Self-assessment

Where is the business with AI, really?

Twenty questions about where a business actually stands with AI, rather than how many tools it has bought.

For the chief executive, finance director, operations director or owner of a business of roughly twenty to five hundred people. 20 questions about 6 minutes

Izam Ryan is a chartered accountant who builds with AI. He has spent years checking other people's figures on behalf of whoever was paying for the answer. More recently he has been directing AI models to design and ship working software. Both land on the same question: not what a tool can produce, but whether anyone can show how it got there.

Most AI checklists list tools and pilots, which is the easy thing to list and the least useful thing to know. These questions ask instead whether there is a reason for doing this that ties to the numbers, whether the work has been re-engineered or merely sped up, whether an AI-assisted result can be checked by someone who did not produce it, and whether anyone knows where the line sits between what the machine decides and what a person decides. When producing an answer becomes cheap, the scarce things are evidence and trust.

It is a thinking tool, not advice, and not a view on your business, which Izam has not seen. Answer from memory, a few seconds each. Having to look something up is itself an answer.

The answers stay in this browser. Nothing is recorded anywhere unless the Send control at the foot of the read-out is pressed.

Section 1 of 4

Why the business is doing this

Most AI spending starts with a tool and then looks for a use. The questions that decide whether it comes to anything are duller: what it is for, who owns it, what gets stopped, and what it does to the price a customer will pay.

01 Is there a stated reason for using AI here that connects to revenue, cost, risk or the service customers get?

Not a technology plan. A sentence a shift supervisor and a finance director would both recognise as being about this business.

02 Do the people who run the business use these tools themselves in their own work?

Directors who have never used one tend to over-estimate what it can do and under-estimate what it takes to make it reliable.

03 Is one named person accountable for AI across the business?

Named, and known to be named by the people affected. A committee with no chair is not an owner.

04 When a new use for AI comes up, is there a settled way of deciding whether to build it, buy it ready-made or wait?

A settled way also says when something gets stopped. Waiting is a legitimate decision, and nobody has to be early.

05 Has anyone worked out what cheaper AI-assisted work does to what this business charges and what its customers will expect?

It applies whether the business sells hours, units, places or a monthly charge. The change usually shows up first in what customers ask for, not in what they pay.

Section 2 of 4

How the work actually gets done

Speeding up an existing process leaves the process intact. The gains that last come from re-engineering the work itself, and they only appear where people are genuinely using the tools and somebody knows what the work used to take.

06 Do you know how many people in the business used an AI tool for work in the last week?

Including tools nobody approved and personal accounts on personal telephones. Unrecorded use is still use.

07 Has any piece of work been redesigned around what AI can now do, rather than having AI added to the existing steps?

Redesigned means the steps themselves changed: an estimate produced differently, a goods-in check done differently, a care plan reviewed differently. Added means the same steps, done faster.

08 Could someone find the current version of a document or a record without asking a colleague where it is?

Findable, current, and with permissions that say who may see what. Roughly what makes information usable by a person is what makes it usable by a machine.

09 Have the people expected to use AI tools been shown how to use them on this business's own work?

A general course is not this. The test is whether someone was shown on a real job of the kind they actually do.

10 Where AI has been introduced, is there a record of what the work took before?

A baseline is simply the before figure: hours, error rate, days to answer a customer. Without one, an improvement can be asserted but not shown.

Section 3 of 4

Work the machine does on its own

An automated step is one that runs without anyone watching it. What it is worth depends entirely on whether the business would notice when it went wrong, and whether it could be stopped.

11 Are there repeatable pieces of work where an AI tool does a defined task from start to finish?

A workflow here means a defined sequence with a beginning and an end: sorting incoming enquiries, drafting a first reply, checking a document against a standard, routing a support ticket.

12 If an AI-assisted step produced a wrong result today, how long would it take for someone to notice?

The honest test is what happened the last time something was wrong, not what the process says should happen.

13 Is it written down which decisions belong to a person rather than to a tool?

Refusing an application, releasing a payment, agreeing a discount, changing what a customer is charged, changing what care someone receives.

14 If the supplier behind an AI tool the business relies on changed its terms or stopped tomorrow, could the work still be done?

This includes tools reached through another system the business already pays for.

15 Could an automated step be switched off today without the work it sits in stopping?

The question is whether the business could still do the work by hand, more slowly, for a few weeks.

Section 4 of 4

What you could show someone else

The scarce things are not tools. They are evidence and trust: whether an AI-assisted result can be traced and arrived at again by someone who did not produce it, and whether anyone can say where responsibility sits when it goes wrong.

16 Is there a short written note of how AI tools may be used here that the people using them have read?

Short means a page. It should say what is allowed, what is not, and who to ask when it is neither.

17 Do you know whether confidential, personal or customer information has been put into AI tools that were never approved for it?

Personal information here means anything about an identifiable person: staff, customers, residents, patients, applicants.

18 Has anyone read what the suppliers of these tools do with what the business puts into them?

The two lines that matter are whether the supplier may train on what you send, and which country it is processed in.

19 If an AI-assisted result were challenged, could someone else trace it back to its sources and arrive at it again?

To reperform something is for a second person, working from the same inputs, to reach the same answer. It is what turns an output into evidence.

20 If an AI tool caused a problem today, is it clear who would be told?

And whether that person has the authority to switch the system off while it is looked at.

0 of 20 answered