An assumption is a conclusion the team has drawn from data and is treating as true, before anyone has checked it.
For example, the analytics show that few new cardholders open the rewards tab. From that, the team concludes that cardholders don't know rewards exist. The numbers are real. The conclusion might not be. People could know about rewards and simply not care, or not find them worth the effort.
Why it matters
Assumptions are easy to miss because they come with numbers attached. They sound like facts, so nobody stops to question them, and teams go on to build plans, designs and budgets on top of them.
When an assumption turns out to be wrong, everything built on it is at risk. Catching them early is one of the cheapest ways to avoid an expensive mistake.
Notice when the team is treating a conclusion as a fact, write it down plainly, and make sure it gets checked before an important decision depends on it.
How we write one
Every assumption is written as one sentence that keeps the data and the belief side by side:
Based on what the data shows, we assume what we believe it means.
Based on low visits to the rewards tab, we assume cardholders don't know rewards exist.
Writing it this way makes the leap from data to conclusion easy to see, and easy to challenge.
How it differs from similar ideas
-
An observation is what we saw: few people open the rewards tab.
-
An assumption is what we think it means: they don't know rewards exist.
-
A hunch is a gut feeling that may have no data behind it at all.
-
A signal is an assumption that has been tested and held up.
A good assumption
-
Points to specific data
-
Makes one claim, not several
-
Keeps the data and the conclusion separate
-
Can be checked with a reasonable test
What happens to it
-
Raised. Someone, or Ray, spots a conclusion being treated as fact and writes it down.
-
Checked. The team designs a quick test, such as a few customer interviews or a short survey.
-
Confirmed or disproven. If it holds up, it becomes trusted evidence. If it doesn't, every plan and decision that relied on it gets reviewed.
-
Retired. Once it no longer matters, it's kept on file for history.
For Ray
This part tells Ray how to find assumptions in an input, such as a report, an analytics export, a transcript or a deck, and how to store them.
How to find one
-
Mark interpretations. Read the input and mark every claim that explains what data means, rather than just reporting it.
-
Split each claim. Separate the data (what was measured or seen) from the conclusion (what it's taken to mean).
-
Check the data is there. The data must appear in the input itself. If it doesn't, the claim is a Hunch.
-
Check there's a leap. The conclusion must go beyond the data. If it only restates the data, it's an Observation.
-
Check it hasn't been tested. If a test has already confirmed it, it's a Signal.
-
Write the sentence. Use the format: Based on [data], we assume [conclusion].
-
Fill the record, score it, link it and store it using the rules below.
Cues to look for
-
Confident words next to numbers: "clearly", "obviously", "this shows", "the data suggests", "proves"
-
A statistic followed by "because", "so" or "which means"
-
Statements about what users want, know, feel or prefer, backed only by behavior data
-
Recommendations that rest on an unstated reason
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 claim…</p></th><th colspan="1" rowspan="1"><p>Store it as</p></th></tr><tr><td colspan="1" rowspan="1"><p>Has no data behind it</p></td><td colspan="1" rowspan="1"><p>Hunch</p></td></tr><tr><td colspan="1" rowspan="1"><p>Only reports what the data shows</p></td><td colspan="1" rowspan="1"><p>Observation</p></td></tr><tr><td colspan="1" rowspan="1"><p>Has already been confirmed by a test</p></td><td colspan="1" rowspan="1"><p>Signal</p></td></tr><tr><td colspan="1" rowspan="1"><p>Is a fixed limit, such as budget or timeline</p></td><td colspan="1" rowspan="1"><p>Constraint</p></td></tr><tr><td colspan="1" rowspan="1"><p>Proposes a solution</p></td><td colspan="1" rowspan="1"><p>Concept</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 assumption</p></td><td colspan="1" rowspan="1"><p>Based on [data], we assume [conclusion].</p></td></tr><tr><td colspan="1" rowspan="1"><p>Data</p></td><td colspan="1" rowspan="1"><p>The data it rests on, quoted or summarized</p></td><td colspan="1" rowspan="1"><p>Short text</p></td></tr><tr><td colspan="1" rowspan="1"><p>Source</p></td><td colspan="1" rowspan="1"><p>The Input and the exact passage, chart or slide</p></td><td colspan="1" rowspan="1"><p>Link to Input</p></td></tr><tr><td colspan="1" rowspan="1"><p>Audience</p></td><td colspan="1" rowspan="1"><p>Who the belief is about</p></td><td colspan="1" rowspan="1"><p>Link to Audience</p></td></tr><tr><td colspan="1" rowspan="1"><p>User need</p></td><td colspan="1" rowspan="1"><p>The need it relates to, if clear</p></td><td colspan="1" rowspan="1"><p>One of the 20 user needs, or blank</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>Untested, In test, Supported, Contradicted, Retired</p></td></tr><tr><td colspan="1" rowspan="1"><p>Confidence</p></td><td colspan="1" rowspan="1"><p>How well the data supports the conclusion</p></td><td colspan="1" rowspan="1"><p>High, medium, low</p></td></tr><tr><td colspan="1" rowspan="1"><p>Stakes</p></td><td colspan="1" rowspan="1"><p>How much would go wrong if it's false</p></td><td colspan="1" rowspan="1"><p>High, medium, low</p></td></tr><tr><td colspan="1" rowspan="1"><p>Raised by, date</p></td><td colspan="1" rowspan="1"><p>The person or asset it came from, and when</p></td><td colspan="1" rowspan="1"><p>Standard on every noun</p></td></tr></tbody></table>
New assumptions always start as Untested.
