What questions should you ask before hiring an AI agency?

Hiring an AI agency is a different purchase from hiring a marketing agency or buying software. Most owners evaluate on the demo and the portfolio, then discover what the contract actually bought a year later. This note covers the five questions that surface what you are actually buying, before you sign.

What should you ask an AI agency before you sign?

Ask questions in five areas before signing with any AI agency. Who owns what gets built. How results will be measured. Where human review sits. Who actually does the work after the contract is signed. And what your firm keeps if the relationship ends. The demo and the portfolio answer none of these.

A demo answers one question, which is whether the agency can build something impressive. The five areas answer a different and more expensive question, which is what your firm holds in twelve months. Plenty of firms have hired an agency that passed the first test and failed all five of the second.

• Ownership. Who holds the repository, the accounts, and the API keys when the build is done.
• Measurement. What revenue the system should create and what margin it should protect, written down before the build starts.
• Review. Which outputs can reach a client without a person at the firm approving them first.
• Staffing. Who scopes the work, who builds it, and who answers when something breaks.
• Exit. What the firm walks away with, in writing, if it leaves in twelve months.

Who owns the system the agency builds?

Ownership is the first question because it decides what every other answer is worth. Ask who holds the repository, the accounts, the API keys, and the prompts when the build is done. If any of those stay with the agency, the firm is renting access to a capability rather than buying one.

The stakes grow with time. A growth system accumulates knowledge of your clients, your voice, and your pipeline every month it runs. If that accumulation lives in the agency's infrastructure, the cost of leaving rises on the same schedule, which is convenient for one party in the relationship and it is not the firm.

We build inside the firm's own accounts and hand over the repository, keys, documentation, and training at handoff. We hold that standard because it forces the next engagement to be earned on results. Whatever agency you talk to, the useful move is asking the ownership question early and watching how quickly it gets answered.

How will the agency measure whether the work paid?

Ask the agency to name the revenue the system should create and the margin it should protect, before the build starts. An agency that builds for outcomes can put numbers next to the work. An agency that cannot is selling activity, and activity is what fills the quarterly review when revenue does not.

Be careful with time-saved pitches. Saved hours do not appear on a P&L unless they become billable work or new revenue, so a proposal built entirely on efficiency is a proposal with no number a CFO can check. The stakes behind this discipline are real. Deltek's industry study found professional-services EBITDA fell from 15.4% to 9.8% over five years, and a purchase that cannot defend itself in margin terms adds to that pressure instead of relieving it.

Ask one follow-up as well. How will the builder define success before work starts. North Signal does not sell a vague AI project. Each $15,000 agent build starts with the business problem, the owner review rules, and the handoff standard written down before the first workflow is built.

Where does human review sit in the agency's process?

Ask which outputs can reach a client without a person at your firm approving them first. The right answer is none. An AI system should draft, prepare, and queue client-facing work, and a person who owns the relationship should approve everything before it goes out.

The reason is asymmetry. A reviewed draft costs the firm two minutes of partner attention. An unreviewed message that misreads a relationship can spend trust the firm built over years, and the firm finds out after the client does. Agencies that promise full autonomy are optimizing for volume, and volume is the wrong goal for a business whose revenue runs on trust.

Treat review gates as a design feature to inspect, not a limitation to negotiate away. Ask to see where approvals happen, who gets notified, and what the system does when a draft is rejected. An agency that has thought about review will show you the mechanism. An agency that has not will tell you the model is very good.

Who actually does the work after you sign?

Ask who scopes the build, who writes the system, and who you call when something breaks. At many agencies the senior people sell and the work moves to whoever is available, so the person who understood your firm in the sales call may never touch the build.

The fix is simple to ask for and revealing when refused. Ask to meet the person who will do the work before signing. Then ask what happens if that person leaves. The honest answer involves documentation and training your own team can use, which is also a preview of how seriously the agency takes handoff in general.

Where will your client data live?

Ask where your client data lives, who can access it, and what happens to it if the relationship ends. Client memory is the most sensitive asset in any growth system, and it should sit in accounts your firm controls rather than on the agency's platform.

This question carries double weight. The client context is what makes the system valuable, because a draft grounded in real relationship history is the difference between attention and spam. The same context is a liability if it lives somewhere the firm cannot see, audit, or delete. A firm that answers to clients for confidentiality should be able to answer this question without checking with its vendor.

What happens if you leave in twelve months?

Ask for the exit answer in writing before the first invoice. Walk through the scenario together. The engagement ends on good terms in twelve months, and you list what the firm still has on day one after. If the honest answer is a login that no longer works, you are evaluating a subscription, not a build.

Good agencies answer this quickly and in writing, because the answer was designed before you asked. Hesitation is information too. A vendor who stalls on the exit question has told you where the power in the relationship will sit a year from now, and you learned it for free.

Bring these questions to every agency you talk to, including us. The Growth Audit Call is where we answer them about our own work, and it doubles as a map of the growth work your firm keeps dropping and what closing that gap is worth. If the timing is right, we are easy to find.