Five steps to Agentic AI

1) Don’t start with the data

There’s a comforting narrative that you need to “fix your data first”. You don’t. You need good data to scale, not to start. For many organisations, achieving high-quality data foundations is a 2–3 year journey. Most CFOs expect AI ROI in under three. If you wait for perfect data, you’ll run out of road before AI can show results.

2) Start small. Start with an SME.

Pick a use case that isn’t heavily dependent on data quality, and pair it with a true subject-matter doer, the person who knows the business process inside out.

Agentic pilots fail 95% of the time, not because of technology, but because the real business problem only becomes clear halfway through. An SME who lives the process avoids that trap. This explains why many more organisations are successful with AI when it’s driven out of back-office, marketing or commercial functions rather than the CTO.

3) Be cheap, deliberately

Your first Agentic step isn’t about money. It’s about organisational learning. You find the blockers you did not expect. You learn how your fellow executives respond when they see an agent in action. You learn how colleagues feel when they think AI might reshape their role. You also get a better sense of how ambitious you can be.

This insight is far more valuable than a big first investment. Low-cost intelligence that saves six figures later. The sweet spot for many clients is a PoC that is under £50,000. This magic number means you don’t have to get an FTE release and org change to get a two-year ROI. The whole process becomes so much simpler.

4) You probably already have the kit.

Most enterprise solutions already had an added agentic AI component. If you have Google Cloud, Amazon Web Services (AWS), ServiceNow etc you can build an agent with tech that your Security team has already signed off. No lengthy onboarding with a known quantity and less risk.

5) Then comes the hard part: scaling

Once a pilot works, the questions evolve fast: How do we measure ROI and build a CFO-ready business case? What governance do we need, and how do we detect and shut down a rogue agent automatically? How does our operating model change when humans and agents work side-by-side?

This is the point the challenge becomes less about experimentation and more about safe, repeatable adoption.

If you’re exploring how Agentic AI can work in your organisation, we’re helping clients through exactly these questions. Happy to share what’s working and what isn’t.