Operational AIProcess and Knowledge

Five Questions to Ask Before You Let Anyone Automate Your Business

The number of firms offering to automate your business with AI has grown enormously in the last two years. Some of them are very good. A lot of them were doing something else in 2024.

You cannot evaluate this by looking at a demo, because demos are the easy part. These five questions are harder to answer well, and the answers tell you most of what you need to know.

1. What happens when it gets something wrong?

Ask this first, and listen for whether they have thought about it before you asked.

A good answer covers three things: which actions require a person’s approval, how you would find out about a failure, and where a correction gets stored so the same mistake stops recurring. A weaker answer is some version of “the model is very accurate now.”

Every system fails. What you are buying is how it behaves when it does.

2. What are you actually connecting to, and how?

There is a large difference between reading and writing live through an API, and someone exporting a file on a schedule. Both get called integration.

Ask which of your specific tools they have connected before, and what happens when one of those tools has no API. The right answer sometimes is “that one has no API, so that step stays manual or needs browser automation, which is more fragile.” Somebody who says everything is possible either has not checked or is not telling you.

3. Who owns it, and what happens if we stop working together?

Ask where it runs, who holds the credentials, and what you are left with if the relationship ends.

This matters more with automation than with a website, because the value accumulates in something less tangible: the written-down knowledge of how your business works. Find out whether you get that, and in what form.

There is no single correct arrangement here. Hosted and operated for you is a legitimate model, and so is installed in your own environment. What is not acceptable is not being able to get a straight answer.

4. How is this priced, and what does phase two cost?

Open-ended AI projects have a way of becoming expensive without anyone deciding they should.

Fixed scope per phase is the arrangement worth asking for: a defined piece of work, a defined price, and a decision point at the end about whether to continue. If someone cannot tell you what the first phase costs before starting, that is worth taking seriously.

Ask what ongoing costs look like too, including model usage, and whether that sits inside a retainer or arrives separately.

5. Can you show me something you run yourselves?

This is the one that sorts the field fastest.

Anyone can build a demo. Running something on your own business every day for a year is different, because you have to live with the failure modes rather than presenting past them. Ask what they automated internally, what broke, and what they changed as a result.

If the answer is that they do not use this on their own operations, that is worth understanding. It is not automatically disqualifying, but it should prompt a follow-up question.


For what it is worth, our answers to all five are on this page, and we run our own helpdesk, intake, DNS and content pipeline through the same system we sell. If you are talking to several firms and want a second read on a proposal in front of you, get in touch.