The "Don't Bother" List: What Most Mid-Sized Companies Get Wrong About AI

Most mid-sized companies don't have an AI problem.

They have a "we bought ChatGPT licenses and nothing changed" problem.

The tools work. The models are capable. What's missing is the boring middle part: figuring out which of your actual workflows are worth automating, and which ones AI will just make slower and weirder.

Why the project list is the wrong starting point

Every AI vendor wants to give you a list of use cases. Chatbot for customer service. AI for your HR process. Predictive analytics for your supply chain. The list is long because it's designed to impress, not to fit your operations.

The uncomfortable truth is that the most valuable thing we can give a new client isn't a list of AI projects. It's the "don't bother" list.

The three things everyone tries first that waste three months. And the two nobody looks at that pay for themselves in six weeks.

What the wasted three months usually look like

The patterns repeat across industries. Teams spend months on:

• An AI chatbot that can't answer questions about their actual products because it has no access to their actual data
• A document summarisation tool that handles the clean PDFs fine but breaks on everything in the real archive
• An AI-generated report that still needs two hours of human review before anyone trusts the numbers

None of these fail because AI doesn't work. They fail because the workflow wasn't mapped before the tool was purchased.

The two things that actually pay off first

The highest-ROI early AI wins are almost never the ones leadership reads about in a McKinsey deck. They're the repetitive, unsexy workflows that live in the middle of your operations:

• Reporting and data re-entry — the work where someone exports a file, cleans it in Excel, pastes it somewhere else, and formats it for a meeting. Every week. Forever.
• Structured data extraction — pulling specific fields out of inbound documents (invoices, contracts, applications) that currently require a person to read and manually enter them

These aren't glamorous. They're also where 300–500 hours a year of senior-ish people's time quietly disappears.

How to build your own "don't bother" list

You don't need a consultant to do a first pass. Ask your team two questions:

1. What's the one task we do every week that a computer should obviously be handling by now?
2. What's the one thing that only one person knows how to pull, and everyone else dreads when they're out sick?

The answers will point you directly at your first real automation — not an AI strategy, just an actual problem with an actual solution.

The point of this exercise

The goal isn't to be contrarian about AI. The goal is to make sure your first move generates a result that justifies the second move. Most companies that are sceptical about AI didn't start with the right problem. Most companies that are enthusiastic about AI haven't started at all.

If you want a structured way to identify your highest-value automation opportunities, our AI Readiness Scorecard walks through this in about five minutes: take the AI Readiness Assessment.