The honest pitch for synthetic research is not "trust us." It is that you can find out what it is worth on your own product, with your own evidence, inside a week, for the cost of a few focused hours. This post is the itinerary we recommend to every new team: five working days from an empty workspace to a grounded panel you have personally backtested, with clear expectations about what each day produces and what you should refuse to trust yet.
Day 1: read a report, then run one
Start by reading, not building. Every new workspace ships with a completed sample simulation, and we published a guided walkthrough of it. Twenty minutes with its layers, the distribution, the segment reversal, the rationales, the drivers, teaches you what a good report looks like so you can recognize a weak one later.
Then run your first simulation the same day. A plain-English prompt is enough: describe the decision, point it at a live URL or paste the two options, and let the default census-grounded panel respond. A run takes minutes, not weeks.
What to expect: a structurally complete report on an ungrounded panel. Read it for the mechanics, not the conclusions. Day 1's panel knows demographics and behavioral priors; it does not know your users yet, and its answers will have the too-general flavor that ungrounded personas always have.
Day 2: feed it your evidence
This is the highest-leverage day of the week. Upload the raw material you already have: five to ten interview or sales-call transcripts, your latest research decks, survey exports, win-loss notes. Connect a qualitative stream if you have one, support conversations or research repositories, so the evidence stays current without manual uploads. If your analytics live in a supported tool, connect that too; behavioral event trails give agents anchors for what users actually do rather than what they say.
What to expect: parsed artifacts becoming attributed evidence, each unit traceable to its source. Nothing exciting is visible yet. Grounding is infrastructure work, and its payoff arrives tomorrow.
Day 3: build the panel that mirrors your users
Now write a panel brief for the population you actually argue about in planning meetings: your target segments, in your proportions. The recruiter builds profiles against your evidence, and this is where you should be pickiest. Check the provenance labels: every attribute of every profile is marked by where it came from, your evidence, an inference, or an assumption. Attributes your evidence cannot support show up as surfaced gaps rather than silently invented defaults.
What to expect: gaps, and that is the correct experience. A panel that claims full knowledge of a segment you have never researched is lying to you. Note the gaps; they double as a map of your thinnest real-world evidence.
Day 4: run the decision you are currently arguing about
Take the live disagreement from this week's product review, the pricing framing, the onboarding step order, the two homepage angles, and run it against the grounded panel. Frame it as a forced choice with a real alternative, including "do nothing," per the scenario-design rules in the grounding guide. Use the actual artifacts as stimuli: Figma frames, the staging URL, the real copy.
What to expect: the day the instrument starts earning attention. Read segments before toplines and rationales before both. The result you are looking for is not "the panel agrees with me"; it is at least one objection or friction point that nobody in the room had raised, phrased the way a customer would phrase it. In our experience that moment, not any accuracy statistic, is what changes a team's posture from skeptical to curious.
Day 5: backtest, then decide what to trust
Before you let simulated evidence near a real decision, make it prove itself retrospectively. Pick one decision you already shipped and measured, an A/B result, a pricing change, a launch that landed or did not, and run it as if it were still open. Then compare honestly:
- Did the panel pick the right winner? Rank order is the fairest first test.
- Does the spread look like reality, or did every agent give one answer in eighteen voices?
- Do the segment differences match what your analytics actually showed?
- Where the panel was wrong, do the rationales reveal a grounding gap you can fix, or a question this method cannot carry?
What to expect: an imperfect but legible result, and a calibrated sense of trust that is specific to your product rather than borrowed from a benchmark. That local evidence is worth more than any number in the validation literature, because it is about your users.
What you should not have by Friday
Honesty about week one, so nothing here reads as overclaiming:
- Calibrated magnitudes. Fresh workspaces have no outcome history, so point estimates start uncalibrated and should be treated as hypotheses. Calibration is earned by the loop, not granted at signup: as shipped outcomes flow back in, forecasts adjust. Week one is for directional reads only.
- Coverage of populations you have no evidence for. The gaps from Day 3 do not disappear by Friday. Where they persist, the honest move is real research first, simulation second.
- A replacement for talking to users. The week's output is a working hybrid split: a fast directional instrument you have personally stress-tested, and a sharper sense of which decisions still deserve recruited humans.
Where engineering fits
If your team lives in the terminal, the whole itinerary above is scriptable: the sentia CLI and MCP server expose the same surface as the app, so a coding agent can index a repo, launch the Day 4 simulation, and pull results without leaving the editor. And the piece that makes Day 5 permanent is the SDK's exposure and outcome events, which tie every shipped decision back to the simulation that predicted it. That loop is what turns a one-week evaluation into an instrument that gets measurably better each quarter.
By Friday you will not have faith in synthetic research, and you should not. You will have something better: a grounded panel, one live decision's worth of output, one backtest's worth of receipts, and a precise, local answer to what this instrument is worth on your product. Signup is free and cardless, the sample report is already waiting, and if you want company on Day 5's backtest, talk to us. It is our favorite conversation to have.