The Re-Research Audit: How Much of Your Budget Buys an Answer You Already Own

Short answer: run this audit on your last twenty studies and count how many were commissioned to answer a question your organisation had already answered and could no longer find. That percentage is your re-research rate, and it is the only research-waste number an executive will accept without argument, because it is measured against your own archive rather than an industry benchmark.

Most research functions have never computed it. The reason is not laziness. It is that no role in the organisation is accountable for knowing what the organisation already knows. The requester is measured on shipping, the researcher on delivering the study they were asked for, the ops lead on throughput, the executive on outcomes. Everyone has an incentive to add a study. Nobody's performance review contains a line about not buying the same answer twice — which makes duplicate research the rare kind of waste with no natural opponent.

First, separate the two reasons a question gets asked twice

This distinction is the whole audit, and getting it wrong makes the number meaningless.

Re-running a pricing study eighteen months after a repositioning is good practice. Re-running an onboarding study because the last one is a slide deck in a Slack thread and nobody is sure what its sample was — that is a retrieval failure wearing a research budget.

The audit measures only the second column. Do not let it drift into an argument about how often findings should be refreshed; that is a different and already-settled question.

The audit protocol

Budget half a day. Do it with one researcher and one person from the requesting side, because the two of you will disagree, and the disagreements are where the findings are.

Step 1 — Take the last twenty studies. Not a sample you choose; the last twenty in chronological order. Selection here is how audits get flattering results.

Step 2 — For each, write the question in one sentence. The decision question, not the method. "Should we require SSO setup during onboarding?" not "enterprise onboarding interviews".

Step 3 — Search the archive for that question, dated before the study started. Use whatever retrieval your team actually has. Timebox it to ten minutes per study — the same ten minutes a requester would realistically have spent.

Step 4 — Classify each study into one of four outcomes:

• A. Novel — no prior work bore on the question.
• B. Refresh — prior work existed and had legitimately expired.
• C. Re-research — prior work existed, was still valid, and was not used.
• D. Partial — prior work would have narrowed the study substantially, even if it could not replace it.

Step 5 — For every C and D, record why the prior work was not used. Force a single cause from the list in the next section. This is the field that turns the audit into an action plan.

Step 6 — Attach a cost to C and D. Use your own numbers, not an industry figure. For each study: incentives paid, recruitment or panel cost, plus loaded hours for design, fieldwork, analysis, and readout. Count D at half weight, since the study would have happened in some reduced form anyway.

Step 7 — Report one number. Re-research rate = (C + 0.5 × D) ÷ 20, expressed as a percentage, with the cost total beside it.

What the number means

There is no published benchmark for this and you should be suspicious of anyone who offers one — the figure depends entirely on your archive's age and your team's turnover. What matters is the internal comparison:

• Under 10% — your retrieval works. Spend your effort elsewhere.
• 10–25% — normal for a team with two or more years of accumulated research and no description discipline. This is where most functions land.
• Over 25% — you are funding an archive you cannot use. The cheapest available improvement in your research program is not better methods; it is making the last two years findable.

Run it annually. The trend is more informative than the level, and it is the only research-ops metric that gets better as your archive gets older — if the underlying practice is fixed.

The four causes, and which are worth fixing

Every C and D traces to one of these. In rough order of how often they appear:

1. The study was never described at a level anyone could evaluate. Prior work exists as tagged fragments, and the requester could not tell in thirty seconds whether it applied to them. Fix: a study-level description for anything ever cited.

2. The finding was found but not trusted. Someone did surface the old study and chose not to rely on it, usually because they could not see who was in it or what it excluded. Fix: the same description, specifically the who-was-not-in-it and what-it-cannot-support fields.

3. The evidence had been stripped of context. The old work survives only as quotes or a headline number, and a decontextualised quote is not usable to defend a decision. Fix: preserve the bond between a quote and its interview.

4. The person who knew left. The finding lived in a head, not a record. This is the one cause you cannot fix retroactively — you can only stop it from recurring, and the mechanism is the same description discipline.

Notice that none of the four are solved by doing more research, buying a panel, or hiring another researcher, which is exactly why the problem persists in well-funded teams.

Why nobody owns this

Every other kind of research waste has a natural opponent. A badly designed study gets challenged by a researcher. An over-scoped study gets challenged by whoever holds the budget. A biased study gets challenged in the readout.

Duplicate research has no opponent because it fails in the gap between roles rather than inside one. The requester genuinely does not know the answer exists — that is what makes them a requester. The researcher is doing exactly their job by running the study they were asked to run; refusing on the grounds that the answer is already in the repository is an awkward act with no mandate behind it. The repository owner is measured on intake and search adoption, not on studies prevented. The executive sees a research request and a research output, which look like a functioning system.

