Concierge Testing SOP

A concierge testing SOP has you deliver your solution by hand to a few real customers before you build any software, and it runs for four to eight weeks. The customer knows a person is doing the work. You do it anyway, carefully, and time every step. By the end you know whether people pay and come back, and you hold a step-by-step recipe of the service that tells you what to automate first. It is the slowest test in Solution Validation and one of the most honest.

What is concierge testing and when should you use it?

In a concierge test, the founder acts as the product. If your future app plans meals, you plan the meals. If it matches clients with freelancers, you make the matches over email. Eric Ries describes this in The Lean Startup with Food on the Table, whose CEO Manuel Rosso served the first customer in person, building her meal plans and shopping lists by hand before any code existed. Paul Graham's essay "Do Things That Don't Scale" makes the same case for founders recruiting and serving users one by one.

Use this SOP:

• When you don't yet know what the service should look like, step by step.
• When the outcome is clear but the process is fuzzy, such as matching, planning, curating or advising.
• When you want paying customers early, even at a small scale, to test the price in real life.

If you already know the process and want to test how people behave with an "automated" product, use the Wizard of Oz testing SOP instead. The concierge test comes first because it designs the service, while Wizard of Oz checks the product experience built on it.

What do you need before you start?

• A confirmed problem from 15+ customer interviews, with names of people who agreed to try something.
• A one-sentence outcome: "Every Sunday you get a matched, vetted sitter for your senior dog."
• A first manual recipe: the steps you think the service takes, even if half of them are guesses.
• A tracking board in Notion, Airtable or Trello, with one card per customer.
• A payment link so customers pay from the first week.
• Blocked time in your calendar. A concierge test with 8 customers easily fills 10 to 20 hours a week.

How do you run it, step by step?

1. Write the outcome and the manual recipe

Describe what the customer gets and how you will produce it, step by step. Keep the recipe on one page. You will rewrite it several times, and that is the point.

2. Pick 3 to 10 customers and charge them

Choose people from your interviews who scored high on urgency. Charge a real price, even a small one, from day one. Payment changes the conversation: free customers praise you, paying customers complain usefully.

3. Deliver by hand and time every step

Do the work exactly as the customer would experience it from the future product. Log each step with the minutes it took, the tools you used and anything that surprised you. A stopwatch app is enough.

4. Check in every week

Ask each customer three short questions: what was useful this week, what did you ignore, and what did you have to do yourself? The last answer shows the gaps in your service.

5. Rewrite the recipe after every round

After each delivery cycle, update the recipe. Mark each step as same every time, varies by customer or dropped. After three rounds, a pattern appears: the steps that repeat are your future software, and the steps that vary are where judgment or a human stays.

6. Rank automation candidates

Multiply each step's minutes by how many times it occurs per customer per month. The top three steps by total time are what you automate first. Steps that take little time stay manual for now.

7. Check the economics

Compare your price with your hours per customer. If you charge $25 and spend two hours, the service only works once automation cuts the time. Unit economics validation turns this into a full margin check.

8. Decide what comes next

Write a one-page memo: how many customers stayed or rebooked, what they paid, the final recipe and the top automation candidates. Then choose: keep serving by hand while you build, move to a Wizard of Oz test, or change the offer.

Template you can copy

CONCIERGE SERVICE LOG (one row per step, per customer, per cycle)
Customer | Cycle # | Step | Minutes | Tool used | Same every time / Varies / Dropped |
Customer reaction | Note

WEEKLY CHECK-IN (3 questions, by message or 10-min call)
1. What was useful this week?
2. What did you ignore or not need?
3. What did you still have to do yourself?

AUTOMATION RANKING
Step | Minutes per occurrence | Occurrences per customer per month |
Total minutes per customer per month | Same every time? | Automate first? (Y/N)

DECISION MEMO
Customers started: __ | Stayed or rebooked: __ | Revenue: $__
Hours per customer, first cycle vs last: __ vs __
Top 3 automation candidates: 1. __ 2. __ 3. __
Decision: build / Wizard of Oz test / change offer

Worked example: sitters for senior dogs

This example is illustrative. A founder plans a marketplace that matches owners of senior dogs on daily medication with trained sitters. Before building it, she matches owners by hand for six weeks and charges $25 per match on top of the sitter's fee.

• 8 owners joined from her interviews and made 23 bookings in total.
• 6 of 8 (75%) booked again without being asked.
• Hours per match fell from 2.4 in the first two weeks to 1.1 by week six.
• The medication handover checklist was identical for all 8 owners, so it became a template.
• Vetting sitters took 60% of her time, the clear first candidate for software.
• Two owners asked if they could refer a friend.

Her memo: the outcome is wanted, the price works only after vetting is partly automated, and the next step is a Wizard of Oz test of a booking app with her still vetting behind it.

What should you measure, and when do you stop?

Stop when the recipe stops changing between cycles, or at eight weeks. If hours per customer refuse to fall, the business is a service with service margins, and that is worth knowing before you raise money on a software story. Scalability testing is the next check if the numbers look promising.

Which mistakes should you avoid?

• Serving friends. They stay out of loyalty. Pick strangers from your interview list.
• Working for free. Free customers never tell you the price is wrong.
• Not timing the work. Without minutes per step, you can't rank what to automate.
• Over-serving. Extra favours you can't automate later create expectations the product can't meet.
• Staying in concierge mode forever. Set an end date and a decision before you start.

How can AI speed this up?

AI tools can take over parts of the manual work during the test: drafting messages to customers, summarizing check-in notes or writing the first version of a checklist. Keep a column in your log for "AI did this step" so you see which steps AI handles alone today. At the end, paste the full log into an AI assistant and ask it to rank steps by total time and group the variations. The human check: you talk to every customer yourself, because the reasons behind a rebooking never show up in a spreadsheet.

The F/MS startup course that builds a startup in 4 weeks gives this kind of hands-on test a deadline, and the free lean canvas template helps you write down the revenue and cost assumptions you are testing.

Where does this SOP fit in your validation plan?

Concierge testing sits in the middle of Solution Validation, after the problem is confirmed and before any build.

1. Before: customer interviews give you your first customers and the outcome they want.
2. Alongside: customer co-creation involves the same customers in shaping the service.
3. After: move to a Wizard of Oz test, and map which steps you own and which you outsource with value chain analysis.

Browse all Solution Validation SOPs.

FAQ

How many customers do I need for a concierge test?

3 to 10. Fewer than 3 shows you no patterns. More than 10 and you can't serve them well by hand.

Should I charge customers during a concierge test?

Yes. Charge a real price from the first week, even if it is lower than your planned price. Payment is what separates polite interest from demand.

How is concierge testing different from consulting?

The goal. A consultant sells hours, and more hours mean more revenue. In a concierge test you time every step so you can remove yourself from as many of them as possible.

When should I stop a concierge test?

When the service recipe stops changing between cycles, or after eight weeks. Then decide whether to build, run a Wizard of Oz test or change the offer.