Three Types of AI Your Firm Will Be Sold: Only One Is Worth Buying

The AI market for accountants breaks into three distinct categories. Two of them are being sold as the third. Knowing how to tell them apart before you sign any contract that involves handing over client data or stakes your reputation on outputs you can’t explain should now be a core skill for every firm owner.

Category 1: Workflow Automation Dressed as Intelligence

The first category is the most common and the most misrepresented. It’s rule-based automation wearing an AI badge.

Workflow automation can conduct tasks like sending reminders and chasing clients for documents. It can also greatly reduce the volume of manual work that wastes hours of preicous time each week.

However, it can’t reason. It can’t apply professional judgement to an ambiguous transaction.

So, how can you tell the AI tool you’re looking at is just workflow automation? Watch a demo of the tool and look out for pre-set conditions and linear if-then logic. Watch out for what happens when you ask an off-script question. Genuine intelligence can handle ambiguity, but rule-based automation can’t. The vendor selling this as AI isn’t necessarily being dishonest, but the gap between what they’re implying and what the product can do is wide enough to matter.

Category 2: Generic Copilots Bolted On

The second category is more sophisticated and arguably more dangerous, because the outputs are genuinely impressive and the professional risk is less immediately obvious.

This is a frontier model, like ChatGPT, Claude, or Gemini, integrated into an existing accounting or payroll product. It can draft client emails and summarise board minutes at speed, which for most firms, is genuinely useful.

What it can’t do is tell you what a 64-8 actually authorises, whether an R&D claim meets the current HMRC threshold, or whether a payroll submission is correct under post-April 2026 reporting rules.

They may think that they can, but they can’t reliably.

These models are optimised to be helpful, and to produce confident, plausible responses. However, as you know, in accounting, plausible is not the standard. Correct is the standard.

So, how can you tell that you’re working with a generic copilot bolted on? The tell is in the data. Ask where client information goes when the AI processes it. Ask whether it feeds into the underlying model’s training. Ask for a written contractual commitment on data residency and zero-training exclusions. If the answer involves vague references to enterprise agreements or a link to a privacy FAQ, you have a generic copilot. GDPR does not pause just because outputs may look initially impressive.

Category Three: Purpose-Built AI

The third category is the one worth the serious conversation.

Purpose-built AI for accounting is architecturally constrained to authoritative sources like HMRC guidance, gov.uk, FRS 102 and 105, and Companies House. Its knowledge base isn’t the internet, it’s the specific legislative and regulatory environment your practice operates in.

Every AI action produces a full audit trail, including timestamps, model version, input, output, confidence score. This isn’t optional for a profession with HMRC enquiries, PI claims, and client disputes that require you to reconstruct exactly what happened.

HMRC guidance changes too, and a vendor treating their AI as a shipped product rather than a system that must keep pace with the regulatory environment is selling you a piece of software that is simply unfit for the job.

How to Tell Them Apart Before You Buy

The three categories may look identical in a demo, but the differentiation happens when you ask questions that some vendors would rather not answer.

Some examples of such questions include:

• Where does client data go, and is that commitment in writing?

• Where does the AI’s knowledge come from, and how is it updated when legislation changes?

• What are the accuracy benchmarks for the specific tasks you’re being sold?

• Can you replay the full reasoning behind any past transaction?

• Why is this model the right choice for compliance work?

A vendor in category three will answer every one of those questions, in plain language, and with evidence.

So, keep in mind that the firms that get AI right in the next two years won’t necessarily be the ones who moved first. They’ll be the ones who asked the right questions before they moved.

In our free guide “Six Questions Every Accountant Should Ask Before Buying AI”, we set out the full due diligence framework in detail. Every question, answered in writing, with evidence.

This content may have been developed with the assistance of AI tools. Content is reviewed by a member of the Bright team before publication. Guidance on legislation can be obtained from your industry body.