A recommendation is a suggested next step that comes out of testing. It always points back to the evidence that prompted it.
For example, in SecondBank's rewards test, participants couldn't work out what their points were worth. The team recommends showing point value in dollars. That advice goes to the people who decide what happens next.
Why it matters
Testing that ends without a clear next step tends to sit in a slide deck and fade. Or the opposite happens: a researcher's favorite idea gets presented as a conclusion, and nobody can tell which parts the evidence actually supports.
Good recommendations close that gap. They turn what was learned into something the team can act on, they show exactly which evidence backs each suggestion, and they keep advice separate from the decision, so the person accountable still makes the call.
State each recommendation as one action with its reason, link it to the signals and observations behind it, and send it to whoever owns the next decision.
How we write one
Every recommendation is written as one sentence that names the action and the evidence behind it:
An action because the evidence.
Show point value in dollars because participants couldn't convert points into what they buy.
Writing it this way forces every suggestion to carry its reason, so anyone can see whether the evidence really supports it.
How it differs from similar ideas
-
A decision is the call someone accountable makes. A recommendation is advice offered to that person; it can be accepted, changed or turned down.
-
A concept is a proposed solution with a name and a shape, such as the points progress bar. A recommendation suggests an action, which may be to build, change or drop a concept.
-
A signal says what the evidence means. A recommendation says what to do about it.
-
Findings are the full bundle from a round of testing. A recommendation is one part of that bundle.
-
A hunch is a belief that may not rest on data. A recommendation must rest on evidence from testing.
A good recommendation
-
Names one specific action
-
Gives the evidence behind it
-
Links to the signals or observations it rests on
-
Is clear about who should act on it
-
Stays advice, leaving the call to the decision owner
-
Can be checked by a later test
What happens to it
-
Suggested. A researcher, designer or Ray proposes it from a round of testing.
-
Captured in Findings. It's added to the Findings for that round, next to the signals it rests on.
-
Considered. The decision owner reviews it alongside other evidence.
-
Accepted, changed or declined. A decision is made, and the recommendation is linked to it either way.
-
Checked. If accepted, later tests or outcomes show whether the advice was right.
For Ray
This part tells Ray how to find recommendations in an input, such as a test readout, a findings deck, a research report, a retro note or a design review email, and how to store them.
How to find one
-
Mark suggested actions. Read the input and mark every statement that proposes what the team should do next.
-
Find the action. State it as one thing to do. If a statement bundles several actions, split them.
-
Find the evidence. Identify the signals, observations or metrics behind it, and link to them. If no evidence is given, leave it blank.
-
Find who it's for. Note the role or team expected to act on it, or who owns the related decision.
-
Link it to the round of testing. Identify the Findings, test and concept it came from.
-
Check whether it's already a decision. If the input shows someone accountable has already made the call, store a Decision instead and link the recommendation to it.
-
Write the sentence. Use the format: [Action] because [evidence].
-
Fill the record, score it, link it and store it using the rules below.
Cues to look for
-
Advice language: "we recommend", "we suggest", "next step should be", "consider", "we should"
-
Recommendations or next steps slides at the end of a readout
-
An action paired with a reason from testing: "because participants..."
