In this guide 6 sections
A concierge MVP shows customers a simple app while your own staff perform the real work by hand behind it. You learn whether people want the service before you pay to automate it. Automate only the steps that staff repeat daily, which you find by watching the manual work.
What a concierge MVP looks like in practice
The customer sees a booking form, a status page and a confirmation message. Behind the form there is no matching engine. A coordinator reads each request, phones two suppliers, picks one and types the result into a back office screen.
Nobody outside learns this. That is the point. They get an answer in the time you promised, and you get a record of every decision a person had to make. That record is the specification for the software you may build later.
This suits services where the hard part is the judgement, such as matching a technician to a job, approving a quote or assigning a hostel bed. It suits poorly where speed is the product, such as live pricing.
Why do the work by hand first?
Automation locks in your assumptions. Code a matching rule before you have matched a hundred real requests, you will code the wrong rule and then defend it.
Doing the work yourself exposes the cases nobody listed. The customer who wants a Sunday slot. The supplier who replies by voice note. The approval that needs a photo of a signed form. Each one would have been a bug report later.
Manual running also tests demand cheaply. If twenty customers do not come back for a second booking, no algorithm would have rescued the idea.
- Real requests replace guessed ones
- Edge cases surface before they are expensive
- Demand is tested with a few weeks of staff time
- Staff learn the process they will later supervise
Which steps to automate first
Keep a tally sheet beside the coordinator. Every time a person repeats a step, add a mark and a stopwatch time. After two or three weeks, three kinds of step stand out: frequent, slow and error-prone.
Automate the frequent and error-prone ones first, since those save the most staff time and cause the most customer complaints. Typical examples are sending the confirmation message, checking a document for missing fields and copying a booking into a calendar.
Leave judgement steps manual for longer. Deciding which of two suppliers to trust is a poor first target. Software can suggest a shortlist while a person still picks.
- Confirmation and reminder messages
- Duplicate and missing-field checks
- Calendar and invoice entries
- Status updates to the customer
- Supplier assignment, last of all
What the operator screen needs
The app is simple, so the back office screen carries the weight. Staff need one queue showing every open request with its age, owner and next action. Without it, requests hide in chat threads and the promised time is missed.
Give each request a notes field, a status the staff can change in one click and a log of who did what. When a customer asks about a booking from last Tuesday, anyone on the team should answer from that screen alone.
Add a way to reassign work when someone is off sick. Add an export so the tally sheet and the log can be analysed in a spreadsheet. Skip the dashboard for now.
- Open queue sorted by age
- One-click status changes
- Notes and attachments per request
- Activity log with names and times
- Reassign and export buttons
Know when to stop being the engine
Set a trigger before you start. For example, when a coordinator can no longer clear the queue by end of day, or when a single step costs more staff hours than the build would. Write the trigger down so it is not argued from emotion.
When it fires, move one step at a time. Automate the highest-ranked step from the tally sheet, run it beside the manual route for a week and compare results. Keep the manual route as a fallback.
Your next move is simple: open a spreadsheet today, list the steps behind your idea and mark which ones a person would perform on day one.
Concierge step tally sheet
Each manual step gets its own line on one sheet. Columns hold the step name, the person who performs it, weekly count, average minutes and mistakes seen. A last column marks it automate now, automate later or keep manual. A small block at the bottom records the trigger that ends the manual phase.
Open a blank worksheet to printQuestions owners ask
Is a concierge MVP dishonest to customers?
No, provided you do not make false claims. Say what the service delivers and by when. How the result is produced behind the screen is your business, as long as data handling is described truthfully.
How long should the manual phase last?
Long enough to see repeat behaviour and the main edge cases, usually a few weeks of real requests. End it on your written trigger, not on a calendar date.
What is the difference from a Wizard of Oz test?
They are close cousins. A concierge test runs as a visible service. A Wizard of Oz test hides the human so the product seems automatic.
Can a concierge MVP handle payments?
Yes. Use a hosted payment page from your provider and reconcile by hand against the booking queue. Keep card details out of your own screens.