A duty that belongs to everyone belongs to no one. The audit's real output is not the percentage — it is the assignment of that duty to a named role and a specific moment.

Assigning the duty: the pre-commission check

Give the duty to the intake step, because it is the only moment when preventing a study is cheaper than running it.

Add one gate to your research intake process. Before a study is scheduled, someone other than the requester spends ten minutes searching the archive and writes one of three answers into the request:

• "No prior work found" — proceed.
• "Prior work found, still valid: [link]" — the requester must read it and either withdraw the request or narrow it in writing.
• "Prior work found, expired because [reason]" — proceed as a refresh, and reuse the old question set so the two waves are comparable.

Three rules make this work rather than becoming theatre:

1. The check is done by someone other than the requester. People cannot search for what they do not know exists, and they are poor judges of whether their question has been asked before.
2. The answer is written into the request record, so it is auditable next year.
3. Withdrawing a request counts as a delivery. If the only visible output of research is studies completed, no one will ever prevent one. This is the incentive fix, and without it the other two are decoration.

How Koji changes the economics

Two of the four causes are structural, and closing a research stack removes them rather than mitigating them:

• Cross-study search is a first-class operation. Semantic and keyword search across every transcript means the ten-minute pre-commission check is realistic rather than aspirational — the difference between a gate that runs and one that gets skipped.
• Descriptions start populated. Briefs, recruitment criteria, fieldwork dates, and the question set are structured objects, so the study record exists before anyone writes prose.
• Evidence keeps its context. Quotes stay linked to the question that produced them and to the interview they sat in, which is what makes an eighteen-month-old finding defensible rather than merely quotable.
• Structured questions make old and new waves comparable. All six types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — store answers in a fixed shape, so a legitimate refresh can reuse the prior question set exactly and produce a real trend instead of two unrelated snapshots.
• The marginal cost of a small confirmatory study collapses. When re-fielding twelve interviews to check whether an old finding still holds costs hours rather than weeks, the choice stops being "trust the stale study or fund a new one" and becomes a cheap verification — which is the correct answer to most Category B cases.

The uncomfortable implication for any research platform, including this one: a repository that only makes it easier to add is not solving this problem. The test is whether it makes prior work evaluable by someone who was not there.

Frequently asked questions

How is this different from proving research ROI?

Research ROI argues the value of the function to people deciding its budget, and it depends on attributing outcomes to insights — contestable, and usually argued rather than measured. The re-research rate is narrower and harder to dispute: it counts studies you paid for against answers you already owned, entirely within your own records. It is a diagnostic for the research team, not a pitch for the executive, though it tends to land well with executives precisely because it is self-critical.

Is not some repetition healthy?

Yes — replication is a virtue, and a deliberate replication is a Category A or B study, not re-research. The distinguishing feature of re-research is that nobody decided to repeat anything. The old work was invisible at the moment of commissioning. If your team knowingly re-runs a study to confirm a finding, record the reason and count it as intentional.

What if our archive is only a year old?

Then run the audit on the last ten studies and expect a low number. The value at this stage is establishing the baseline and the intake gate before the archive gets big enough for the problem to appear — which is typically around the two-year mark, or immediately after the first researcher departure.

Who should run the audit?

Someone who did not commission the studies being audited. A researcher auditing their own intake decisions will systematically classify borderline cases as Category A, not from dishonesty but because they remember the reasoning that made each study feel necessary. Pairing with a requester-side colleague is the practical version of research independence.

Does this mean we should re-run fewer studies?

Not necessarily — it means you should know which column each re-run belongs in. A team with a 20% re-research rate and a good refresh cadence should not run less research; it should redirect the wasted fifth into new questions. The output of this audit is a reallocation, not a cut.

How long does the audit take?

Half a day for twenty studies once your archive is searchable, most of it in step 3. If step 3 alone takes more than ten minutes per study, stop the audit and record that fact — it is a stronger finding than any percentage you were going to produce, because it means your requesters have no realistic chance of finding prior work either.

Related Resources

• How Long Is User Research Valid? — insight decay, the legitimate reason to re-run a study
• The Study-Level Description — the fix for the most common cause of re-research
• Why a Tagged Quote Is Not Evidence — why surfaced findings still fail to be trusted
• Research Request and Intake Process — where the pre-commission check belongs
• Search Across Interview Transcripts — what makes a ten-minute check realistic
• Structured Questions Guide — the six question types that make a refresh comparable to the original wave