Everyone wants AI. But what do we actually want it to change?

By Abi Taylor, New client relationship manager, Panintelligence

I spent a day at Big Data London talking to data leaders, product teams, finance teams, founders and software businesses, and unsurprisingly almost every conversation came back to AI.

What was more interesting was how often the conversation started with the technology rather than the problem.

People know they want AI. Boards are asking what the strategy is. Customers are asking when it is coming. Competitors are talking about it. Teams are running pilots, testing tools and trying to work out where AI belongs in their organisation.

But the question I found myself asking most often was much simpler:

What do you actually want AI to change?

That usually shifted the conversation.

Because ask someone what they want from AI and the answer can be broad. Ask a Finance Director what makes month end painful, and suddenly the problem is very specific. Ask a Head of Data where their team loses time, and they can tell you exactly which requests keep landing with analysts. Ask a software business what its customers are demanding, and the answer is often that they expect more intelligence inside the product they already use.

That is where I think the AI conversation needs to move next.

AI is not the outcome.

The outcome is reducing the time it takes to produce a board pack. It is answering a question while it still matters rather than three days later. It is spotting a risk early enough to do something about it. It is helping a customer understand what their data is telling them without waiting for an analyst. It might even be creating a new revenue stream inside a software product.

Those are business outcomes, and importantly, they are outcomes you can measure.

We need to stop looking for AI use cases

One of the things that struck me at Big Data London was how much pressure there is on organisations to find an “AI use case”.

I think that is probably the wrong way round.

If you begin with AI, you naturally start searching for somewhere to put it. That can lead to interesting demos, clever pilots and plenty of enthusiasm, but not necessarily much value.

If you begin with the business problem, the conversation changes.

• What takes too long today?
• Where are people waiting for information?
• What work is repetitive?
• Where are expensive decisions being made too late?
• What do customers want from your product that you cannot currently provide?

Once you understand that, you can ask whether AI is the right way to improve the outcome.

That distinction matters because not every problem needs AI. Sometimes the answer is better reporting, better data, better automation or a better process.

And that is absolutely fine.

In fact, one of the lessons we have learned at Panintelligence is that AI becomes much more valuable when the foundations underneath it are already strong.

One of our customers, Denmaur, reduced standard finance reporting from around 40 hours a month to under four using dashboards and automated reporting. That is around a 90% reduction without generative AI being involved at all.

That matters because the next step is not to replace that foundation with AI. It is to build on it.

Once the data is trusted and the reporting process works, AI can help explain what the numbers mean, answer questions more naturally and help teams look forward rather than only backwards.

The real opportunity is in the work people already do

This is where the opportunity becomes much more tangible.

Think about monthly reporting.

In many businesses, senior people still spend hours every month writing commentary around numbers that have already been produced. They are not creating new insight for all of that time. Much of the work is explaining what moved, whether it was expected and what needs attention.

That is a job AI can help with.

If the first draft of that commentary can be generated from the same governed information behind the dashboard, the person responsible can review, challenge and refine it rather than starting from a blank page.

The outcome is not “we have AI-generated summaries”.

The outcome is that experienced people spend less time describing performance and more time deciding what to do about it and once you frame it that way, you can begin to calculate the value.

If 12 monthly packs each require three hours of commentary, that is 432 hours every year. At an illustrative £50 an hour, that represents £21,600 of time. If AI reduces the writing effort by 60%, nearly £13,000 of capacity is released.

The figures will be different in every organisation, but that is exactly the point.

The business case should be built around your problem, your people and your baseline, not around somebody else’s claim about what AI can deliver.

The same principle applies to analytics teams.

Most organisations I speak to do not have analysts with spare capacity. They have a queue.

And a surprising amount of that queue is made up of questions that are important to the person asking them, but do not necessarily require an analyst to answer.

• Why was revenue down last month?
• Which customers have stopped ordering?
• Which branches are behind target?

The business user knows the question. What they often do not know is where the answer sits or how to navigate the data model.

Natural language can remove some of that friction.

If people can ask those questions in plain English and get to the right governed information, analysts are freed to spend more time on the work that genuinely needs their skills.

Again, the interesting thing is not the technology. It is the capacity you create.

Twenty two-hour analyst requests a month at £50 an hour equates to £24,000 of time over a year. If 40% of those become genuinely self-service, the organisation releases almost £10,000 of analyst capacity.

But there is another benefit that is harder to capture on a spreadsheet.

The answer arrives while the question is still relevant.

That can change the quality and speed of the decision that follows.

Better AI should make existing analytics more useful, not compete with it

There was another theme running through many of the conversations I had.

Businesses do not necessarily have a shortage of dashboards.

They have a shortage of people who confidently know what to do with them. That is an important distinction.

A dashboard can be beautifully built, full of accurate information and still leave someone asking, “So what?”

I think this is one of the most practical roles AI can play in analytics.

Not replacing the dashboard, but helping people understand it.

If an AI capability can explain what has changed, highlight unusual movement and give someone a sensible place to start investigating, it lowers the barrier between seeing a number and understanding what it means.

That is particularly important if organisations want data to be used beyond specialist analytics teams.

