Acquisition.md
← Playbooks
Growth · 16 min

Customer Research Playbook

The interview → insight → action loop.

Every marketing team's real problem is that nobody has talked to a customer this month.

tl;dr
  • 2-week research sprints beat 90-day persona theatre every time.
  • Clay + Attio turns interviews into actions in 1 workflow.
  • 11x handles outreach; humans handle 8 hard questions and 1 decision.
  • PostHog + Common Room expose the 20% of users driving 80% of signal.
  • Framer + Relume ships the winning message before the next board meeting.
Action plan
Action plan0/3 · 0%
Progress saved locally. Sign in to sync

1. Build the pipeline before you book the first call

A 2026 research system starts in Clay, not in a calendar link. Pull 200 target accounts from Attio, enrich with Common Room, Apollo AI, and Keyplay, then use Clay AI prompts to score for trigger events, product usage, and revenue potential. Route the shortlist into 11x or Lindy for outreach, with n8n handling consent, reminders, and transcript capture. Every interview should land in Attio with the account, persona, and theme already tagged. If you cannot answer who, why now, and what changed in 30 seconds, the pipeline is broken. Cursor or Claude Code should own the SQL that joins interview notes to PostHog and activation data.

  1. 01

    Assemble the research universe

    Export 200 accounts from Attio, enrich in Clay with Common Room and Apollo AI, then add 3 trigger fields: hiring, usage, and funding.

    Clay prompt: 'Score accounts 1-5 for urgency using hiring, tech stack, and recent product event evidence.'
  2. 02

    Automate outreach and scheduling

    Use 11x Alice or Lindy to send 40 personalised invites and book 12 interviews without a human SDR queue.

    11x prompt: 'Invite 3 operators per account; ask for 25 minutes; include 1 reason now based on Clay signals.'
  3. 03

    Wire the data path

    Send recordings to Attio, transcripts to n8n, and summaries to Claude 4.5 Sonnet for structured tagging.

    n8n workflow: 'On meeting ended, fetch transcript, classify theme, write 5 fields to Attio, alert in Slack.'
IfThen
Clay cannot enrich 70% of the target list with at least 3 usable signalspause outreach and fix the data model before another interview is booked.
Attio does not store the interview outcome in 1 structured objectthe research programme is already dead and needs a schema reset.

2. Run interviews like a revenue system, not a therapy session

Each call needs 8 questions, 1 decision rule, and 1 explicit next action. Use a script in Claude Code or Cursor that adapts by persona: buyer, user, blocker, or champion. Record in Fireflies or Google Meet, then push the transcript into Claude 4.5 Sonnet for a forced output: pains, triggers, alternatives, budget, and buying process. 11x can follow up with no-show rescue and thank-you emails, while Lindy can request artefacts such as screenshots or pricing pages. The point is not notes; the point is a reproducible chain from verbatim quote to pipeline change in Attio.

  1. 01

    Standardise the question set

    Keep 8 core questions: trigger, current workaround, last purchase, stakes, budget owner, blockers, alternatives, and success metric.

    Prompt: 'Ask only 8 questions; stop after 25 minutes; extract exact phrases with quotes.'
  2. 02

    Capture verbatim and classify

    Push transcripts into Claude 4.5 Sonnet and ask for 1 theme, 1 subtheme, 3 quotes, and 1 risk per interview.

    Claude prompt: 'Return JSON with theme, quote, alternative, and buying-stage.'
  3. 03

    Automate follow-up

    Use 11x or Lindy to send a summary, request missing artefacts, and book the second call if buying intent is real.

    Lindy workflow: 'If budget owner mentioned, send calendar options for a pricing review within 3 days.'
IfThen
A call produces no exact quote and no buying triggermark it as invalid and do not count it in the sprint.
The prospect cannot name a workaround in under 20 secondsyou are too early or speaking to the wrong persona.

3. Turn 15 interviews into 3 decisions, not 30 slides

Synthesis in 2026 is a database job with a language model on top. Use Cursor or Claude Code to query Attio notes, PostHog events, and Common Room signals, then cluster by job-to-be-done, trigger, and objection. Clay Chat should generate a theme score, while n8n writes the winning themes back into Attio and posts the top 5 to Slack. The output is not a deck; the output is 3 decisions: what to say, who to target, and what to stop doing. If 2 patterns dominate 60% of interviews, the rest is noise. One operator plus Claude 4.5 Sonnet should replace the old 3-hour whiteboard ritual.

