A first call with a vendor is forty-five minutes and most of it gets spent on their deck. These eleven questions take about twenty minutes, and they sort a shortlist better than a month of proposals. We are asked them by the buyers we most want to work with, and several of them make us uncomfortable, which is roughly the point. Ask all eleven, in order, and write the answers down while you are still on the call.
The short version
Question eight, show us your actual vetting questions, separates most reliably, because a real process can be produced in ten minutes.
A vendor who refuses nothing has no specialism, and no specialism means no depth.
Ask which named engineers and how many other clients each one carries right now.
Working hours belong in the statement of work as times, not adjectives.
Score on the call. Memory flatters whoever presented last.
The eleven questions
Copy these into your call notes. They are ordered so that the cheap disqualifiers come first and you can stop early without wasting anyone's afternoon.
What a good answer sounds like
The point is not the answer itself but how quickly and specifically it arrives. A vendor who lives their process answers these from memory. A vendor who wrote the process for the website has to go and look.
The ninth is more useful than it looks. Every vendor with real clients has had a bad project. A vendor who cannot describe one is either new, or is managing you rather than talking to you, and both are worth knowing on the first call.
Question two is the one worth pushing hardest on, because a named engineer you can interview is the whole difference described on the hire React developers page.
Why question eight separates best
Ask a vendor to send the actual questions they put to candidates: the real ones, not a diagram of six stages with arrows between them.
A vendor with a genuine process can send them in ten minutes, because the questions already exist in a document somebody uses every week. A vendor without one sends a slide about culture fit, or promises to follow up and does not. It is the closest thing to a lie-detector available on a first call, and it costs you nothing to ask.
Where our interests differ from yours
We publish this question knowing we will be asked it, which is a reasonable thing to be suspicious of. The defence is that it is only useful if the answer is checkable: ask for the questions, then ask one of them to the engineer they propose and see whether the answer sounds rehearsed or lived.
"A vendor with a real process can send you their vetting questions in ten minutes, because the questions already exist. A vendor without one sends a slide about culture fit."
Answers that should end the call
These are not red flags to weigh against other factors. In our experience they are terminal, and continuing costs you weeks.
Score it on the call, not afterwards
Three to five vendors is the right shortlist. More than that and nobody reads the proposals carefully, including you. Score each answer out of three while you are still on the call, because memory systematically flatters whoever presented most recently.
One sheet, all vendors
Eleven rows, one column per vendor. It fits on a page and it makes the decision arguable rather than felt.
Weight two, seven and eight
Named staff, wrong-fit handling and real vetting questions predict the engagement better than the rest combined.
Note the hesitation
Which question made them pause is often more informative than the answer they eventually gave.
Three questions for the engineer, not the salesperson
The eleven above are for whoever runs the commercial conversation. If you get a second call with the person who would actually write your code, and you should insist on one, these three are worth more than any of them.
The second is the one to weight, and it is the question we started asking after a placement went wrong for exactly this reason. Someone who says they read the error properly, reproduced it, then asked in the channel after thirty minutes is describing behaviour that works in a distributed team. Someone who describes four days of solitary investigation is describing behaviour that produces an unpleasant surprise in week nine, however good the code is.
The third question sorts on curiosity. A strong engineer immediately wants to know about test coverage, deployment, who reviews, and where the last three production incidents came from. A weaker one asks which ticketing system you use.
What to do with the answers
A scored sheet is only useful if it changes what you do next. Three habits make the difference between a process and a ritual.
Send the sheet back
Tell the two leading vendors where they scored badly and give them one chance to answer properly. The reaction is often more informative than the original answer.
Ask for one thing in writing
Usually the vetting questions or the standard MSA. A vendor who promised on a call and does not deliver within two days has told you about their delivery.
Stop at two
Take two into a paid pilot, not four. Beyond two you are diluting your own reviewer attention, which is the scarce resource in any trial.
Where our interests differ from yours
The advice to shortlist tightly favours whoever is already leading, and if that is not us, it works against us. It is still the right advice: a wide pilot round produces shallow evaluation, and shallow evaluation is how vendors with good sales and mediocre engineering get selected.
Questions people ask
Where to go next
Once the shortlist is down to two, stop asking questions and run a paid pilot. Before the first invoice, the contract checklist covers what questions five, six, ten and eleven turn into on paper. Our own answers to all eleven are effectively the offshore web development page.
Yogesh Jadhav
Founder of reactdevstudio. Ten years as a working full stack developer, across SaaS products and client work.
Related, Group A only
Three more on offshore work
Procurement and process posts only, scored within Group A, so a buyer is never handed a React internals piece.