Scoring confidence
Confidence measures the size of the leap, not how sure the author 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>The data is specific and recent, and the conclusion is one small step from it</p></td></tr><tr><td colspan="1" rowspan="1"><p>Medium</p></td><td colspan="1" rowspan="1"><p>The data is real, but other explanations are just as plausible</p></td></tr><tr><td colspan="1" rowspan="1"><p>Low</p></td><td colspan="1" rowspan="1"><p>The data is vague, old or from a small sample, or the conclusion goes well beyond it</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.
Judging the stakes
<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 live Decision, Bet or Concept depends on it</p></td></tr><tr><td colspan="1" rowspan="1"><p>Medium</p></td><td colspan="1" rowspan="1"><p>It shapes an open Test, Opportunity or Initiative</p></td></tr><tr><td colspan="1" rowspan="1"><p>Low</p></td><td colspan="1" rowspan="1"><p>Nothing depends on it yet</p></td></tr></tbody></table>
Stakes can change. Recheck them whenever a new link is added.
Linking
-
Check for duplicates first. If an existing assumption makes the same claim about the same audience, add the new source to it instead of creating another.
-
Link to its source Input.
-
Link to its audience and, if clear, a user need.
-
Link to anything that depends on it: Tests, Concepts, Decisions and Bets.
-
Note conflicts. If another assumption draws the opposite conclusion from the same data, link the two as contradicting.
When to store and when to ask
-
Store automatically when confidence is high and stakes are low or medium.
-
Add to the review digest when confidence is low, stakes are high, or it contradicts another assumption.
When to flag
-
A Decision or Bet depends on an assumption that's still Untested
-
A high-stakes assumption has been Untested for more than 30 days
-
Two assumptions from the same data contradict each other
-
An assumption is Contradicted but a live plan still depends on it
Worked example
Input: a quarterly product review deck, slide 7.
"Only 12% of new cardholders visit the rewards tab in their first month. Clearly they don't know the program exists, so we should add a rewards banner to the home screen."
What Ray finds:
-
"Only 12% of new cardholders visit the rewards tab in their first month" is the data.
-
"They don't know the program exists" is the conclusion. It goes beyond the data, so it's an assumption.
-
"Add a rewards banner" is a solution, so it becomes a separate Concept, linked to this assumption.
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>Based on 12% of new cardholders visiting the rewards tab in their first month, we assume they don't know rewards exist.</p></td></tr><tr><td colspan="1" rowspan="1"><p>Data</p></td><td colspan="1" rowspan="1"><p>12% first-month rewards tab visits</p></td></tr><tr><td colspan="1" rowspan="1"><p>Source</p></td><td colspan="1" rowspan="1"><p>Q3 product review deck, slide 7</p></td></tr><tr><td colspan="1" rowspan="1"><p>Audience</p></td><td colspan="1" rowspan="1"><p>New cardholders, first 30 days</p></td></tr><tr><td colspan="1" rowspan="1"><p>User need</p></td><td colspan="1" rowspan="1"><p>Findable (Basics)</p></td></tr><tr><td colspan="1" rowspan="1"><p>Status</p></td><td colspan="1" rowspan="1"><p>Untested</p></td></tr><tr><td colspan="1" rowspan="1"><p>Confidence</p></td><td colspan="1" rowspan="1"><p>Medium: the data is specific, but low interest would explain it just as well</p></td></tr><tr><td colspan="1" rowspan="1"><p>Stakes</p></td><td colspan="1" rowspan="1"><p>High: the banner concept depends on it</p></td></tr><tr><td colspan="1" rowspan="1"><p>Raised by, date</p></td><td colspan="1" rowspan="1"><p>Q3 product review deck, October 2026</p></td></tr></tbody></table>
Flag raised: a concept depends on an untested, high-stakes assumption, so it goes to the review digest with a suggested test.
Edge cases
-
Two conclusions in one sentence. Split them into two assumptions that share the same data.
-
Stated as a question. "Do people even know about rewards?" is a Test question, not an assumption.
-
People disagree. If two people read the same data differently, store both and link them as contradicting.
-
Already part of the plan. If an assumption is buried inside a decision or roadmap, still pull it out. These are the most important ones to catch.