  1. 01

    Run theme clustering

    Export notes from Attio, use Claude 4.5 Sonnet to cluster into 5 buckets, then merge the low-signal buckets.

    Claude prompt: 'Cluster 15 interview notes into 5 themes and rank by revenue impact.'
  2. 02

    Join qualitative and product data

    Use Cursor SQL against PostHog to compare interview themes with feature adoption and retention cohorts.

    SQL rule: 'Join tagged themes to 30-day retention and flag themes with >15% churn delta.'
  3. 03

    Publish the decision memo

    Write a 1-page memo in Framer or Notion and push the 3 decisions into Attio as tasks owned by Growth, Product, and Sales.

    Decision memo template: 'Say this, target that, stop this.'
IfThen
A theme appears in <20% of interviewstreat it as anecdote unless it maps to a high-value account cluster in Clay.
Two themes explain >60% of painship only those 2 and freeze new research until the next sprint.

4. Ship changes fast enough to change what you hear

Research only compounds when the feedback loop is visible in product, pricing, and messaging. Use Framer + Relume to publish the new landing page in 1 day, Mutiny or Intellimize to personalise the headline by segment, and Attio to tag every affected account. PostHog should track whether the new message increases activation, while Common Room should show whether target accounts engage again. n8n can open a task when a theme crosses the threshold, and 11x can re-contact prior interviewees with the updated narrative. If the change did not move a metric in 14 days, the insight was weak or the action was too slow.

  1. 01

    Ship the message change

    Move the strongest theme into Framer + Relume and change 1 headline, 1 proof point, and 1 CTA.

    Headline rule: 'Mirror the exact phrase said by 3 buyers in interviews.'
  2. 02

    Track the effect in product and pipeline

    Use PostHog for conversion and Attio for pipeline movement by theme-tagged accounts.

    PostHog check: 'Compare activation by cohort before and after the landing page swap.'
  3. 03

    Close the loop with re-contact

    Ask 11x or Lindy to send the updated page to prior interviewees and collect 1-line reactions.

    11x prompt: 'Send update to all interviewees from segment A and ask if the new message matches their pain.'
IfThen
A new message lifts conversion by <10% after 14 daysrevert and test a different pain point, not a different colour.
Interviewees confirm the new message in 3 of 5 follow-upspromote it into outbound, ads, and the homepage immediately.
Pitfalls

What goes wrong

Mistake

Treating 5 interviews as truth

Fix

Use 12-15 interviews per segment and require 2 recurring patterns before changing strategy.

Mistake

Storing notes in Google Docs with no schema

Fix

Write structured fields into Attio and let Claude 4.5 Sonnet normalise the text.

Mistake

Hand-transcribing transcripts for 3 hours

Fix

Use n8n plus Claude Code to extract themes in under 10 minutes per call.

Mistake

Asking opinion questions instead of recall questions

Fix

Use 'tell me about the last time' and 'what happened next' on every call.

Mistake

Launching a new landing page without measuring the cohort

Fix

Track the affected accounts in PostHog and Common Room before and after the change.

FAQ
How many interviews are enough for a 2026 growth team?▾

For one segment, 12 to 15 interviews is the operating minimum. Fewer than 8 usually gives you founder bias, while more than 15 without a decision is procrastination. Use Clay to pre-qualify the list, then stop when 2 themes explain most of the pain and one message clearly wins.

Should product, marketing, and sales run separate research?▾

No, 1 Attio schema should serve all 3 functions. Split the questions by persona, not by department, or you will create 3 contradictory truths. Claude 4.5 Sonnet can generate role-specific summaries, while n8n writes the same tagged outcome into pipeline, onboarding, and homepage tasks.

What is the fastest way to get useful interviews?▾

Use Clay plus 11x or Lindy to recruit from a target account list, not from random inbound leads. Ask for a 25-minute working session, reference 1 trigger event, and offer 1 concrete takeaway. The fastest interviews are the ones where the prospect already feels the pain in the last 30 days.

How do I avoid false signals from loud customers?▾

Weight every interview by account fit and product relevance in Clay. A loud low-fit customer should not outrank a quiet high-fit buyer. Cursor SQL against PostHog can show whether the loud segment actually activates, retains, or expands, which is the only signal that matters.

What should we ship first after research?▾

Ship the message, then the onboarding, then the product. Framer + Relume lets you change the narrative in 1 day, while Mutiny or Intellimize can personalise it for the highest-value segment. If the message does not move conversion in 14 days, do not hide behind a roadmap.

Get one system like this in your inbox each week

No fluff, no upsell. Unsubscribe anytime.