Short answer: shortlist three agencies, ask each the same five questions, read the exclusions in their proposals rather than the inclusions, and buy a short paid trial slice from your favourite before committing to the build. The whole process takes about three weeks and removes almost all of the risk.
Step 1 — write down what "done" means, roughly
Before talking to anyone, write two pages: who the user is, the one journey that matters, what has to be true for the app to be worth having, and the hard constraints (offline, hardware, regulation, launch date).
Two pages, not twenty. Its purpose is to let three agencies respond to the same thing so you can compare their answers rather than their imaginations.
Step 2 — shortlist three
Three is the number. Below it you have no comparison; above it the evaluation costs more than the spread between candidates.
Filter on evidence, not positioning:
- Apps live in the stores today that you can install.
- At least one with a constraint resembling yours — offline, background location, payments, hardware.
- A published rate, or at least a rate given without three meetings.
- A size that matches your project. A five-hundred-person firm will staff your four-person project with whoever is free.
Step 3 — the five questions
Ask all three candidates the same five, and write down the answers.
1. "Which named engineers will be on this in month nine?"
The most revealing question in the process. Agencies that rotate staff cannot answer it in writing. Ask for names in the contract with notice before any change.
2. "Show me an app you built with a constraint like ours, and tell me what went wrong on it."
The second half matters more. Every real project has a disaster; a team that describes one clearly has actually shipped and has learned something. A team with no failures has no memory.
3. "I change my mind in week five. Walk me through exactly what happens."
Good answer: re-estimate against acceptance criteria, written decision, revised date, published day rate. Bad answer: "we're flexible".
4. "Who owns the repository, the store accounts and the signing keys?"
One correct answer: you, from the first commit. Losing a signing key to a departed supplier is close to unrecoverable.
5. "What would you tell us not to build?"
Tests whether they will ever say no. The most expensive agencies are the ones that agree with everything, because you pay for every bad idea you had in month one.
Step 4 — read the proposal backwards
Start with the exclusions. Most proposals are written to look complete; the exclusions tell you what will become a change request.
Then check four things:
| Look for | Good sign | Warning sign |
|---|---|---|
| Definition of done | Acceptance criteria per screen and state | Prose paragraphs |
| Change pricing | A day rate in euros | "Discussed case by case" |
| Ownership | Yours from commit one | Transferred at project end |
| Operational scope | Crash reporting, analytics, CI, forced upgrade | Only the client binary |
If crash reporting, analytics, CI and forced-upgrade handling are absent, the price is not lower — it is incomplete, and you will buy them later at a worse moment.
The cheapest proposal is usually the one that excluded the most, and the exclusions are where projects go to die.
Step 5 — buy a trial slice
Then stop evaluating and start delivering. Two to four weeks, fixed price, one real increment of the hard part — the background sync, the payment path, the offline behaviour — in production, with tests, CI and documentation.
You learn things no reference call reveals: how they write, how they handle a surprise, whether the named engineers actually appeared, whether questions get answered the same day.
If it goes badly you have spent a month. If it goes well you have a working piece of your product and a supplier you have genuinely tested.
Ours is €9,500 fixed, with a dedicated senior team at €7,000 a month afterwards and a published €450 senior day rate. See what we have built or describe the app.
Frequently asked
How many agencies should I shortlist?
Three. Below that you have no comparison; above it the process costs more than the difference between candidates. Spend the saved time on a paid trial slice with your favourite instead.
What is the single most revealing question?
Which named engineers will be on this in month nine. Firms that rotate people to keep a bench utilised will not answer it in writing, and that answer predicts more about your project than any portfolio.
How do I read a proposal properly?
Check what is explicitly excluded, whether 'done' is acceptance criteria or prose, what a change costs in euros, and who owns the repository, store accounts and signing keys. The exclusions tell you more than the inclusions.
Should I pay for a trial?
Yes. A two-to-four-week paid slice delivering one real increment to production tells you more than every reference call combined, and the downside is a month rather than a year.