For years, the industry has talked about democratising data. In reality, giving more people access to dashboards is not the same as giving more people confidence in using data.

AI has the potential to close some of that gap.

But only if it is grounded in information the organisation already trusts.

Looking forward is where some of the bigger value sits

Reporting has traditionally been very good at telling us what happened.

The more interesting question is what happens next.

If we can see demand starting to soften, capacity heading towards a problem or customer behaviour beginning to change, we have an opportunity to act before the impact is fully felt.

That is where predictive analytics becomes much more commercially interesting.

The value is not the forecast itself.

The value is what somebody can do differently because they saw it sooner.

If a problem costs £100,000 once it happens but £10,000 to address when the warning signs first appear, the value is not measured in the accuracy of a chart.

It is measured in the £90,000 of impact that may have been avoided.

That is the level at which AI and predictive analytics need to be discussed.

Not “what model are you using?” But “what decision does this allow us to make earlier?”

Software companies have another challenge altogether

Some of my most interesting conversations at Big Data London were with software vendors, because their AI challenge is slightly different.

Their customers are asking for AI inside the product.

That creates a real commercial opportunity, but also a difficult build decision.

Adding intelligent analytics is not simply a matter of connecting an LLM and putting a chat box on the screen.

You have to think about permissions, tenant separation, data context, governance, security, user experience and ongoing maintenance.

All of that requires engineering effort, and every day spent building supporting analytics infrastructure is a day not spent developing the functionality that differentiates the core product.

That is where embedded analytics platforms such as Panintelligence can play a different role.

If analytics and AI can sit within the existing product experience, under the vendor’s own brand and within the same tenant and permissions model, the conversation becomes commercial rather than purely technical.

• Could AI support a premium tier?
• Could it help strengthen retention?
• Could it improve adoption?
• Could it differentiate the product in a competitive deal?
• Could it help the vendor respond to customer demand without building an entire AI capability from scratch?

Those are the questions worth answering.

A simple model starts to make the opportunity real. If 200 customers are offered an AI-enhanced tier, 15% adopt it and it adds £1,200 per year, that is £36,000 of additional ARR.

That may or may not be the right number for a particular business but now you have something testable.

The difficult bit is not the demo

The other thing I came away thinking about is how easy it is to make AI look impressive in a controlled environment.

The hard part is making it trustworthy enough to use every day.

AI does not remove the need for good data governance. It should not bypass permissions.

It should not expose information a user was never entitled to see.

And it should not create a separate version of the truth that sits alongside the metrics people already use to run the business.

This is one of the areas where Panintelligence can help.

Our approach is to bring AI into the governed analytics environment, so it works with the same data, semantic context and permissions as the dashboards and reports around it.

That means AI becomes part of the existing decision-making environment rather than another disconnected tool.

For regulated organisations, that is essential.

For everyone else, it is still important.

Because an AI pilot that never gets through security, governance or compliance has no production value, however impressive the demonstration was.

So what should businesses do next?

I do not think most organisations need another conversation about why AI matters.

They know.

The harder question is where it matters enough to invest.

My starting point would be very practical.

Find one piece of work that currently takes too long, costs too much or happens too late.

Measure the baseline.

• How many hours does it take?
• How many people are involved?
• How many requests are sitting in a queue?
• What happens when the answer is delayed?
• What does the problem cost today?

Then test whether AI can materially change that outcome.

This is also how we want to work with customers at Panintelligence.

Not by turning up with a generic AI demonstration and asking you to imagine what you could do with it.

• Bring us the problem.
• Bring us three questions your team asks every month.
• Bring us the reporting process that still takes two days.
• Bring us the forecast that only one person knows how to produce.
• Bring us the customer requirement you cannot currently satisfy.

We can then look at whether Panintelligence can help automate the work, explain the data, make the answer self-service or identify something earlier.

And, crucially, we can agree what success should look like before we start.

• If it saves time, measure the time.
• If it releases analyst capacity, measure the capacity.
• If it improves adoption, measure the adoption.
• If it helps identify risk sooner, measure how much sooner.
• If it creates a new revenue stream, measure the revenue.

That is how AI moves from an interesting technology conversation to a business case.

My biggest takeaway from Big Data London

The appetite for AI is not the problem.

If anything, there is too much pressure to “do something with AI” before organisations have decided what they want that something to achieve.

The businesses that create real value from it will not necessarily be the ones with the biggest transformation programmes or the most ambitious strategy decks.

I suspect they will be the ones that are much more disciplined.

They will identify a real problem. Understand what it costs today.

Decide what a better outcome looks like. Then use AI where it genuinely helps them get there.

That is also how I think the conversation around AI needs to mature.

We should spend less time asking:

“What can AI do?”

And much more time asking:

“What should change if this works?”

That is a question every business can answer.

And once you can answer it, the route to proving whether AI is valuable becomes much clearer.

At Panintelligence, that is where we can help. We can take the problem you already have, work with the data and analytics environment you already use, and help you test whether AI can deliver a measurable outcome.

Because ultimately, the best AI use case is not the cleverest one.

It is the one that changes something that matters.