April 27, 2026
Recently we published a framework for delivering AI in customer experience. It covered six foundations, from data and infrastructure through to people and culture, that we believe will separate the businesses scaling AI from the ones still stuck in proof of concept.
That framework assumes you have already done the hard work that comes before it. Finding the right use case, building a business case that holds up to scrutiny and knowing whether your organisation is in a strong position to deliver.
Almost every business we speak to has a list of AI ideas, often, multiple people in the business have their own list… and in some cases, individuals have started to implement AI across their workflow or in the experiences they deliver to customers.
Examples we have seen range from conversational interfaces for help & support to building AI-driven Tinder style applications (I personally think swiping left or right will be a thing of the past sooner rather than later!)
The problem we have found is not a shortage of ideas, it’s that most of them start in the wrong place.
Don’t start with ‘AI is the answer’
When the starting point is "what can we do with AI?", the conversation naturally gravitates towards technology.
-
What's possible
-
What looks impressive
-
What are competitors doing in the market
What it doesn't produce is a clear answer to the only question that matters.
Does this align to our business objectives, and will it positively impact or help our customers?
The businesses building genuinely valuable AI experiences aren't starting with a list of AI capabilities and looking for somewhere to apply them. They're starting with a willingness to rewire their business.
Step 1: Map the end-to-end workflow
Map the end-to-end experience for a key business function. As a digital consultancy, for us, this typically means the marketing and sales journey. The full arc from brand awareness to continued loyalty and everything in between.
Crucially, map both sides – the external experience your customer has and the internal processes your team runs to deliver it. The handoffs, the manual steps and the moments where things slow down, fall apart or rely on someone knowing something that isn't written down anywhere.
The best AI opportunities don’t just sit in the sit front-end experience. They also tend to live in the operational layer behind it, where high-volume, repetitive decisions are still being handled inconsistently or where the right data exists but nobody's joining it up.
Step 2: Create a longlist
Once you have the journey mapped, think about where the biggest challenges are to solve, are there specific things your customers struggle to do? Is there a clear drop-off point along a journey? Or is something always taking longer than it should? Then ask why – why do these challenges exist? And there lies where you should start to think: How can we solve this problem?
Typically, there are three levers to pull: strategy, operations or technology. The latter including AI, of course.
At this point think how you will use these levers to solve the root cause of the problem and you will begin to assemble a list of use cases.
In our experience, a genuinely useful list of AI use cases is always shorter than the original challenges list. That's a good sign – it means thinking has been applied and we’re not just looking to solve everything with AI.
Step 3: Whittle it down
In our opinion a simple ‘value vs. effort’ matrix is not enough to create a list of meaningful AI use cases. Once you have a shortlist of genuine candidates, you need a way to decide which to back first. Value vs. effort is the default, it's not sufficient.
A use case can look high-value and low-effort until you ask the right questions.
-
What value does it actually drive? Be specific – revenue growth through conversion or through retention? Cost reduction through efficiency or through fewer errors? The use case needs to connect to a measurable outcome.
-
How probable is that value? This is where most prioritisation frameworks fall short. Probability isn't only technical risk. It's also about how direct the link is between implementation and outcome; how much depends on people changing how they work; how well-scoped the delivery is and how resilient the value is if conditions change. A use case with a plausible value story but high probability discounts on all four dimensions is a much riskier bet than it first appears.
This isn't a full business case – that comes next – but it's the rigour that separates a considered shortlist from a backlog nobody ever acts on.
Finding a genuinely good use case is just the beginning
The right AI use case isn't always the most technically exciting one on the list. It's the one that sits at the intersection of a genuine customer or business need, real technical feasibility and your organisation's readiness to change around it.
That last criterion tends to be the one that decides whether AI programmes scale or stall.
In the next piece, we look at how to build the business case once you've found your best candidate. What the numbers need to show, how to pressure-test the value claims and even create rapid prototypes before you take them to a decision-maker.
