Why this conversation, and why now?
Agentic AI is easy to talk about. Getting it from pilot to production inside a regulated business is where it gets interesting.
That was the whole theme of the latest Future Forward, which Walmer Group co-hosted with Blott and IFX Payments. A full room, two strong panels and some useful conversation about what it takes to make this work when regulators, boards and customers are all watching.
Three things stood out.
1. The hard part is no longer building
Building an agent is not what holds most firms back any more. The harder work happens earlier: choosing the right problem, and deciding what success looks like before anyone starts.
How many pilots have stalled because nobody defined that? Without an agreed measure of success, a pilot cannot prove it worked, so nobody can justify scaling it. It simply carries on being a pilot.
2. Data matters more than the model
A lot of attention goes on which model to use. In practice, the data the model works with matters more.
More data is not better if it is messy or poorly defined. An agent working from inconsistent data will act on it with confidence and get the answer wrong. In financial services and insurance, where data often sits across old systems and product lines, getting it in order is unglamorous work. It is also where much of the value is decided.
3. Governance has to be designed in from day one
When an agent can make decisions or take actions with limited human involvement, governance cannot be added later. It has to be part of the design from the start.
That includes knowing how you revert to manual when something goes wrong, because eventually something will. A fallback that only exists on paper is not a fallback. The people, processes and systems to carry on without the agent need to be ready before they are needed.
Thank you
Thank you to our speakers and moderators across both panels:
- Steven Martin, Monzo
- Umang Sota, Checkout.com
- Kuldeep Mann, S&P Global
- David Scott Turner, Flech (moderator)
- Hasani Jess, Aventum Group
- Katya Kinane, Nettle
- Heather Williams, Brit Insurance
- Michael Robinson, Gable Group (moderator)
A big thanks also to the IFX Payments team for hosting, and to everyone who came along.
The next Future Forward will be announced shortly. If you would like to hear about it first, get in touch.
Key takeaways
- The hard part of agentic AI is no longer building. It is choosing the right problem and defining success before you start.
- Data quality matters more than the choice of model. More data is not better if it is messy or poorly defined.
- Governance has to be designed in from day one, including a tested way to revert to manual.
Frequently asked questions
What is agentic AI?
Agentic AI refers to systems that can plan and carry out multi-step tasks, make decisions and work with other software with limited human involvement, rather than only answering questions or generating content.
Why do agentic AI pilots stall?
Common reasons include no agreed measure of success, poor or inconsistent data, and governance that was not designed in from the start, which makes it hard to scale safely in a regulated business.
What is Future Forward?
Future Forward is an event series co-hosted by Walmer Group and Blott, focused on AI, deep tech and how innovative companies grow and are funded.