-
Proposals to iterate, retest, ship or drop a concept
What doesn't count
<table xmlns="http://www.w3.org/1999/xhtml" style="min-width: 50px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>If the statement...</p></th><th colspan="1" rowspan="1"><p>Store it as</p></th></tr><tr><td colspan="1" rowspan="1"><p>Records a call that someone accountable has made</p></td><td colspan="1" rowspan="1"><p>Decision</p></td></tr><tr><td colspan="1" rowspan="1"><p>Describes a proposed solution with a name and shape</p></td><td colspan="1" rowspan="1"><p>Concept</p></td></tr><tr><td colspan="1" rowspan="1"><p>Says what the evidence means without suggesting an action</p></td><td colspan="1" rowspan="1"><p>Signal</p></td></tr><tr><td colspan="1" rowspan="1"><p>Suggests an action with no testing behind it</p></td><td colspan="1" rowspan="1"><p>Hunch or Ask</p></td></tr><tr><td colspan="1" rowspan="1"><p>Names something to keep monitoring</p></td><td colspan="1" rowspan="1"><p>Watch item</p></td></tr></tbody></table>
The record
<table xmlns="http://www.w3.org/1999/xhtml" style="min-width: 75px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>Field</p></th><th colspan="1" rowspan="1"><p>What to fill in</p></th><th colspan="1" rowspan="1"><p>Values</p></th></tr><tr><td colspan="1" rowspan="1"><p>Statement</p></td><td colspan="1" rowspan="1"><p>The one-sentence recommendation</p></td><td colspan="1" rowspan="1"><p>[Action] because [evidence].</p></td></tr><tr><td colspan="1" rowspan="1"><p>Action</p></td><td colspan="1" rowspan="1"><p>What should be done</p></td><td colspan="1" rowspan="1"><p>Short text</p></td></tr><tr><td colspan="1" rowspan="1"><p>Evidence</p></td><td colspan="1" rowspan="1"><p>The signals, observations or metrics behind it</p></td><td colspan="1" rowspan="1"><p>Links to Signal, Observation or UX Metric</p></td></tr><tr><td colspan="1" rowspan="1"><p>Concept</p></td><td colspan="1" rowspan="1"><p>The concept it concerns, if any</p></td><td colspan="1" rowspan="1"><p>Link to Concept, or blank</p></td></tr><tr><td colspan="1" rowspan="1"><p>For</p></td><td colspan="1" rowspan="1"><p>The role or team expected to act</p></td><td colspan="1" rowspan="1"><p>Role or team</p></td></tr><tr><td colspan="1" rowspan="1"><p>Findings</p></td><td colspan="1" rowspan="1"><p>The round of testing it came from</p></td><td colspan="1" rowspan="1"><p>Link to Findings</p></td></tr><tr><td colspan="1" rowspan="1"><p>Suggested by</p></td><td colspan="1" rowspan="1"><p>The person or role who proposed it</p></td><td colspan="1" rowspan="1"><p>Name or role</p></td></tr><tr><td colspan="1" rowspan="1"><p>Source</p></td><td colspan="1" rowspan="1"><p>The input and exact passage or slide</p></td><td colspan="1" rowspan="1"><p>Link to Input</p></td></tr><tr><td colspan="1" rowspan="1"><p>Status</p></td><td colspan="1" rowspan="1"><p>Where it stands</p></td><td colspan="1" rowspan="1"><p>Suggested, Captured, Accepted, Changed, Declined</p></td></tr><tr><td colspan="1" rowspan="1"><p>Confidence</p></td><td colspan="1" rowspan="1"><p>How clearly it was stated and supported</p></td><td colspan="1" rowspan="1"><p>High, medium, low</p></td></tr></tbody></table>
New recommendations start as Suggested. Only a linked Decision can move one to Accepted, Changed or Declined.
Scoring confidence
Confidence measures how clearly the recommendation was stated and how directly it traces to evidence, not how good the idea sounds.
<table xmlns="http://www.w3.org/1999/xhtml" style="min-width: 50px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>Level</p></th><th colspan="1" rowspan="1"><p>When to use it</p></th></tr><tr><td colspan="1" rowspan="1"><p>High</p></td><td colspan="1" rowspan="1"><p>A specific action, stated in writing, with the evidence behind it named and traceable to this round of testing</p></td></tr><tr><td colspan="1" rowspan="1"><p>Medium</p></td><td colspan="1" rowspan="1"><p>The action is clear, but the evidence is only implied or comes from a summary</p></td></tr><tr><td colspan="1" rowspan="1"><p>Low</p></td><td colspan="1" rowspan="1"><p>The action is vague, or no evidence from testing is given</p></td></tr></tbody></table>
If any part of the record can't be filled from the source, leave it blank and lower confidence by one level.
