Frameworks & methodology: how to turn customer research into revenue decisions

Thirty transcripts sit in a shared drive. The pricing committee meets Friday. The question on the agenda is whether a new integration goes into the mid tier with a price uplift or stays locked in the top tier as a reason to upgrade, and the only thing between those interviews and that decision is a method for reading them. That method is what this lesson gives you: how many people you talk to, how you code what they said, and how a coded finding becomes a number a finance team will sign off on.

From transcript to a decision with a number

A research programme that ends in themes has stopped one step short. The output of a study should be a decision candidate written in the same grammar as a business case: the finding, the segment it applies to, the number of accounts affected, the move (price, package, roadmap slot, or messaging), and the revenue at stake with its assumptions visible. Everything below is in service of that one sentence. Assume the evidence split the foundations lesson draws between what people say and what they do, and assume the progress-making frame it sets out for why customers switch. This rung is about arithmetic and sampling discipline.

Sub-concept 1: how many interviews, and which ones

The two questions teams get wrong are sample size and sample frame, and the second one costs more.

On size, the academic work is more useful than the folklore. Guest, Bunce and Johnson's 2006 study found that in a reasonably homogeneous sample, new codes had almost stopped appearing by the twelfth interview, with most of them showing up in the first six. Griffin and Hauser's voice-of-the-customer research put the number higher, in the range of 20 to 30 interviews per segment, to surface something like 90% of the needs you will ever hear. The reconciliation is the word "segment". Twelve interviews will saturate one buying situation. If you sell to three (self-serve teams, mid-market with a security review, enterprise with procurement), you need twelve to fifteen in each, so 40 interviews, not twelve. Teams that pool 15 interviews across three segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition → end up with themes that are true of nobody in particular and a positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition → claim that survives no sales conversation.

On frame, run a quota rather than a convenience list. A workable split for a 24-interview study: eight recent wins, six competitive losses, six churned accounts, four evaluations that ended in no decision. The last group is the one everybody skips and the one that explains flat pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition →, because no-decision buyers tell you which internal problem your product failed to outrank. Interviewing only happy customers produces research that reliably recommends what you already built.

Cadence matters more than volume. A weekly interview habit, which Teresa Torres codified in Continuous Discovery Habits and trains teams to run, keeps recruitment machinery warm so that a pricing question in March does not require six weeks of scheduling before anyone can answer it.

Sub-concept 2: the switch interview

Bob Moesta's switch interview targets the moment of change: from a competitor to you, from you to a competitor, or from nothing to buying. Four questions in sequence: when did the first thought arrive that something had to change, what turned that into an active search, why this product over the shortlist, and what almost stopped the purchase. You are reconstructing a timeline, not collecting product feedback, so you push for dates, calendar events, and who else was in the room.

Drift's move from live chat to conversational marketing is the cleanest public example of what that timeline produces. The problem their buyers described was not chat as a feature. It was the gap between a prospect filling in a form and a human replying, sometimes days later, by which point intent had gone cold. That reading pointed the roadmap at real-time routing to sales reps rather than at chat features, and it pointed the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → at removing lead capture forms from their own site, which then became part of the positioning. Note the shape: one timeline finding, two decisions, one of them a roadmap trade-off with engineering cost attached.

Sub-concept 3: coding transcripts so a finding can be argued with

Coding is where most research dies quietly. The method that survives scrutiny:

Pass one, open codes. Tag every distinct statement of a problem, a workaround, a trigger, or a rejected alternative. Twenty-four interviews typically yield 60 to 120 raw codes. Do this in a spreadsheet with one row per code instance and columns for account, segment, ARRARRAnnual Recurring Revenue (ARR) is the normalized, predictable revenue a subscription business expects to earn from active contracts over a single year.View full definition →, and role.

Pass two, collapse to candidates. Merge codes that describe the same job in different vocabulary. Then apply two thresholds before anything becomes a decision candidate: the code appears unprompted in at least a third of interviews within one segment, and the accounts carrying it hold at least a quarter of that segment's revenue. Count accounts, never mentions. One articulate customer who repeats a complaint eleven times will otherwise outvote nine quieter ones.

