GUIDE · NEED VALIDATION

When a user says “I want it”, is it a need?

Everyone is polite in an interview, and everyone ticks "willing" in a survey. Judging whether a need is real comes down to one criterion: is there observed behavior, or only a statement? This guide gives you the four common misreadings and how to spot them, a behavior checklist you can use without installing anything, and a way to grade your conclusions.

Start with the conclusionTHE ONE-LINE RULE

Saying it does not count; doing it does. A verbal statement ("I want it", "I would buy") is evidence, but only the weakest level (E1: the problem is stated); observed substitute behavior, workarounds, and payment are what make E2 and E4+/E5 evidence. A need backed only by words stays at "needs validation", however often it is repeated.

Four common misreadingsFOUR MISREADINGS

MisreadingWhat it looks likeHow to spot it
Taking politeness for a needThe person nods and says "that sounds good, I'd want it"Follow up: "What did you actually do last time this problem came up?" If no concrete behavior comes back, it is politeness, not a need
Taking a feature request for a painThe user hands you the solution: "just add Excel export"Ask: "which step of your current process would that replace?" — first recover the pain and scene that were skipped
Taking one case for a populationOne heavy user uses it daily, so "the market is large" followsA single case can show a need exists (E2), not how common it is; prevalence needs a market estimate with a complete denominator, and the two are written separately
Taking intent for payment60% of survey respondents tick "willing to pay"Intent is E1. Price anchors, real purchases, continued use, and repurchase are E4+/E5 — stated intent is never paying behavior

Substitute-behavior checklistE2 SIGNALS CHECKLIST

Look for these behavioral traces in your target users' communities, tickets, and interview notes — any single piece of E2 evidence is harder than ten sentences of "I want it":

When you search a target community, use the language of the problem rather than the language of the solution: how do I, workaround, spreadsheet, 有没有办法, 手动 — what follows those words is the real scene.

Write your conclusions down, gradedGRADE YOUR CONCLUSIONS

Evidence-strength grading (E0–E5; the full definitions Askovo uses internally are in the glossary):

LevelEvidence formQuestion answered
E0Activity, role, or scene contextThe scene occurs
E1Unmet job, goal, or difficultyThe problem is stated
E2Current practice, workaround, failure, or switching costSubstitutes and their friction
E3Explicit acceptance of or preference for the research conceptThe concept is accepted under stated conditions
E4+Price anchor, purchase or payment intentDirect commercial intent
E5Paid possession, continued use, repurchaseRealized behavioral evidence

Two hard rules: a need judgment reaches "validated" only with at least one piece of E2 (observed behavior) — E1 verbal statements alone stop at "needs validation"; and commercial judgments accept only E4+/E5 — counter-evidence and substitute behavior will not lift them to "observed".

Three steps you can take todayDO IT WITHOUT ANY TOOL

  1. Pick one target community (GitHub issues, a Reddit sub, a vertical forum), search it for 30 minutes in the language of the problem, and note the scenes that come up at least 3 times.
  2. Collect 5 pieces of E2 evidence for each scene (workaround / self-build / paying), each with a link to the original text — no link, no entry.
  3. Write the conclusions on one page, graded E0–E5: which judgments have E2 behind them, which have only E1, which have no evidence at all. A cell with no evidence says "unknown", not zero.

These three steps need no tooling. Their value is this: before you write the first line of code, you already know which judgment holds up.

Where Askovo automates part of thisWHERE THE PRODUCT HELPS

In the process above, the two steps that cost the most manual effort and drift most easily are "collection" and "not overclaiming": collecting a large corpus from authorized sources, registering the evidence behind every judgment, and naming each conclusion that outruns its evidence. askovo do "<your question>" --dry-run first shows how it intends to break the question down (runs locally, uploads nothing); askovo demo runs the synthetic sample in a second.

The boundary is stated here too: the synthetic sample proves that the structure and the reasoning chain are self-consistent, not real demand or a market; semantic review always stays with humans. The same discipline applies to Askovo itself — see the claims page.

Verify it locally in 30 seconds Commission a real research task