A concierge experiment uses people to deliver a proposed outcome while the team studies the workflow. It is useful when the uncertainty concerns value or process rather than technical feasibility. The customer should understand what service is actually being supplied.
Identify the question the experiment can answer
A manual report can test whether a decision-maker uses an insight. It cannot establish that an automated system can produce that insight accurately or cheaply at scale.
GOV.UK’s discovery guidance supports examining the problem before committing to a solution. The founder’s experiment should similarly isolate a defined uncertainty instead of pretending to be a complete product trial.
Agree the output, delivery frequency and access required. Use authorised data and avoid presenting human analysis as an autonomous AI capability.
Record the effort behind the result
An illustrative B2B intelligence startup might prepare weekly account briefs by hand. If a customer finds them useful, record which sections affect action and which are ignored. Also record research, checking and revision time.
The result may reveal that the value lies in a short verified update rather than a large database. It may also reveal that customers expect judgment which is difficult to automate responsibly.
Keep founder relationships separate from product evidence. A friend accepting a report to support the venture is different from a buyer paying for recurring work.
Decide what should become software
Automate stable steps only after understanding their inputs and exceptions. Retain human judgment where it is material to the service, and include that cost in the business model.
The next test could reduce personal involvement while maintaining the promised output. If customer value disappears as soon as the founder stops providing bespoke advice, the team may be building a professional service rather than the intended SaaS product. That can be viable, but it is a different operating model.
Describe the manual service to customers in terms they can verify: what arrives, who reviews it and what the company does when information is missing. Keep experimental automation backstage only when that accurately reflects the offer. A service test should never depend on customers misunderstanding whether a person performed the work.
GAT’s analysis of scaling bottlenecks provides a useful companion. A manual experiment succeeds when it reveals both the value customers want and the work required to supply it repeatedly.
Published by the Global Apex Tech Editorial Desk. Partner involvement, when applicable, is disclosed above the headline. For editorial questions or source material, contact editor@globalapextech.org.