More AI ideas do not mean more AI value
It has become relatively easy to come up with potential applications for AI. Ask ten teams where AI could help them and you are unlikely to leave the room short of ideas.
That can create a problem of its own. When every function has its own shortlist, pilots can end up being driven by enthusiasm, available budget or the most vocal sponsor. What is often missing is a consistent way to compare one opportunity with another. Before funding an AI use case, I think there are four basic questions worth asking.
- What is it worth? What business outcome are we trying to change, and can we measure it?
- Are we ready to deliver it? Is the necessary data usable? Can the solution connect to the systems and processes it will depend on?
- What could stop it? Are risk, compliance, security and accountability understood early enough to shape the solution?
- Can we reuse what we build? If it works, can it be applied elsewhere, or are we creating another isolated solution?
These questions sound straightforward. In practice, they can be difficult to answer consistently across a large organisation. That matters because the objective should not be to get as many use cases into a pilot as possible. It should be to invest in the opportunities that have a credible route to business value.
A successful pilot answers only part of the question
Getting a pilot to work is important. But a controlled experiment is very different from a live banking environment. Production means connecting to real systems. It means working with real data and existing architecture. Someone needs to own the solution once it goes live. Risk and compliance teams need to understand it. Monitoring, validation and human oversight need to be clear. And the people who will actually use the solution need to be ready for it. This is where promising AI initiatives can lose momentum.
The issue is not necessarily that the technology failed. Often, questions that were relatively easy to ignore during experimentation become unavoidable when the organisation tries to put the solution into production.
Can the data actually be used in this context? How much work is needed to integrate the solution with existing platforms? Who is accountable for its outputs? Where does a human need to remain in the process? What happens after launch?
If those questions first appear at the end of a pilot, there can be a significant amount of work still to do.
Measure the outcome, not the pilot
There is another question I think financial institutions need to ask earlier: what exactly will make this investment worthwhile? AI pilots naturally tend to focus on whether the technology performs. Does the model produce a useful answer? Is the output accurate enough? Can the task be completed?
Those are necessary questions, but they are not the same as measuring return. The real measure sits in the process the AI is intended to improve. Perhaps the objective is to reduce the time required to complete a task. It might be to improve quality, reduce operational effort, lower risk or help colleagues make better decisions. In another case, it could be about customer experience or revenue.
Whatever the objective is, it needs to be explicit. If business and finance cannot agree what outcome an AI initiative is expected to deliver, it becomes very difficult to decide later whether that initiative deserves to scale. This is why value needs to be part of the conversation before the pilot starts, not something calculated once the technology has already been built.
Governance should not arrive at the end
The same is true of governance. In financial services, risk and compliance cannot simply be the final hurdle between a successful pilot and production. Their requirements influence how the solution should be designed in the first place.
What data can it use? What needs to be validated? How will the organisation explain or monitor its outputs? Where is human oversight required? Who owns the decision when AI is involved?
Bringing those questions in early does not mean slowing innovation down. Quite the opposite. If teams understand the boundaries from the start, they can design within them. The alternative is to build a promising pilot and then discover that significant parts of it need to change before the business can use it.
For me, that is an important distinction. Good governance is not simply about approving AI. It is about making AI deployable.
The real capability is a repeatable route to production
One successful AI solution is useful. A repeatable way of getting the right solutions into production is much more valuable.
That means being able to assess opportunities against the same criteria, stop ideas that do not justify further investment and move the strongest ones forward with a clear understanding of what production will require.
It also means looking beyond the individual use case.
Is the data ready to use, rather than simply available somewhere in the organisation? Can new AI solutions connect to existing platforms without major rework? Are operations teams involved before launch? Can something built for one function or market be reused elsewhere? And once a solution is live, does somebody clearly own it?
These are less exciting questions than what the latest model can do. But they are the questions that determine whether AI remains a collection of experiments or becomes a capability the organisation can use repeatedly.
Where GFT's AI Experience Framework fits
This is the thinking behind GFT's AI Experience Framework, or AIX.
AIX helps organisations look at the business ecosystem first, identify and prioritise the AI opportunities that are worth pursuing, and define the technology and roadmap needed to move them forward. Governance is considered as part of that journey, rather than as a separate exercise once a solution has already been built.
The aim is not to generate another list of AI ideas. It is to create a clearer route from opportunity to business outcome and, where the case is strong enough, to scale.
Where is your organisation getting stuck?
There is no single starting point. For some financial institutions, the challenge is still identifying where AI can make a meaningful difference. Others have no shortage of ideas but need a better way to decide which ones deserve investment. Some have already proved that the technology works and are now facing the harder task of getting it into production. So rather than asking how many AI initiatives are under way, I would start with three simpler questions.
- Can you identify the opportunities that are genuinely worth pursuing?
- Can you prove the business value, not just the technical feasibility?
- And when something works, can you take it into production and scale it without starting from scratch?
Identify. Prove. Scale. Knowing where that process becomes difficult is often the best place to start.