The client
A professional services firm. The founder is a well-networked consultant with a portfolio career — advisory work, peer conversations, mentorship, and client engagements across multiple domains.
They sent us 176 recorded calls. We ran them through our standard pipeline.
What went wrong
The report declared: "You're running a recruiting agency disguised as a consultancy."
It scored the business 4/10 on product-market fit and recommended fundamental restructuring. The tone was adversarial — designed to challenge the founder with uncomfortable truths.
The problem: there were no uncomfortable truths to deliver.
This wasn't a failing business. It was a varied professional life. The calls included client projects, peer networking, mentorship sessions, industry conversations, and personal check-ins. The system treated all of them as sales data and diagnosed the diversity as dysfunction.
A well-networked professional with a portfolio career doesn't have a "product-market fit problem." They have a rich professional life that doesn't fit into a single-business-model framework.
Why it happened
Our pipeline had one persona: the "ruthless PE operating partner." Every report was written from the perspective of someone diagnosing a company for acquisition — looking for inefficiencies, misalignment, and structural problems.
That persona works when the data is a sales team's call archive. It doesn't work when the data is a consultant's varied professional conversations.
Three specific failures:
1. Wrong audience assumption. The system assumed every call archive belongs to a company with a sales team. It doesn't.
2. Forced adversarial tone. The report was engineered to find problems. When the data didn't contain obvious dysfunction, the system manufactured it by reframing professional variety as strategic confusion.
3. Single-lens analysis. One report, one persona, one audience. No option for a gentler, more appropriate analysis.
What we built because of it
This failure led to the most significant product evolution in Moat's history: the multi-lens report suite.
Instead of one report from one persona, we now generate nine reports from the same evidence base — each written for a different audience with a different tone:
- CEO Brief: direct but calibrated — brutal only when data warrants it
- Sales Intelligence: coaching-oriented, improvement-focused
- Conversation Quality Audit: constructive, never shaming
- Voice of Customer: research synthesis, letting customers speak for themselves
- Product Signals: neutral, evidence-based, analytical
- Customer Health: account-level, early-warning oriented
- Market Intelligence: opportunity-focused, analytical
- Promises & Risks: factual, risk-oriented, no judgment
- Diagnostic Scorecard: quantified health assessment
The system also now detects whether the data represents a company or a person, and adjusts which reports to generate and what tone to use.
The lesson
The most important thing an intelligence system can do is know when its framework doesn't fit the data.
A system that always finds problems will always find problems — even where none exist. That's not intelligence. That's confirmation bias with a confidence score.
We'd rather generate a report that says "your data doesn't show dysfunction — here's what it does show" than force a narrative that doesn't match reality. The 176-call engagement taught us that.
The worst kind of intelligence is the kind that's confidently wrong. Our most important product improvement came from a report that should never have been written the way it was.