Applying Jobs-to-be-Done Without Pretending
Most JTBD analyses are theatre. Here's how the Customer Job Agent grounds every job statement in evidence — and what to upload to make it work.
The problem with most JTBD analyses
Jobs-to-be-Done is a powerful framework — when applied rigorously. The problem is that most analyses cut corners: they restate features as "jobs," confuse demographics with motivations, and produce job statements that no real customer would recognise.
The Customer Job Agent in QuantShift AI is built to avoid those failure modes. Three rules drive it.
1. Every job is grounded in evidence
The agent will not surface a job that can't be supported by a direct quote or specific data point from the documents you uploaded. If you upload nothing, it falls back to public proxy data and clearly labels the analysis as lower-confidence.
2. Every job follows the structured format
"When [situation], I want to [motivation], so I can [outcome]." Vague jobs that don't follow this structure get rejected. This isn't pedantry — the structure forces you to identify the trigger, the motivation, and the desired outcome separately, which is what makes JTBD useful in the first place.
3. Trends matter as much as state
A job that's becoming more prominent over time is more strategically valuable than one that's stable. The agent estimates trend direction (growing / stable / declining) for every job — even when data is thin, it makes a defensible call and says so.
What to upload
The agent works best with first-party data: customer interview transcripts, support ticket exports, NPS surveys with open-ended responses, win/loss notes. Upload them as PDF or CSV. If you connect Notion via MCP, the agent can also pull research docs and meeting notes from there.
We don't currently support direct platform exports (Zendesk, Intercom, etc.) — you need to compile the data yourself first. We're keeping the data ingestion surface intentionally small while we get the analysis layer right.