Reuse a consistent method for sales-audience research
Package a repeatable research method so an assistant can analyze sales conversations and related lead metrics with the same business definitions each time.
How this works
Package a repeatable research method so an assistant can analyze sales conversations and related lead metrics with the same business definitions each time. Repeated audience analysis becomes comparable only when terms such as qualified lead, objection, and buying trigger keep the same meaning across runs. A reusable method should preserve those definitions while allowing new evidence to challenge old assumptions.
Join authorized sales-call evidence to lead and performance records at a declared grain. Extract passage-level observations first, then aggregate themes using the team's documented method. The model supports interpretation; joins, counts, and rates require deterministic checks.
Before you start
Research method
Obtain the current definitions, coding rubric, and decision questions.
Joinable records
Confirm call and lead identifiers and the permitted reporting period.
Analysis owner
Identify who reviews ambiguous coding and approves methodological changes.
Prepare the inputs
- transcripts
- Required
Call text with IDs
Preserve speaker turns and dates.
- lead_metrics
- Required
Rows at a declared grain
Include qualification status and outcome window.
- method
- Required
Versioned instructions
Define categories and exclusions.
- join_keys
- Required
Call-to-lead mapping
Detect one-to-many and missing matches.
- comparison_period
- Required
Date interval
Use mature, comparable outcomes.
Implementation, step by step
Freeze the method and sample
- Record the method version, period, and eligible call population.
- Keep later definition changes out of the current comparison unless the whole sample is recoded.
Checkpoint: A reproducible analysis specification.
Validate joins
- Connect transcripts to leads using approved identifiers.
- Check duplicates, missing matches, and multiple calls per lead before computing counts.
Checkpoint: A join-quality report.
Extract evidence before themes
- Find explicit needs, constraints, objections, and buying triggers in each conversation.
- Attach passages and mark ambiguous statements rather than assigning confident labels.
Checkpoint: A call-level evidence table.
Apply consistent coding
- Use the documented rubric to classify evidence and review uncertain examples.
- Retain contradictory observations and avoid turning an analyst's expectation into a label.
Checkpoint: A reviewed coded dataset.
Aggregate with clear denominators
- Calculate theme counts and outcome comparisons at the chosen lead or call grain.
- Separate association from causality and show missing-data coverage.
Checkpoint: An audience-analysis table.
Write the reusable brief
- Explain recurring patterns, counterexamples, and decisions the findings could inform.
- Save corrections to the method as a new version for subsequent runs.
Checkpoint: A comparable research brief and method change log.
Decisions along the way
One lead has several calls
Aggregate according to the declared method and avoid duplicate lead counts.
Outcome data is immature
Exclude it from outcome comparisons while retaining transcript evidence.
A new theme appears
Use an open-code field and review a taxonomy update.
What you will produce
audience-evidence.csv
CSVSupports traceable analysis.
Include: lead_id · call_id · theme · passage · certainty · outcome
audience-research-brief.md
MarkdownEnables repeatable team discussion.
Include: method_version · sample · patterns · counterexamples · decisions · limits
Check the result
- Reconcile call and lead counts before analysis.
- Inspect a sample of codes against full conversation context.
- Recalculate denominators and rates independently.
- Confirm period comparisons use the same definitions.
When something goes wrong
Identifiers do not join reliably
Pause outcome analysis and report transcript-only findings.
The prompt changes labels mid-run
Reapply one version to the full sample.
A theme is inferred without explicit evidence
Remove the unsupported code or mark it as an analyst hypothesis.
A worked example
SYNTHETIC EXAMPLE · NOT A REPORTED BUSINESS RESULT
Request
Compare audience objections across two sales cohorts.
Inputs
Synthetic example. Synthetic lead L1 has two calls expressing an integration concern; L2 has one price objection. Both calls for L1 belong to one lead.
Expected result
Synthetic expected result. At lead grain, report one integration-concern lead and one price-objection lead, not two versus one. Any outcome association remains descriptive.
Run a small pilot
Start small
A small batch with repeated leads and some missing outcomes.
Record the baseline
Record analyst time and coding disagreements under the current method.
Agree success criteria
Agree join accuracy and coding consistency before routine reuse.
Stop or revise
Stop comparative conclusions if definitions or identifiers cannot be reconciled.
Keep the scope clear
- Use only authorized call records and necessary customer context.
- Do not turn unreviewed transcript labels into automatic employment or customer decisions.
Check before using
Validate input selection and metric definitions. This workflow assists interpretation; it does not establish causation or replace research judgment.
Business activities
- Run sales conversations and proposals
- Research customers and markets
FROM A RECIPE TO YOUR WORKFLOW
Bring the company context together.
This workflow needs reliable inputs before it needs a better prompt. Moat8 Vault brings working information into a readable archive for the AI tools you use.
For this recipe: Sales-call transcripts · Lead and performance data · Research instructions and definitions
Start with one recurring task, agree which sources to use, and review the first result with your team.
Discuss this workflow