← Blog

How to choose an AI consulting firm: seven questions

I first wrote this guide in 2019, when companies were hiring data science firms. Today they’re hiring AI firms, and there are far more of them to choose from. The questions haven’t changed much. The answers have.

These are the seven questions I’d ask before signing with any outside AI firm, including ours.

1. What could AI do for your business?

Start with what you want to improve. AI pays off when it makes a prediction or a decision that changes a metric your business already tracks: orders that ship late, customers who leave, stock that sits too long, crews sent to the wrong job.

Here are two examples from our own work. For Irth, we built a model that scores every call-before-you-dig ticket for damage risk. Half of all incidents fall in the riskiest 8% of the tickets it flags, so field teams know where to go first. That’s the damage-risk model we built for Irth. For a Fortune 250 consumer goods company, we built one analytics platform for seven brands, and a portfolio report that took an analyst about a week now runs in minutes. That’s the analytics platform we run for seven consumer brands.

If you can name the metric and where it’s recorded, a good firm can tell you quickly whether AI can improve it.

2. Is it the right time to invest?

AI won’t fix an unstable platform or missing data. Before you bring anyone in, ask your data and engineering teams:

  • What data do we collect, and how far back does it go?
  • How reliable is it, and who would know?
  • Can the metric we care about be calculated from it today?
  • Is our platform stable enough to build on?
  • Who on our side will give an outside team access and answer its questions?

If those answers are hard to get, that’s worth knowing. It usually means the first job is getting the data ready. I wrote about that in why so much of an AI project goes to cleaning data.

3. What are the risks of hiring an outside firm?

I’d watch for four.

Nobody has time to get the firm the data. An outside team needs access to your data. If your people can’t pull it or explain it, the work stalls in the first month.

Your own team feels threatened. If your engineers see the firm as their replacement, they’ll resist it. Talk to them before you sign. Tell them what the firm will do and what stays with them.

Leadership isn’t behind it. If you’re doing this because the board keeps asking about AI, think again. Projects without a senior sponsor rarely make it into daily use. I cover this and four other causes in why AI projects fail to pay off.

Nobody planned for production. A model only pays off once it runs inside your systems and people use it every day. That takes time from your engineering team, so budget for it from the start. I cover what’s involved in putting AI into production.

4. What do you get from an outside firm?

Companies usually bring in outside help for one of three reasons. They have no AI team, they have one that’s overloaded, or they have one that needs a different structure. In each case, a good firm gives you four things.

Speed. Senior engineers who have done the work before skip the learning curve. In my experience, what a senior team delivers in six months can take a newly hired junior team a year and a half.

Less to manage. Experienced engineers need little supervision, and a good firm explains its work to your executives in plain language.

Costs you can defend. The fee should be a small fraction of the value the work creates, and both should be agreed in writing before work starts.

Quality you can check. AI that’s built badly breaks the first time the data changes. Ask every firm for work that is running in production today, at scale, and ask to speak with that client.

Two more things belong in the contract. You should own everything built for you: the code, the models and the documentation. And the work should happen inside your systems, so your data never has to leave. That’s how we set up every engagement.

5. How do you find and interview a firm?

Above all, you need a firm you trust. Ask peers who have put AI into production who they used. Look for people who write in detail about the kind of work you need, and write to them directly. Talk to several firms. A good one will tell you quickly whether it can help, and if it can’t, it will often point you to someone who can.

Be careful with claims. A lot of firms put “AI” on their website because it helps them sell. I’ve seen phrases from my own first calls show up on other firms’ websites a few days later. The fastest test I know is to ask to speak with the person who leads the firm’s technical work. In most cases, you’ll know right away.

In the interviews, look for these:

  • Similar work. A firm that has built something close to what you need, in a business like yours.
  • Your systems. Experience with your cloud, your data warehouse and your tools. If they don’t know them, your team ends up doing more of the work.
  • Your choice of models. Ask which AI models the firm builds on and how hard it would be to switch later. We use models from Anthropic, OpenAI and Google, chosen per task, and the rest of the system stays the same when the model changes. You can see how in how Bennett ONE works.
  • Clear communication. Ask how they kept a past client informed. If they can’t describe it, you won’t know what they’re doing once they’re inside your company.
  • The people. Meet the engineers who will actually do the work. With us, the same senior engineers stay with you from the first study through live operation.

6. How will you measure success?

Agree on one metric before any work starts: where it’s recorded, where it stands today, and what you want it to be. Late shipments, churn and cost per order are all good examples. Every result should be reported against that starting value.

That’s how we start every engagement. The first two calls end with one workflow and one metric, and a fixed-fee Discovery measures where that metric stands before anyone builds anything. You can read the steps in how we work with clients.

Stay open to changing course if the firm finds something worth more than the original goal. Just change the metric on purpose, in writing.

7. What happens when the engagement ends?

The firm will probably want to keep working with you. That’s good for both sides when the work is paying off, because a team that already knows your data and your people is worth a lot.

Decide early how it ends. With Bennett ONE, you choose how it runs after go-live: our engineers run it, we run it together with your engineers, or your team runs it and we stay on call. You can switch at any time, and everything we built is yours in all three. The Bennett ONE questions and answers cover the details.

Whatever you agree, put the hand-off plan in writing at the start.

Talk to us

If you’re weighing an outside firm right now, the first call with one of our senior engineers is free and takes 30 minutes. We’ll tell you plainly whether we’re a fit.

Let's talk about what you want to improve

Talk to an engineer