Checklist
Software development partner evaluation checklist: 43 checks before you sign
How do we choose a development partner without getting burned?
- Checks
- 43
- Sections
- 7
- Last updated
Your progress
0 of 43 checked
Saved in this browser
A software development partner evaluation checklist is a set of yes or no checks you apply to every studio or agency on your shortlist before you sign. A partner worth hiring diagnoses before it quotes, labels its proof, names who does the work, shows its engineering standards, keeps code in your accounts, and puts terms and AI use in writing.
This checklist is for a founder, CTO, or head of product at a seed or Series A startup with a funded plan, no dependable team to build it, and an outside partner to choose. The same checks apply to studios, agencies, and freelance teams, so every name on the shortlist is judged on the same terms.
How do you vet a software agency with this checklist?
#Vet a software agency by scoring it on evidence at three points: after the first call, after the proposal walkthrough, and after a reference call. The first call shows how a partner diagnoses, the proposal shows its scope and terms, and the reference shows how it handles problems.
- Mark a check yes only for something you saw: a document, a repository, a contract clause, or a person you met.
- Count “we usually do that” and “we can add that later” as no.
- Score each partner in its own column, and note the evidence behind every yes.
Our guide on choosing a software development company explains the reasoning behind each group of checks. A studio that sells builds wrote this checklist, so apply every check to us as well.
Section 010 of 7 checked
Did they diagnose before they quoted?
- A quote produced without those answers prices a product the partner imagined, not yours. The gap returns later as change requests for things nobody asked about, such as a CSV import or a CRM sync.
- A partner that knows the outcome, such as “three pilot customers onboarded by March”, can cut scope to protect it. One that only knows the feature list builds all of it and misses the date.
- Inherited code hides problems no call can surface, such as a schema with no tenant boundary or migrations that were never committed. A price given without reading the code rests on assumptions you pay for later.
- A range early lets you check budget fit without anyone guessing at detail. Written assumptions, such as “Stripe Checkout, not a custom billing flow”, show exactly what would move the price.
- A partner that agrees with everything is waiting for tickets, and you lose the senior judgment you are paying for. A useful challenge sounds like “launch with email sign-in and add single sign-on when the first enterprise customer asks for it”.
- Scope arguments usually start where each side assumed the other had something covered. An exclusion such as “no data migration from the old system” settles the question before it costs money.
- A short fixed-price step shows how the partner works on your problem for a fraction of the build budget. A partner that only sells the full build asks you to take all the risk at once.
Section 020 of 5 checked
Is their proof labeled and checkable?
- AI design tools make polished screens cheap, so an unlabeled mockup can pass for a shipped product. The label tells you what you are judging: engineering, design, or neither.
- Screenshots hide everything behind the interface. A live product lets you test sign-up and error states, and a public repository shows the commit history, pull requests, and CI runs.
- Problems are normal on any project, and how the partner handled them is something no portfolio can show. A reference past launch can also tell you whether handover worked and whether their own engineers could take over.
- A case study built by a team that has since left proves little about the team you would get. Ask who led it and whether that person is free for your timeline.
- A line such as “cut load times in half” can mean one page on one fast laptop. A partner with real results can explain the baseline, the method, and the date.
Section 030 of 6 checked
Who does the work, and how will you hear from them?
- The person on the sales call is often not the person in the repository. The lead's judgment on the data model, code review, and trade-offs is what you are paying for, so judge it directly.
- Undisclosed subcontracting puts your code and data with people you never vetted. If their contract with the partner has no IP assignment, the chain of ownership to you can break.
- Teams of every size lose people. A good answer names a second engineer who already reviews the code, plus architecture decision records and runbooks that keep the reasoning in your repository.
- A blocking question asked at 5pm and answered the next afternoon costs a day each time it happens. Agree an overlap window and a written reply time, such as one business day, before kickoff.
- Status reports and percentages hide problems a working build exposes in minutes. A preview link per pull request lets you check progress without waiting for a meeting.
