How Matillion Secures Its AI Agents and Knows What They’re Worth

Growing AI agent use, safely and with purpose

Matillion are betting big on agents: their goal is to move every employee beyond chatbots to agentic AI that delivers real outcomes, safely but quickly. Hack days and internal events encourage people to experiment, and the strongest ideas move into a central AI or engineering team to be governed. This year raised the stakes, with a new MCP layer connecting agents directly to Matillion’s systems.

Matillion brought in Geordie as a security proof of concept, and it quickly became something broader: a single view of the agent estate that showed security, GRC, finance and senior leadership the risks Matillion’s agents carried and what the spend behind them was delivering.

Today, that view gives Matillion the confidence to push adoption faster.

Risk Intelligence: Surfacing and remediating the most critical agentic risks

Matillion saw a 50% increase in the size of their agentic estate quarter-on-quarter. "We end up with a lot of risks as a result of that increase in activity, but Geordie's helped us put a light on those,” Graeme explained. “Having Geordie as part of our stack has really enabled us to accelerate and continue that growth." Geordie identified and prioritized the most relevant risky behavior arising from agents. Beam, Geordie’s remediation engine, can act on those risks alongside the agent itself, escalating from monitoring, to guidance that adjusts how an agent proceeds as it works, to policies that must always hold. Because Beam works through the agent’s own context rather than a gateway, risks can be managed without stopping the work.

Matillion has started enabling Beam across its engineering teams, where agent use is most advanced. There, its policies cover risks such as MCP servers with hard-coded credentials and agents running on non-enterprise accounts. Beam also fits neatly into how the security team already works, with the SOC handling alerts and GRC looking after the policies.

Value Intelligence: Seeing the work behind the spend

Matillion initially engaged Geordie to solve a security problem: the concern of losing track of what people were doing once they handed work to agents. But as agentic work became more and more integral to the business, broader questions of cost and value became more important.

Alongside security, Matillion has seen some of Geordie’s biggest impact on the ROI side. With near-universal adoption, the question has moved from who is using AI to who is getting value from it. In his words, "we're not necessarily wanting to see less usage now, but we're wanting to see more impactful usage."

Geordie's Value Intelligence module gave him the visibility to answer that question, showing where spend goes, which tasks drive it and which ones are good value for money. Matillion’s AI team could tell departments how much they were spending, but not the use cases behind it. "Geordie has given us visibility into those use cases on a department, organization or even individual basis, and tied the cost back to those core pieces of work."

The right model for the job. Knowing what each task was lets the team identify work a cheaper model could have done just as well. In a single month, Geordie identified $15,000 of spend that could have run on more efficient models. With agent activity growing 45-50% a quarter, a gap like that widens fast unless it’s addressed at the source. Beam policies can flag, steer or block the use of models a team doesn’t need, keeping agents on the right models as usage scales.

Build once, reuse everywhere. The same view showed teams building the same thing twice. Presentations are a simple example: a deck built through Claude Cowork in finance and a similar one in security could deliver largely the same outcome at very different costs. Matillion combined those efforts into one centralized skill that carries best practice forward at a more predictable price.