Pass three, look for the disconfirming case. For each candidate, name the interview that contradicts it and write why. A candidate with no counter-example usually means you only sampled people who agreed with you.

The second-order effect is worth planning for. Once you name a job, sales starts qualifying against it, so within two quarters your inboundinboundA strategy that attracts prospects organically via valuable content (blog, SEO, social) rather than interrupting them.View full definition → pipeline skews toward buyers who already have that job. Your next study, sampled from that pipeline, will confirm the finding by construction. Hold back a recruitment channel that is not your own funnel (a panel, a partner list, a competitor's user community) or the research becomes an echo.

Sub-concept 4: attaching the revenue number

A worked packaging example. The coded finding: mid-tier buyers in the security-review segment consistently name an audit log as the thing blocking rollout beyond one team. Segment has 320 accounts, 55% on the mid tier, average 35 seats. Moving the feature down a tier with a $5 per seat per month uplift gives 320 × 0.55 × 35 × 5 × 12, about $370k of gross new ARR. Apply a haircut for accounts that refuse the uplift or renegotiate, say half, and you are arguing for roughly $185k against a build cost of two engineers for a quarter. Write the ratio down as a rule before you run the study: if expected annual revenue does not clear build cost by 3x, the finding is real but it does not get the roadmap slot this quarter.

Then a quantitative gate before you spend on acquisition. Sean Ellis, who used it across Dropbox, LogMeIn and Eventbrite before founding GrowthHackers, popularised the 40% test: ask users how they would feel if they could no longer use the product, and if fewer than 40% say "very disappointed", scaling paid spend buys you churn. Two edge cases matter. In B2B the respondent is usually the user while the renewal is signed by procurement, so the score can sit at 55% while a contract dies in a committee nobody surveyed. And the question is asked of current users, which excludes everyone who already left, so the number is structurally optimistic. Run it per segment against the persistent profile the go-to-marketgo-to-marketThe strategy defining how you'll launch a product: target segments, channels, value proposition and coordinated action plan.View full definition → foundations lesson describes, not as a single company-wide figure.

Real-world cases

IKEA's entry into India shows the full chain. Before opening in Hyderabad in 2018, the company reported carrying out around a thousand home visits across Indian cities, watching how people cook, store clothes and use small kitchens. The findings did not stop at insight. They became assortment and price decisions: hundreds of items priced under 200 rupees, cookware sized for local pans, and a restaurant menu built around Indian food. Each of those is a margin and volume call with a forecast behind it, made on ethnographic evidence rather than on a transplanted European range.

Drift's case shows the discipline of stopping at one decision. The switch interviews said the pain was response delay, so the visible bets were routing and form removal. Had the same transcripts been coded on mentions, chat customisation requests would have won, and the roadmap would have shipped features for the category the company was trying to leave.

Action items

• Rewrite your last research readout as decision candidates: finding, segment, accounts affected, proposed move, revenue at stake, assumptions listed. Anything that will not fit that format goes back for more interviews.
• Set your sample quota now for the next study, including six no-decision evaluations, and book a non-funnel recruitment channel so the sample cannot drift toward your own pipeline.
• Agree the thresholds in advance with product and finance: the share of interviews a code must appear in, the revenue coverage it needs, and the expected-return multiple required for a roadmap slot.

Common mistakes that kill results

Running research after the brief is written. Validation research produces confirming quotes because the interviewer already knows the answer, and it spends customer goodwill on a decision that was never open.

Coding by mention count. This is the most common arithmetic error in qualitative work and it hands your roadmap to whoever talks longest. Deduplicate to accounts first, then weight by revenue.

Treating satisfaction data as buyer motivation. NPSNPSNet Promoter Score (NPS) measures customer loyalty by asking how likely customers are to recommend a brand, then subtracting detractors from promoters.View full definition → and CSAT measure how existing users feel about what they already have. Neither one tells you the trigger, the shortlist, or the alternative that nearly won, so positioning built on them optimises for feature contentment.

Pooling segments to hit a sample size. Twelve interviews spread across self-serve, mid-market and enterprise buyers saturate nothing and average away the very differences that pricing and packaging decisions depend on.