5 mistakes companies make with AI

Most companies are not getting AI wrong on purpose. They are getting it wrong by drift.

AI is fizzing into organisations quietly, one curious person at a time. Someone tries a tool, it helps, word spreads a little, and slowly the company is using AI. No plan, no direction, no standard. This feels fine. It can even feel modern. And it quietly caps how much you will ever get out of the technology.

This is a generational chance to transform an entire organisation in one move, and a chance that big deserves a decision from the top, not a slow drift from the edges. Strategy here should be as top-down as it is bottom-up. You will adjust the plan as you go, and that is expected. Having no plan at all is the real mistake.

Underneath the five mistakes below sits a single enemy: the unstructured infiltration of AI. Each mistake is a symptom of it.

1. Individual discovery solves individual problems

You mandate nothing. You let everyone find AI on their own. So each person solves their own small problem, in their own corner, and none of it adds up.

When discovery is individual, the wins stay individual. When discovery is collective, the wins compound across the whole team, building on each other’s prompts, projects and agents. This is the first and most common symptom of the unstructured approach.

2. You optimise for volume over quality

The message becomes “just use it more”. More prompts, more tokens, more output. Usage is easy to count, so usage becomes the target, and usage is a poor target.

Two things happen. Costs climb quietly and can spiral fast once everyone is running heavy models on everything. And quality gets no scoreboard, so the team steers toward doing more rather than doing better.

3. You lock AI away from your own data

AI is only as useful as what it can see. Many companies block or heavily limit access to the documents, tools and data stores where the real knowledge lives. You end up with a capable assistant that has been shut out of the filing room.

Access with sensible guardrails is where the value sits. Locking everything down by default guarantees shallow results.

4. You are not recording your meetings

This is the single highest-leverage habit to start today, and it takes no extra effort. Record and transcribe your meetings.

Knowledge and memory are the foundation for getting real scale from AI, and most of a company’s thinking happens out loud in meetings that vanish the moment they end. Capture them and you build an organisational memory that every future workflow can draw on. It is the easiest first win available.

5. You are not dictating

Writing takes time. Speaking does not. The fastest way to move knowledge out of people’s heads and into a form AI can use is the voice.

Dictate your thinking, your updates and your instructions, and let AI structure them afterwards. This very piece started as a voice note. Teams that still type everything are paying a tax on every idea they capture.

The pattern under all five

Look closely and the same villain sits under every one of these. The unstructured infiltration of AI. It arrives one curious person at a time, with no shared plan and no standard, and that drift produces exactly these five symptoms.

Drift is fine if you are happy with weak, local, occasional gains. If you want to transform the organisation and get ahead of competitors who are already moving fast, drift will never take you there. At the minimum you want to keep pace. Ideally you use this moment to pull ahead.

Begin with an audit of what you already have

Before you roll out a single tool, understand what you actually have. Map every department, every role, every process, and where your data lives.

There are two reasons. First, you cannot rebuild a process you have never mapped. Second, and more important, the audit gives you space to ask the question this technology demands: what are we doing today that we should not be doing at all in an AI-native world?

Apply AI to your existing processes without asking that question and you get small gains bolted onto old habits, and you never reach the point where the process itself is redesigned. This technology is transformative enough to deserve a from-scratch question, and you earn the right to ask it once you can see the whole picture.

How the training works

On day one, before any fixing and before any teaching, we run the audit. We set up calls with everyone who is going to be trained, and we understand each department and each person’s role. People need to be heard first.

That mapping becomes two things at once. A clear picture of how the company actually runs, which AI then helps keep updated over time. And the foundation for the real conversation: what do we keep, what do we change, what is repeatable, and what is not.

From there we apply the right framework for your processes and your industry. A few we lean on:

• Fast to check, delegate it fully. If you can verify the output in seconds, let AI run it.
• Hard to check, keep it in human hands. If verifying is as slow as doing the work, AI saves you nothing and hides the risk.
• Valuable judgment, keep some of it in-house. Hold on to the calls that carry real judgment so the skill does not leave the building.
• The trust matrix. It decides where it is worth letting AI run for long stretches, and where you want small human checkpoints along the way.

We choose the framework that fits where you already are, so the transformation works with your business rather than against it. You can see the full training programme here.

Book a consultation