How should we document test results so they are not lost?
Our answer
Keep one shared test log your business owns. Write each test's plan before launch, then record its result with scope, dates, spend, uncertainty and what else changed, losers included. Save platform reports before they expire.
- Confidence
- Medium
- Strongest proof
- 3 official rules
- 1 practitioner, 10 opinions or claims
- Sources
- 11 sources
- 14 statements
ALG
CE
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentResults get lost in scattered files and missing context. Test history can sit in agency decks, spreadsheets and dashboards, and a bare return figure means little without scope, dates, outcome, spend and uncertainty (Sellforte). Common Thread Collective adds the dates and promotion.
A test's plan belongs on paper before launch. Write the hypothesis, primary metric, duration and decision rule first (Atticus Li, Topsort). Record losing tests as well as winners (Convert, EntreResource).
Platform data expires. From June 2026 Google keeps hourly, daily and weekly Google Ads data for 37 months and recommends exporting what you need beyond that. Meta keeps unique-count and hourly breakdowns for 13 months.
Our judgment
Keep the plan and the result in the same entry of one log the business owns. The sources name three ways results get lost (scattered files and memory, missing context, and platform data that expires), and one owned entry written before and after each test covers all three. Save each test's platform report and daily data in the log rather than relying on the platform, applying Google's advice to export granular data and Meta's limit on some breakdowns to test records. Start in a free sheet or page with fixed required fields and tags, and move to a dedicated library only when tests run by the tens or feed a mix model. That reconciles the advice of Convert's interviewees to start simple with Atticus Li's warning that unstructured logs decay and Sellforte's point that many experiments need a library that shows how each feeds an estimate.
Our judgment, built from Sellforte, Atticus Li, Convert, Common Thread Collective, Google and Meta: our reading of these sources, not a finding from any one of them.
What the sources say
14 statements from 11 sourcesGoogle Ads keeps hourly, daily and weekly reporting data for 37 months from June 1, 2026, and monthly or longer data for 11 years; reach and frequency are kept 3 years.
“Reach and frequency metrics will be available for 3 years only”
CTC says a ROAS target from a Black Friday lift test may not fit January; keep the test dates and promotion attached so a seasonal result does not become a permanent budget rule.
“Keep the test dates and promotional context attached to the result so a seasonal finding doesn't become a permanent budget assumption.”
Relevant Audience, an agency, says plan Google Ads experiments for a month or more with the first one to two weeks as ramp-up, and warns a 10% split on a small campaign may never reach significance.
“Too small a split on a small campaign. A 10% treatment on a low-volume campaign may never reach significance.”
inBeat says the brand should hold Meta admin rights and own its pixels, Conversions API, catalogs and customer data. Moving in-house takes about a quarter if phased, with current agencies kept live meanwhile.
“takes about a quarter when you run it in phases and keep existing agencies live until the new setup proves itself”
Sellforte, which sells a test library, says to record each experiment's design, scope, dates, outcome measured, spend and result with its uncertainty, including inconclusive tests.
“A valid experiment with an inconclusive result may still provide information.”
Stackmatix, a paid media agency, says to keep offers, season and product changes constant during a geo holdout and write the plan down first so the result holds up.
“Document the plan before you start so the result is defensible when someone asks why you paused spend in a market.”
Sellforte, which sells testing and mix models, says to pick tests by impact, uncertainty, feasibility and cost: plan slow geo tests quarterly and run faster platform studies monthly.
“Testing the largest channel every quarter can repeat what the team already knows.”
EntreResource suggests a Meta creative test log with hypothesis, control, variable, KPI, budget, dates, result and next action, set review dates in advance, and record losers as well as winners.
“Failing to document the results of both winners and losers.”
Google announced that from June 1, 2026, daily, hourly and weekly Google Ads data is kept 37 months and monthly or longer data 11 years; export older granular data first. BigQuery copies are unaffected.
“If you require granular historical data beyond 37 months, we recommend exporting it prior to the June 1, 2026 deadline.”
Convert, an A/B testing vendor, suggests one shared test repository with the page, hypothesis, variants, dates and results, losers included, owned by one team, starting with simple tools.
“Don’t hide tests that lost.”
Atticus Li suggests writing a short pre-registration before each test: the hypothesis, primary metric, stopping rule and decision criteria, timestamped and not changed after launch.
“Small teams benefit more from pre-registration, not less.”
Atticus Li, a CRO consultant, says test libraries in spreadsheets fail within a quarter as the reasons get lost and old tests get repeated; enforce a few required fields and add tags later.
“The most tenured IC becomes the real database. When they leave, the experiment knowledge base resets.”
Topsort, a retail media platform, advises writing the hypothesis, success metrics and test length before an incrementality test, and not stopping early or changing settings mid-test.
“Changing test parameters mid-flight”
Meta stopped returning 7-day view and 28-day view attribution windows in the Ads Insights API from January 12, 2026; 1-day click, 7-day click, 28-day click, 1-day engaged view and 1-day view remain.
“On January 12, 2026, we will deprecate certain attribution windows available today with the Ads Insights API’s”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it2 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- What the research saysWhat academic research adds, kept apart from the answer
- How sure we areWhat its confidence rests on, and what would change the answer