Emad Ghorbaninia runs 40 to 50 agents inside his company. He still feels uneasy reaching for five different analytics tools to infer what a customer wants, when a five-minute conversation would tell him directly.
The short answer: The data he trusts most is not tracked but asked for, and anything that reaches a client passes a human review first.
What Emad Ghorbaninia trusts, and for what
| Evidence | Trusted for | Not trusted for |
|---|---|---|
| A direct conversation with the customer | What the customer wants | He named no limit, and trusts it more than anything tracked |
| Analytics tools | Inferring what customers want, which he accepts only reluctantly | Replacing the question |
| Human design, review and publishing | Anything that leaves the company | Nothing is exempt |
| Agents that build skills for other agents | Cutting his own instruction load | Client work without review |
A confessed conflict about tracking
Emad does not take analytics as a given good. He has to argue himself into reaching for it, since a short conversation with the customer answers the question sooner and more honestly. A production incident years ago shaped his view too: in his account, a third-party SDK crashed because it was tracking far more than anyone knew, and login stopped working across a large share of app store apps. He now sees how openly you track and whether you should track at all as one question.
A hard rule for anything client-facing
He spent years teaching before he built, and went all in on AI in January. His agency now replicates what a senior-staffed traditional agency does, with four co-founders who each bring two decades of experience. The company rule is strict: text, video, image or audio that reaches a client is designed, reviewed and published by a person, whatever the content type. The four co-founders each own a domain and all of them comment on everything.
Speed moved the rigour upstream
He has stopped instructing agents directly and started asking agents to build the skills that instruct other agents. One client engagement took three months to scope and produced a 200-page requirements document, which no model was given whole.
Because output is no longer the constraint, he removed the middle layer from his own process and put himself in front of the customer. The freed time goes on architecture, code quality and security.
What to do with this
Before adding a tool to infer what customers want, check whether you could simply ask them, and be open with users about whatever you do collect. Our reading, not his: tracking still earns its place for questions no conversation can answer, such as what happened across every account last week.
For seven checks drawn from across the series, see the questions to ask before trusting a number.
Frameworks this interview tests
- Jobs to be done (JTBD): Clayton Christensen's jobs to be done says customers hire a product to make progress. The job decides which events matter, so interviews come before the tracking plan.
Read next in The Question
Source and disclosure
Read Issue 11 on A Founder's Friend. The full interview is for subscribers. This page paraphrases it and quotes nothing that sits behind the paywall.
Skene sponsored this research and commissioned it on a single question. A Founder's Friend ran the interviews, wrote them up and published them, and the series states that Skene paid for the research and did not direct the questions, the analysis or the conclusions. Each of the thirteen interviews was reviewed and signed off by the person in it, named or anonymous by their choice. Appearing in the series is not an endorsement of Skene, and nobody here is presented as a Skene customer.