Some questions are not answerable by watching behaviour. Why people hesitated, what they thought a feature did, whether the price felt fair — analytics can show you the drop, never the reason. Research studies are how Sightspool asks, under the same governance as everything else.
It starts with an escalation
A committed assumption can be escalated for qualitative research. This never touches its verdict — the research state runs parallel to it, because a study produces evidence, not a measurement. Escalation is always proposed by a person or the agent and confirmed by a person; no background job can decide to start researching your customers.
The brief has two authors
The agent frames the study: the research question, the decision it must change, the evidence already available, and the cohort worth recruiting. It deliberately leaves the design blank. A human researcher then designs it — method, sample size, participants, discussion guide and bias risks. Approval re-parses the brief and refuses an incomplete design.
Two delivery routes
- In-product — a micro-interview served through the Sightspool SDK to the relevant cohort, at the moment they are in the product.
- Your own PostHog — a survey created, launched and stopped through PostHog's surveys API using your key. Responses are read back as citable evidence.
The PostHog route exists because survey delivery is something PostHog already does well, and rebuilding it would be worse for you. Responses count against your own PostHog allowance, so the response target caps the survey's response limit — Sightspool never commits your spend, and launching is always your call.
Choosing who is asked
You can target the audience: people who signed up more than a month ago, people who never finished onboarding, people who downgraded, people who returned after going dormant, or the specific funnel drop the assumption is about. The audience compiles to a cohort in your own PostHog, which evaluates it — Sightspool describes the audience in plain language before you launch so you can see exactly who will be asked.
Closing the study
When a study stops, you trigger the synthesis. A routed lens reads the responses and posts one proof-gated finding citing that exact read — the same proof gate every other finding passes. The brief then settles, the assumption's research state settles, and the study's judgement lands as a graded read to be scored later against whatever measurement eventually says.