Pythian unveils Gemini Enterprise AI model for staff

Pythian has introduced an internal AI operating model built around Google Cloud's Gemini Enterprise, based on its deployment across 500 staff in 27 countries.

The consultancy developed the framework after its own rollout highlighted why many corporate AI projects stall at the pilot stage or deliver limited returns. In its view, companies often focus too much on buying licences and making tools widely available instead of redesigning broader workflows where the financial impact is greater.

Its model combines strategy, deployment, execution and ongoing operations in a single structure, with four elements: advisory work from its Field CTO team, tooling deployment, a dual centre of excellence, and an XOps layer for monitoring AI systems in production.

The Field CTO team is responsible for governance and identifying use cases before development begins. This process uses 16 agentic patterns, including automated document processing and runbook creation, to produce a prioritised list of projects.

At the deployment stage, Pythian connects AI systems to tools such as customer relationship management platforms, enterprise resource planning systems and database estates. That allows models to draw on internal business context rather than operate in isolation.

Internal test

The dual centre of excellence is split between people productivity and process productivity. The first group focuses on adoption and change management, including building no-code agents for non-technical functions such as human resources and procurement. The second builds more complex coded agents and workflows tied to data platforms.

The XOps component is meant to address what Pythian sees as a common weakness in AI projects after launch. Production systems need ongoing monitoring, prompt tuning and observability because model performance and prompt structures can drift over time.

Pythian linked the model to measurable changes in its own operations, saying active user engagement with Gemini Enterprise and earlier AI tools rose threefold while database incident resolution times fell by 80%.

One internal workflow handles about 15,000 database tickets each month. According to Pythian, the system reads tickets, searches knowledge bases and prepares short runbooks before an engineer starts work.

The company presented this as an example of the difference between small time savings for individuals and larger changes to business processes, arguing that the latter is more likely to produce "million-dollar outcomes".

Customer work

Beyond its own operations, Pythian cited several customer deployments to illustrate the broader model. In one knowledge management project, it said autonomous IT support agents were deployed across 10,000 consultants.

According to Pythian, that installation automated 10% of 20,000 annual IT tickets into "no-touch" resolutions, resulting in savings of more than 1,000,000 operational hours.

In supply chain work, the consultancy said it built custom tools on Gemini Enterprise that reduced forecast-matching cycles from weeks to two or three days across 70 manufacturing sites worldwide. It also pointed to a retail project in which Gemini agentic AI and computer vision were used to automate product onboarding in stores, cutting a manual task that had taken 20 minutes to a process measured in seconds.

Pythian used those examples to support a broader argument that AI projects should be judged on workflow redesign and return on investment rather than on narrow measures such as minutes saved per user. It contrasted what it called a tool-centric approach with an operating model that sets priorities, assigns specialist delivery teams and manages systems after deployment.

Pythian's account also reflects a wider issue in the enterprise AI market, where many organisations have moved quickly to adopt foundation-model tools but are still trying to identify repeatable business value. One obstacle, it argued, is that teams often build custom agents without establishing the operational processes needed to maintain them once they are live.

That concern has grown more prominent as businesses shift from experimentation to production deployment. Questions around governance, model drift, observability and lifecycle management are increasingly central as companies try to integrate AI into core operations rather than keep it at the edge of the business.

Pythian said its own experience prompted a shift away from broad access to AI tools and towards a more structured model aimed at high-value processes. The company said: "By proving this complete model internally first, Pythian drove a 3x surge in active user engagement and cut our database incident resolution times by 80%."