Software development companies are difficult to compare because they all present the same way. Similar case studies, similar process diagrams, similar language about partnership and quality. The differences that matter are not visible on a website and rarely surface in a first call unless you go looking for them.
What follows are the questions that produce different answers from different firms, and what the answers tell you.
Ask them to describe your data model back to you
After a couple of conversations, ask the firm to sketch the main entities in your business and how they relate. Not screens, not features, the underlying things: what a customer is, what an order is, what happens to a record when someone cancels.
A firm that can do this roughly correctly has understood your problem. A firm that cannot has been listening for keywords, and will discover the real complexity later while spending your budget.
This one question does more sorting than any amount of reference checking, and it costs nothing.
Ask what they will not build
A proposal that only lists inclusions has not been thought about. Every project has a boundary, and if the firm has not drawn it, you will discover its location during a disagreement about whether something was in scope.
The answer you want is specific and slightly uncomfortable. Something like a named feature they think you do not need in the first version and their reason. A firm willing to argue you out of scope in a sales conversation is a firm that will hold a line later, which is what you are paying for.
Our own statements of work carry an explicit exclusion list alongside assumptions and risks. Scope gets fixed before price, in that order, because a price attached to an undefined scope is not really a price.
Ask what happens when the estimate is wrong
Estimates are wrong. The useful question is what the contract does about it.
An answer that no estimate ever changes is either untrue or means the padding is large. An answer that everything is billed by the hour puts all the risk on you. A workable answer names a mechanism: a change process, a phased structure where each phase is separately priced, or an agreement that the scope flexes while the budget holds.
Ask specifically about the last project where the estimate was wrong and what happened commercially. The willingness to answer that matters as much as the answer.
Ask who is actually writing the code
The gap between the people in the sales process and the people on the project is a real and common problem, and it is worse the larger the firm.
Ask for names, ask whether those people are currently on other projects, and ask what happens when someone leaves mid-project. Then ask how many people will be on your project, and treat a large number as a warning rather than a reassurance. Small projects succeed far more often than large ones, and team size is a decent proxy for project size.
Check ownership before anything else
You should own the code, the repository, the deployments, the infrastructure accounts, and for AI work the model accounts and prompts. This should be true from the first commit rather than on final payment.
Firms that hold any of these are not necessarily acting in bad faith, since some genuinely believe hosting is a service they provide. The effect is the same either way. If leaving is expensive, the relationship has a lock in it, and the time to find out is before you sign.
We hand over all of it. It is worth asking any firm to state the same in writing.
What rates actually tell you
Accelerance's 2026 rate data puts senior developers at $64 to $76 per hour in Central and Eastern Europe, $60 to $75 in Latin America and $31 to $41 in Asia, with rates falling in every region during 2025. Clutch's directory shows most app development firms in the $25 to $49 per hour bracket, with Indian firms below $25 and Australian firms between $100 and $149.
Those ranges are wide, and the position within them tells you less than people assume. A team at half the rate that needs three times the hours costs more and takes longer. Compare total scope against total price, and treat a much lower total for the same scope as a signal that something has been left out rather than as a discount.
Signals that should end the conversation
No questions about your data. A firm that can quote after hearing your feature list has quoted a guess.
A quote broken down by screen. Cost lives in roles, integrations and data migration. A price list of screens means nobody has thought about the parts that carry risk.
Reluctance to start small. A firm that will only take the whole programme is managing its own revenue predictability, not your risk.
A team assembled after you sign. Recruiting for your project after the contract means the timeline starts with hiring, and you will not know who you are getting.
Vague answers about ownership. This one is worth walking away over, regardless of how good everything else looks.
A workable sequence
Start with a small, separately priced piece of work rather than the full build. A paid discovery or a contained first deliverable tells you more about how a firm works than any reference call, and if it goes badly you have lost a small amount rather than a large one.
Judge that first piece on whether the artifacts arrived. Was there a real scope with exclusions, did software reach a real environment early, did the estimate hold or was the deviation explained before it became a problem. Those are observable, unlike the qualities that get discussed in sales conversations.
Then commit to the larger work with the firm you have now actually seen work.
Frequently asked questions
Agency, freelancer or in-house?
Freelancers work well for contained, well-specified pieces and carry concentration risk on anything long. Agencies cost more per hour and bring continuity and a broader skill mix. In-house makes sense once the software is central enough to need continuous change, which is usually later than people expect.
How do I compare quotes that differ by a large factor?
Compare the assumptions, not the totals. Large gaps almost always mean the firms have quoted different projects. Ask each to list what they have assumed and what they have excluded, then compare those documents rather than the numbers.
Should I pay for discovery?
Yes, and be sceptical of firms that give it away, since unpaid discovery is a sales activity and produces a document shaped like a sales document. Our audit starts at $3,000 and is credited toward the build, which keeps the incentive on producing a real scope.
How do I check references usefully?
Ask past clients what went wrong and how it was handled. Every project has something. A reference who cannot name anything either did not have a real project or is not being candid.
How much should the first project be?
Small enough that being wrong about the firm is survivable. Our focused builds run $5,000 to $15,000, which is a reasonable size for a first piece of work with a new partner.
Where to start
If you are evaluating firms and want a scope you can hand to any of them, our audit produces exactly that. It starts at $3,000 and is credited toward the build if you work with us.