To choose an AI automation agency, ask what it would fix first, how you will both know it worked, who owns what it builds, what it costs to run each month and who looks after it once it is live. An agency worth hiring answers those questions in plain words before it talks about tools, and tells you when an off-the-shelf product would do the job.
A disclosure first: CentriWeb is an AI and automation studio, so we are one of the businesses you might be comparing. We don’t name any agency here. The questions below work the same whoever is across the table, us included.
What an AI automation agency actually does
The label covers very different businesses. Before you compare prices, work out which kind you are talking to.
- Builders. They connect your existing software, set up AI that answers enquiries or handles admin, and test it against your real work.
- Advisers. AI consulting without the build: workshops, a roadmap, a list of tools. Useful if you have people to carry it out.
- Resellers. They set up one platform for you, often on a monthly fee. Fine if that platform suits you, less so if every problem gets the same answer.
- Trainers. They teach your staff to use general AI tools well. Often the cheapest useful first step.
None of these is wrong. Problems start when you hire one kind expecting another, such as paying for a roadmap when you wanted something working by the end of the quarter.
Before the first call: three things to have ready
You will get far better answers if you arrive with one clear problem instead of a wish list.
- The job that costs you most. Missed calls, slow replies to enquiries, quotes nobody follows up, hours of data entry. Pick one.
- Rough numbers. How often it happens a week, how long it takes each time, and what a lost customer is worth. Estimates are fine.
- Who owns the process. The person who does the job today needs to be in the conversation, because they know where the exceptions are.
A useful filter: AI and automation are worth paying for when they save time, bring in more money, or keep money the business already makes. If the job you have picked does none of those, it can wait.
Ten questions to ask an AI automation agency
1. What would you fix first, and why that?
A good answer points at the part of your business that costs the most and explains the reasoning. Be wary of a proposal that starts with ten things at once. One working system teaches both sides more than a big plan.
2. How will we know it worked?
Ask them to write down the measure before they build: replies sent within an hour, calls answered after hours, admin hours saved. That means recording where you are today. Without a starting figure, nobody can say whether it worked.
3. Is there a cheaper product that already does this?
Plenty of jobs are handled by a feature in software you already pay for. An agency that says so, even when it costs them the work, is telling you something useful about how it will treat you later.
4. What will the AI answer from?
AI that answers customers should answer from your own information: your services, prices, policies and calendar. Ask how they load that information, how they keep it current, and what the AI does when it doesn’t know. See our guide to reducing wrong AI answers for what good looks like.
5. What happens when the AI can’t help?
Every system needs a handover: to a person, a callback or a message in the CRM. Ask them to show you the handover, not describe it. Ask who gets notified and how fast.
6. Where does our data go?
Customer details pass through the AI model, the automation tool and any platform in between. Ask which companies handle the data, where they store it, and whether any of them use it to train their models. If your business is covered by the Privacy Act, read our guide to AI and the Privacy Act and check the rules for your situation.
7. Who owns the build and the accounts?
You want the software accounts, phone numbers, automation workflows and AI settings in your business’s name, with the agency given access. If you part ways, you should keep everything that runs your business. Our guide on who owns the code goes through what to put in writing.
8. What will it cost to run every month?
The build is only one part. There are model usage charges, software subscriptions, text and call charges and upkeep. Ask for each line separately, with the assumptions behind it. Our guide to AI automation costs explains each part.
9. Who looks after it, and what does that include?
Automations break when an app changes, a form gets a new field or a staff member leaves. Ask what monitoring is in place, who fixes things, how quickly, and what counts as a change you pay extra for.
10. Can we try it on something small first?
A short first stage on one job, with an agreed measure, lets you see how they work before you commit to more. Ask what you keep if you stop after that stage.
Pay attention to how they run that first stage. Do they ask to sit with the person who does the job today? Do they test with real enquiries from your inbox, or made-up ones? Do they tell you what went wrong as well as what went right? The way an agency handles a small piece of work is the best preview you will get of how it handles a large one.
Warning signs
- Guaranteed results. Nobody can promise a number of extra sales from software they haven’t seen working in your business.
- Tool first, problem second. If the pitch is about a platform before they have asked how your business runs, the platform is what they sell.
- Accounts in their name. Convenient at the start, a problem the day you want to leave.
- No testing plan. AI that talks to customers should be tested on real questions, including awkward ones, before it goes live.
- “It will replace your staff.” Most good systems take the repetitive part of a job and hand the rest to a person.
- Monthly fees with no description. You should know what the fee covers and what it doesn’t.
What a good first meeting feels like
You should come away knowing more about your own business than when you sat down. A useful first conversation spends most of its time on how enquiries, bookings and admin move through your business today, and where they get stuck. Tools come up near the end, if at all.
Notice who does the talking. If you barely got a word in, that pattern tends to continue.
Comparing proposals side by side
Lay the proposals out against the same questions. Differences show up quickly.
| Question | A clear answer | A reason to ask more |
|---|---|---|
| What gets fixed first | One named job, with reasons | A long list with no order |
| How success is measured | A number recorded before and after | “You’ll see the difference” |
| What the AI answers from | Your own information, kept current | General AI knowledge |
| Handover to a person | Shown working, with notifications | Described but not shown |
| Ownership | Accounts in your name | Accounts in theirs |
| Running costs | Itemised, with assumptions | One bundled figure |
| Upkeep | What is covered and response times | “We’re always here” |
When you don’t need an agency at all
Sometimes the right answer is to do it yourself or buy a product.
- Your software already does it. Many CRMs and booking tools have built-in reminders, follow-ups and simple chat.
- The volume is low. If the job happens a few times a month, a checklist may be enough.
- Someone on your team enjoys it. An owner or manager with time and interest can set up a lot with standard tools and some patience.
- The process isn’t settled. If your team does the job three different ways, agree the steps before paying anyone to automate them.
If you are not sure where AI fits in your business yet, our guide to where to start with AI is a better first read than any proposal. If you do want it built and looked after, our AI services page explains how we work.