07 · Data & AI

AI Is a Business Tool

Turning AI ambition into practical business outcomes.

1813 AdvisorySelected experience

Don't start with AI

The organisation wanted to use AI.

More accurately, it needed to demonstrate that it was using AI.

The UK insurance company had around £500 million of GWP and 300 employees. Expenses were outside the tolerances set by the Board, while the parent organisation and executive leadership team wanted evidence of AI adoption.

There was no shortage of ideas.

There were vendors ready to provide solutions.

What was missing was a clear answer to a more important question:

What problem are we trying to solve?

That was where I started.

AI is not the strategy

There was no Data or AI strategy.

The conversation had become focused on technology: what AI could do, which solutions were available and how quickly they could be deployed.

I challenged that approach.

AI isn't the objective.

It is a tool.

I created a simple framework around three business outcomes:

Increase revenue.

Improve operational efficiency.

Improve personal productivity.

Every AI opportunity would need to connect to one of these.

That gave the organisation a way to prioritise rather than simply accumulate AI ideas.

I recommended starting with revenue.

We had data. We weren't using it well enough.

The organisation held substantial data across underwriting, claims, reserving and finance.

The problem wasn't necessarily data quality.

It was fragmentation.

Information was held across multiple applications and file stores. There was no centralised data lake, data council, enterprise data dictionary or clearly defined data owners.

Reporting was largely siloed.

The organisation had focused historically on security and accuracy, but accessibility and consistency were less mature.

Good data that cannot easily be found, understood and combined is difficult to turn into value.

And then we found something more fundamental.

The data we didn't have

The business received submissions before underwriting, but this information wasn't captured in a structured way.

There were no formal records of Not Taken Up or Declined to Quote submissions.

The business therefore knew a great deal about the risks it had written, but much less about the opportunities it had considered and rejected.

That limited its ability to learn from its own underwriting decisions.

It also created an opportunity.

If we captured that information, we could use AI to help identify which submissions deserved an underwriter's attention.

This is where garbage in, garbage out becomes important.

The issue isn't simply whether data is good or bad.

It is whether you have the right data to answer the question you are asking.

Find a problem worth solving

The first use case became very specific.

We would capture pre-underwriting information and use AI to assess and prioritise submissions.

The proposition was to:

Automatically pre-process submissions and rank those most likely to be within risk appetite, most likely to convert to a firm order and most likely to produce a profitable outcome.

Now we had something worth doing.

It wasn't an AI project looking for a purpose.

It was a business problem for which AI could provide a solution.

That distinction determines whether AI becomes another technology experiment or becomes part of how the business operates.

Prove value before scaling

This is where my view differed from the executive leadership team.

I wanted to start small.

Prove the technology.

Prove the value.

Learn from the implementation.

Then expand.

The executive leadership team wanted broader adoption. They were under pressure to demonstrate AI across a wider user community and address more complex lines of business.

There was a genuine difference in philosophy.

I was advocating controlled implementation.

They wanted broader adoption.

Ultimately, we followed their preferred approach.

It added cost and time.

There were months of planning and replanning, with decisions made, reversed and remade.

The lesson wasn't that the executive team was wrong.

It was that the order in which you do things matters.

Proving that something works is often more valuable than proving that you can deploy it everywhere.

Build the foundations while solving the problem

I didn't want the AI project to become another isolated technology implementation.

Alongside it, I presented a Data Strategy built around a small number of principles:

Store once. Use many times.

Fix data at source.

Create common definitions.

Establish ownership.

Prioritise use cases.

Build incrementally.

I recommended a Data Council, an enterprise data dictionary, clear data ownership and enterprise-wide Master Data Management.

The objective wasn't to create technology projects for their own sake.

It was to make information a business asset that could support better decisions and future AI initiatives.

Data belongs to the business

This was one of the more difficult organisational changes.

People assumed data was somebody else's responsibility.

Business teams could be defensive about data quality.

Different parts of the organisation sometimes used different definitions for the same thing.

Technology couldn't fix that alone.

The business needed to take ownership.

Data quality is a business responsibility, not a technology responsibility.

That becomes increasingly important as AI is introduced.

A person can recognise that information looks wrong.

An automated model may simply use it.

The more automated the decision-making becomes, the more important the quality and context of the underlying data.

Don't build tomorrow's technology today

I also recommended an AI orchestration layer to manage interactions between AI agents, applications and data stores.

But the principle was more important than the architecture.

I didn't want a collection of disconnected AI solutions.

Equally, I didn't want the organisation spending years designing the perfect AI environment before delivering anything useful.

The approach was incremental:

Solve a real problem.

Capture better data.

Learn.

Build the next capability.

AI changes the business, not just the technology

The organisation also needed to establish its new operating model.

There was significant pressure to move quickly and cheaply.

But AI wasn't simply a technology implementation.

It changed how people worked, what information needed to be captured and where decisions were made.

It also created new responsibilities around data.

The technology therefore had to evolve alongside the operating model.

Automating an existing process doesn't necessarily improve it. Sometimes it simply makes the existing process faster.

What changed

The organisation took its first practical steps into AI.

The pre-underwriting use case provided a real business application rather than an AI demonstration.

Data quality improved.

Decision-making became faster.

A largely manual process gained new capability.

More importantly, the organisation now had:

• a Data Strategy;
• a framework for prioritising AI opportunities;
• clearer data ownership;
• improved data quality;
• a practical AI use case; and
• a pathway for future initiatives.

The conversation had changed from:

“We need to demonstrate that we are using AI.”

to:

“We are using AI to improve something that matters to the business.”

That is the difference between adopting technology and using technology to run a better business.

The lesson

The mistake businesses make when they start talking about AI is treating AI as the destination.

It isn't.

AI is a tool.

Before asking:

“What can AI do for us?”

ask:

“What are we trying to achieve?”

Then:

“What problem are we going to solve?”

And only then:

“Can AI help us solve it?”

That sequence matters.

Sophisticated technology cannot compensate for an organisation that hasn't defined the problem.

And sophisticated AI cannot compensate for information that is incomplete, inconsistent or poorly understood.

Garbage in, garbage out.

But even that doesn't go far enough.

Sometimes the problem isn't bad data.

Sometimes it is the absence of the data you actually need.

The business wasn't capturing the submissions it chose not to pursue. Until we recognised that, no amount of sophisticated AI would have solved the problem.

AI doesn't remove the need for good data. It makes good data more important.

The real opportunity isn't to become an organisation that uses AI.

It is to become an organisation that uses information better to make better decisions.

AI is simply one of the tools that can help you get there.