- Decisions made on calls and never written down get argued again weeks later. A written update also shows early when the timeline depends on feedback you owe.
Section 040 of 6 checked
Can they show their engineering standards in practice?
- Anyone can say “we review all code”. A real pull request shows whether review catches design problems or only approves, and whether a failing check actually blocks the merge.
- A coverage number can climb on easy code while the paths that lose money go untested. One test proving that account A cannot read account B's data guards against a breach-level bug.
- Every team ships a broken release eventually. Feature flags or a one-step redeploy of the previous version turn it into a short incident, as our guide to safe releases and rollbacks explains.
- Without them, a broken sign-up flow surfaces as a customer email days later. Error tracking with a tool such as Sentry can go in within a day, while structured logs and OpenTelemetry traces are far harder to retrofit across code written without them.
- Keys committed to Git stay in its history after the file is deleted, and vulnerable packages ship without anyone noticing. NIST's Secure Software Development Framework gives buyers a shared vocabulary for asking suppliers about these practices.
- Six months later, nobody remembers why the team chose Postgres over Firestore or one service over three. A short record per major choice lets your next engineer change course without repeating the debate.
Section 050 of 6 checked
Will you own the code, the accounts, and the knowledge?
- A repository in the vendor's account turns handover into a negotiation and puts the history at risk if the relationship ends badly. Owning it from the first commit costs nothing.
- A domain or Apple developer account registered to a contractor can take weeks to recover. Paying providers directly also shows you the real running costs from the first month.
- A single owner is a single point of failure, whether that is your cofounder or the partner. With two owners and MFA, one lost phone or one departure cannot lock the company out.
- Offboarding without a list leaves old deploy keys and API tokens active for months. A list of who holds what, plus secrets in a manager you control, turns the exit into a checklist rather than a search.
- Code alone is not a handover. An engineer who has never met the partner should be able to run, change, and deploy the product from what sits in the repository.
- A handover document nobody has tested is a guess. One real deploy while the partner can still answer questions finds the missing steps while they are cheap to fix.
Section 060 of 7 checked
Do the contract and payment terms protect you?
- Paying an invoice does not by itself transfer rights in code. In the US, the UK, and Canada a contractor generally keeps the copyright unless a written assignment says otherwise, and German law grants exclusive usage rights instead of an assignment, so have a lawyer read the clause.
- Partners often build on code they reuse across clients, and that code stays theirs. Without a license that survives a sale of your company, you may own the product but not the parts it cannot run without.
- Payments on calendar dates keep flowing whether or not anything works. A split such as 40% at start, 30% at a working midpoint, and 30% at launch keeps money and progress in step.
- Unpriced changes either get folded quietly into the estimate or arrive as a surprise invoice. A quote per change keeps scope and budget visible to both sides.
- A termination clause written only for the vendor can leave you paying for work you cannot access. Check the notice period and exactly what transfers on exit.
- Usage-priced services grow with every active user, and a markup hidden in a hosting bill repeats every month. Listing them in the proposal shows the real cost of running the product.
- Under UK GDPR and the EU GDPR, a partner that handles personal data on your behalf acts as a processor, and Article 28 requires a contract with specific terms. Enterprise customers' security questionnaires often ask for it as well.
Section 070 of 6 checked
Is their AI use written down and reviewed?
- AI coding tools are a normal part of how good teams ship quickly. Without a written policy, you cannot tell whether the speed comes with review or with debt you inherit later.
- AI-drafted code can read cleanly and still skip an authorization check on one endpoint. Our AI code review workflow shows how a named engineer who can explain every line stops that from merging.
- Whether prompts are retained or used for training depends on the AI vendor's plan and settings. Ask which vendors they use under which terms, and require written approval before real customer data goes into a prompt.
