result fell between P10 and P90, which should be about 80% if the forecast is honest. Lag-adjusted figures are estimates, labeled, and never used in public results.

Conflicts of interest

When we run a firm’s media and also report its results, we are grading our own work. So:

1. Every channel we manage shows its latest test result beside its attributed cost per signed case, or the words “no holdout yet”. No proposal, review or case study presents attributed results as proof of what our spend caused.
2. Causal claims need a test that meets the holdout-test rules, for our work as for anyone’s.
3. Before we recommend raising a managed channel’s budget by more than 30%, we show its marginal or test evidence, or say there is none and offer a step test.
4. Our pay is never a percentage of spend: flat fees, no markup, no share of legal fees, nothing per lead, client or case.
5. When we measure a firm’s current agency, we say in writing: “Probative also manages ads for law firms. While we measure your current agency, we won’t offer to take over its work unless you ask in writing. Your agency can see the same Ledger you see.”
6. Any commission we earn on a tool is disclosed before the firm buys, and any referral fee we pay a partner is disclosed to the firm referred.
7. No firm’s data reaches another firm. Bids, budgets, values and results stay inside each firm’s own accounts and reports.
8. Benchmarks only with consent: a firm may opt in, by an addendum its responsible attorney signs, to share firm-level totals, combined only in cells of at least five consenting firms, with no firm above 25% of a cell and data at least three months old.

The 40-point measurement checklist

Use it on any agency’s reporting, ours included. A firm that can answer yes to all 40 knows which dollars sign cases. (CC BY 4.0: adapt it freely, with attribution.)

Ownership and access

1. Every ad, Local Services Ads, analytics, call-tracking and case-management account is in the firm’s name and on its payment method.
2. The agency works through user access the firm can remove at any time.
3. The firm can export its full history from each platform without the agency.
4. The domain, website, Business Profile and tracking numbers belong to the firm.
5. The firm holds a current exit kit: an admin list, exports and dashboard copies.

Definitions 6. The agency has written definitions of a lead, a qualified lead and a signed case. 7. A short call doesn’t count as a lead, and the minimum length is written down. 8. Duplicates, existing clients, job seekers and spam are excluded by rule. 9. “Signed” means an engagement agreement executed, as recorded in the firm’s system. 10. Cost per signed case states which costs it includes.

Tracking 11. Click IDs (GCLID, GBRAID and WBRAID) are captured into the firm’s system. 12. Calls are tracked with numbers tied to channels and visits. 13. The system has required fields for source, language, case type and stage dates. 14. At least 85% of paid leads carry a click ID or call ID. 15. Tracking outages are logged with their dates.

Reconciliation 16. The agency’s signed count is reconciled to the firm’s records every week. 17. The gap between the two is printed, not hidden. 18. Platform-reported conversions are compared with the firm’s outcomes. 19. Late signings restate earlier periods on a stated schedule. 20. Unattributed signed cases are shown on their own line, not spread across channels.

Uploads and bidding 21. Qualified and signed stages go back to Google as conversions every day. 22. The conversion names sent to platforms are neutral and carry no case details. 23. Values come from the firm’s own fee history, in bands, recalibrated on a schedule. 24. The responsible attorney has signed off on what is uploaded. 25. No Customer Match or lookalike audiences are built from client lists, and remarketing runs only where each platform’s policy allows it.

Intake 26. Answer rate is measured from call logs, by hour and language. 27. Speed to lead is measured for forms and missed calls. 28. Missed calls that Local Services Ads bills for are counted and audited. 29. Calls are graded on a written rubric, with consent. 30. Leads untouched for 24 hours are listed every week.

Incrementality 31. The agency can say which channels have been tested for incrementality. 32. Tests are written down before they start, with a decision rule. 33. Tests have a power analysis and a minimum detectable effect. 34. Readouts show intervals, not just point estimates. 35. Brand search is labeled “not proven incremental” until it has been tested.

Reporting honesty 36. Reports lead with signed cases and cost per signed case, not with clicks or rankings. 37. The firm can see the raw rows behind any number. 38. Errors are disclosed in writing and corrected. 39. Comparisons use equal windows on the same basis. 40. The agency’s fee is not a percentage of the firm’s ad spend.

Changelog

Under consideration for a later version, not in force: folding Local Services Ads into the value model once its migrated campaigns expose conversion goals; a consented, published signed-case benchmark; and coding rules for self-reported “ChatGPT or AI” sources as that channel grows.