Selection in this market is hard because the thing you are buying is invisible until it is too late to change your mind. The way through is to stop trying to assess capability from claims and instead assemble evidence you can verify yourself, which is more available than most buyers realise.
Why top 10 web development companies lists are close to worthless
Published top ten lists reflect firm size, review solicitation and marketing spend, and several are pay to appear. None of them measure the only thing that matters to you, which is how a specific team behaves on your specific problem. Use them to collect names if you have none, then discard the ordering entirely. A list is a starting point, and treating it as a recommendation is how buyers end up with the best marketed rather than the best suited.
The four pieces of evidence worth gathering
References at your size and in your situation, asked the failure questions rather than the happy ones: what went wrong, who noticed first, was it late. The names and tenure of the people who would actually be assigned. The firm's attrition rate as a number. And a written exclusions list showing what the proposal does not cover. Those four are all obtainable by asking, and together they predict outcomes far better than any credential.
Buy a small piece of real work
The single most informative step available is paying two finalists for a small, genuine piece of the work, ideally a real integration or a slice of the actual system. You learn how they ask questions, how they handle ambiguity, what their code review looks like and whether their estimates mean anything. Two or three weeks and a modest sum, against a decision that will run for years.
Then settle the terms that make a mistake survivable
Repository, pipeline, hosting and accounts in your name. Documentation in your repository. An express written copyright assignment, since under 17 USC 101 commissioned software is not automatically a work made for hire. Measurable acceptance criteria for performance and accessibility. With those in place, choosing wrong costs you a quarter rather than the product, which is the real objective of a selection process.
Questions people ask about how to choose a software development company
Are top 10 software development companies lists useful?
Only for collecting names. They reflect firm size, review solicitation and marketing spend, and several are pay to appear. None measure how a specific team behaves on your specific problem, so discard the ordering.
What evidence should we gather?
References at your size asked the failure questions, the names and tenure of the people who would be assigned, the firm's attrition as a number, and a written exclusions list. All four are obtainable simply by asking.
What single step helps most?
Paying two finalists for a small piece of real work. Two or three weeks tells you how they ask questions, handle ambiguity and whether their estimates mean anything, against a decision that will run for years.