How do we build an AI roadmap?

If you set off on a journey without knowing the destination, you’ll likely end up getting lost.

And yet that happens a lot with AI, and reports place the failure rate for AI projects between 80% (RAND Corporation) and 95% (MIT). Without a clear idea of where you’re going, and how you’re going to get there, it’s very easy to take a wrong turn and become another example of AI deployment ‘gone bad’.

We grabbed Nick Benzie, Cirrus Head of Solutions Enablement (and, when it comes to AI roadmaps, something akin to a human sat nav), to explain how a good implementation plan for CX AI comes together.

Start with ‘why’, not with AI

An AI roadmap shouldn’t begin with “Where can we use AI?”

It should begin with “What are we trying to change?”

There are lots of reasons you might want to deploy AI within an organisation. But whether these are overtly financial goals such as boosting revenue or reducing operating costs, or more customer-focused outcomes such as extending your service availability or improving accessibility, the technology should always solve a defined problem rather than become an objective in its own right.

So, before you start introducing all-singing all-dancing AI software into your business, be sure you know what you’re looking to achieve and why.

Outline what success looks like

Not every AI outcome needs to be (or can be) measured in pounds.

For charities or purpose-led organisations, success might mean serving more people, increasing accessibility, or improving the level of support available to achieve better real-world outcomes.

What does matter, though, is a sense of what a successful AI deployment will look like and how you’ll go about quantifying that.

If you’re looking to improve customer satisfaction, for example, do you have the right tools in place to accurately analyse their sentiment and capture feedback?

Not only is this useful for assessing your AI deployment, it’s also key for refining your AI over time.

Get your AI ducks in a row

Failed AI deployments often trace back to not getting the basics right around knowledge, integrations, processes, readiness and governance.

When building out an AI roadmap, then, it’s not simply a question of whether AI can achieve what you want (for the most part, yes it can), but a question of whether your organisation is ready for it.

You’ll need to have data and organisational knowledge that is accessible to AI agents, and APIs (Application Programming Interfaces) in place to allow different software to communicate with one another. That’s particularly key as we move away from one-off AI projects and towards true AI orchestration across a business.

Having clarity about your AI governance and compliance might not be the sexiest part of a roadmapping process, but it is absolutely fundamental to success.

Prepare your people, not just your systems

There are other human considerations here, and lots of them. Not all organisations are culturally ready for change, and a workforce that isn’t prepped to embrace AI can feel disenfranchised and limit the potential benefits of a deployment.

Perhaps part of an AI roadmap needs to include a communications campaign to reframe AI not as a replacement or a threat, but a tool to support, supplement and enhance what your people already do.

Sometimes there’s also a need to upskill your people, bring in additional technical expertise, or create new roles to make sure you have the right skills in place to get the most from your investment in AI.

Use evidence, not assumptions

There’s no such thing as a one-size-fits-all AI roadmap because the starting point is different for every organisation. What’s right for someone else will likely not be right for you.

Setting your priorities needs to start with what’s actually happening in your organisation today. Your customer data can tell you a huge amount: why people contact you, the nature of those interactions, where frustrations arise, the choices customers make, and where your existing processes are falling short.

Bringing that evidence together will give you a clearer picture of where AI could have the greatest impact, which use cases are worth pursuing, and what dependencies you’ll need to tackle first.

Ultimately, your roadmap should follow the evidence in your own data, not assumptions about what your customers or your organisation need.

Don’t chase the shiny object

Start where you can make a useful difference. You can save the flashy stuff for later.

For a lot of organisations, automated summarisation is often a strong first use case because it is relatively easy to deploy, saves time, creates consistency, and delivers measurable value quickly.

Other organisations may find workforce management, quality assurance or coaching a more valuable first step.

Those early deployments do more than deliver immediate benefits. They give you a chance to learn what it takes to embed AI within the organisation, understand what works and, crucially, prove its value before moving on to something more ambitious.

Without that established baseline it’s easy to end up funding a series of disconnected experiments and hoping something sticks.

Expect a diversion or two

A roadmap gives you a direction of travel, but it doesn’t mean you have to rigidly follow the exact route from A to B.

Your organisational priorities may shift as the needs of the business evolve. An AI deployment might prove more successful than you expected (or considerably less so), and the data and lessons you gather as you go may point you towards opportunities you hadn’t considered at the outset.

There’ll be changes beyond your business too, as new AI capabilities emerge or regulatory requirements shift.

So, set a destination, but be prepared to adjust the route. And if you’ve put the right governance and fundamentals in place, navigating those diversions should be a whole lot smoother.