- The US Copyright Office concluded in January 2025 that AI output is protected only where a human author determined enough of its expressive elements, and that prompts alone are not enough. Some AI-drafted lines may carry no copyright at all, so the contract has to say that you receive every deliverable and the partner keeps no claim to any of it.
- Building a first version with AI tools is a sensible speed trade that can prove demand quickly. A good partner checks it for exposed secrets, permissions, and missing tests before adding features, the gaps our AI-built app risk checklist covers.
- A model that is sometimes wrong can still be useful if everyone agreed in advance what happens then, such as a human review queue. A pilot that measures accuracy and cost on your own data turns those limits into numbers.
0 of 43 checked
How do you score a software development partner evaluation checklist?
#Score each partner group by group, not by total. A partner with 38 yes answers can still be the wrong choice if its five gaps sit in ownership and contracts, because those are the hardest to fix once work has started.
Which failures are red flags?
#A red flag is a failed check that gets expensive to fix once the contract is signed. Watch for these:
- A fixed price for a full build before anyone asked about your users, data, or code.
- The repository, hosting, or domain in the vendor's accounts, or reluctance to move them.
- No named lead you have met, or subcontractors you hear about late.
- An IP clause that is missing or vague, or one that keeps rights with the vendor.
- No written exclusions or acceptance criteria in the proposal.
- Unlabeled proof, or concepts shown as shipped client products.
- No answer on who reviews AI-drafted code or where your data goes.
One red flag is a reason to ask more questions and get the answers in writing. Two or more usually mean you should keep looking.
When does an outside review make sense?
#Get an outside review when the partners you are comparing would build on code you already have, or when your plan is still too loose for two partners to quote the same thing. A software audit from $2,500 reviews the code or the plan and ends in a written report every partner can quote from, and you keep the report whichever partner you choose. When you are ready to build, our product development service is priced by milestone and starts with a 45-minute discovery call.
Checklist updates
Get an email when this checklist changes
We revise checklists when tools or advice change. Leave your email and we'll tell you when this one is updated. You'll also get the monthly newsletter, and every email has an unsubscribe link.
FAQ
Questions about this checklist
Straight answers on what it covers, who should run it, and what to do with the results.
The criteria that matter most are the ones that are expensive to fix later: who owns the repository and accounts, what rights the IP clause gives you, who leads the work, and whether every change is reviewed and tested. Price and portfolio polish matter too, but a cheaper quote with your code in the vendor's account usually costs more by the first handover.
Yes. Most checks apply unchanged, especially ownership, IP assignment, and code review. Adjust the team group: a solo freelancer has no second engineer, so ask who reviews their code and what happens if they become unavailable. Decision records, runbooks, and accounts in your name matter even more with one person, because nobody else holds the context.
Yes, within limits. Client code is usually confidential, so expect a walkthrough on a screen share, an open source repository, or the partner's own product rather than files sent to you. What you want to see is a pull request with real review comments and the CI checks that ran. A partner with no code it can show at all should be able to explain why.
For each partner, plan on a discovery call, a proposal walkthrough, a reference call, and an hour to score the results. That usually fits into one to two weeks, and two or three partners are enough to compare. If the decision needs more certainty than calls can give, a small paid audit or pilot tells you more than another round of meetings.
Score us the same way. We have no client case studies yet, so our proof is labeled Open Source, such as Commitscape, whose code, pull requests, CI runs, and decision records are public. Every project has a named lead, the repository and accounts are yours from day one, code and IP transfer on final payment, and our engineering standards are written down. A pilot sprint from $4,500 is a fixed-price way to see how we work before a build.
Related
The reasoning behind the checks
Guides that explain why these checks matter and how to fix what fails.
Next step
Need help closing the gaps?
Tell us what you're building and what the checklist turned up, or book a call if you're not sure where to start.
On the discovery call
- 01What prompted this now, and where things stand
- 02What a good outcome looks like
- 03Whether we're a fit, and what a next step could be
45 minutes. No slides, and no long questionnaire.