Linking
-
Check for duplicates first. If an existing recommendation suggests the same action for the same concept, add the new evidence and source to it instead of creating another.
-
Link to its evidence and to its source input.
-
Link to its Findings and concept.
-
Link to the decision that accepts, changes or declines it, once one exists.
-
Note conflicts. If two recommendations suggest opposite actions, such as ship and retest, link them as conflicting.
When to store and when to ask
-
Store automatically when confidence is high.
-
Add to the review digest when confidence is medium or low, when it conflicts with another recommendation, or when it recommends a bet or another high-stakes change.
When to flag
-
A recommendation has no evidence linked to it
-
A recommendation has been waiting more than 30 days with no decision
-
Two recommendations from the same Findings suggest opposite actions
-
A recommendation is being acted on as if it were a decision
-
Findings were completed with no recommendation at all
Worked example
Input: a design review email from the lead researcher after the SecondBank rewards value test.
"Most people couldn't convert points into anything real; P4 literally asked if the number was money. We should show point value in dollars next to the balance. I'd also retest the progress bar with that change before Q1."
What Ray finds:
-
"Show point value in dollars" is the action.
-
"Couldn't convert points" and P4's question are the evidence, already stored as a signal and an observation.
-
"Retest the progress bar" is a second action, so it becomes a separate recommendation.
-
No one has made the call yet, so these stay recommendations for the product lead to decide on.
Stored record:
<table xmlns="http://www.w3.org/1999/xhtml" style="min-width: 50px;"><colgroup><col style="min-width: 25px;"><col style="min-width: 25px;"></colgroup><tbody><tr><th colspan="1" rowspan="1"><p>Field</p></th><th colspan="1" rowspan="1"><p>Value</p></th></tr><tr><td colspan="1" rowspan="1"><p>Statement</p></td><td colspan="1" rowspan="1"><p>Show point value in dollars because participants couldn't convert points.</p></td></tr><tr><td colspan="1" rowspan="1"><p>Action</p></td><td colspan="1" rowspan="1"><p>Show point value in dollars next to the balance</p></td></tr><tr><td colspan="1" rowspan="1"><p>Evidence</p></td><td colspan="1" rowspan="1"><p>Signal: point value is unclear; Observation: P4 asked if the number was money</p></td></tr><tr><td colspan="1" rowspan="1"><p>Concept</p></td><td colspan="1" rowspan="1"><p>Points progress bar</p></td></tr><tr><td colspan="1" rowspan="1"><p>For</p></td><td colspan="1" rowspan="1"><p>Product lead</p></td></tr><tr><td colspan="1" rowspan="1"><p>Findings</p></td><td colspan="1" rowspan="1"><p>Rewards value test, round 1</p></td></tr><tr><td colspan="1" rowspan="1"><p>Suggested by</p></td><td colspan="1" rowspan="1"><p>Lead researcher</p></td></tr><tr><td colspan="1" rowspan="1"><p>Source</p></td><td colspan="1" rowspan="1"><p>Design review email, October 2026</p></td></tr><tr><td colspan="1" rowspan="1"><p>Status</p></td><td colspan="1" rowspan="1"><p>Captured</p></td></tr><tr><td colspan="1" rowspan="1"><p>Confidence</p></td><td colspan="1" rowspan="1"><p>High: specific action with named evidence from this round</p></td></tr></tbody></table>
Next step: stored automatically and added to the round 1 Findings, with a prompt to the product lead to accept, change or decline it.
Edge cases
-
Recommendation dressed as a decision. "We're going with dollars" from a researcher who doesn't own the call is still a recommendation. Store it that way and prompt the owner.
-
Many actions in one line. "Add dollars, redesign the tab and email everyone" is three recommendations. Split them so each can be decided on.
-
No evidence given. A suggestion with no testing behind it is a Hunch or an Ask. Don't store it as a recommendation.
-
Declined advice. Keep declined recommendations linked to their decision. If later evidence supports them, they can be raised again.