Pictet turns weeks of work into hours with Claude Code

Incremental gains at a bank that bets on automation

When generative AI arrived, Pictet moved quickly, building an internal chat assistant for the whole group while central teams automated pockets of back-office work. But the limits soon showed. While central teams built with AI, everyone else was left using chat tools for Q&A, and coding assistants gave engineers only limited lift. End users couldn't build anything complex on their own.

Some back-office work sat out of reach entirely: a relatively simple task such as checking a large number of internal directives against the latest regulations across jurisdictions was sized at two weeks for a team of two to three, and the tools in place couldn't handle it.

The gap wasn't abstract for a firm like Pictet, which focuses on a limited number of high net worth clients rather than millions of retail accounts. "We try to be very high-touch at the front, but as automated as we can at the back, so that we can provide the best service possible," said Steve Blanchet, Head of Group Technology Strategy and Innovation at Pictet. "It's a bit different than many banks out there."

Agentic AI that fit the bank's risk framework

Pictet's bet was on agentic AI: models that carry out multi-turn tasks end to end. Claude Code brought that capability to engineers, and Claude Cowork extends it to colleagues outside engineering. "We wanted to equip our developers with the best tools available, because we think it's a big area of opportunity," Blanchet said. "Agentic AI is perceived as a big accelerator, whether for coding or for business process automation."

By then the group already had a training partner in place: Artefact, a data and AI consultancy it first brought in for generative AI expertise before its AI Center of Excellence existed. "We wanted to have specific expertise around generative AI," Blanchet said.

Getting Claude approved at a Swiss private bank took close work with the risk, information security, and data protection teams, a relationship the group had been building for three years. The group had a clear AI policy when adopting Claude, which was built onto its existing risk framework. It helped that Pictet's AI use was deliberately internal-facing, aimed at improving operational processes rather than client-facing products.

The policy also sets clear boundaries. Business use cases must run in the EU, and for some only in Switzerland. Data can never persist for model training, so that, in Blanchet's words, "the data just stays with us."

The rollout followed the same logic that eased the risk conversation in the first place. "What also helps managing the relationship with risk and compliance is this rollout per persona and per kind of use case," said Xavier Meyer, Head of Cloud Engineering at Pictet. It started with developers, where residency constraints are loosest.

A hackathon as the springboard

The developer rollout opened with a Claude Code hackathon, co-organized with Artefact. Every team built a working prototype with real code behind it, a step past the design concepts similar events had produced in earlier years. The winning tool, PicAccess, gave employees changing roles or joining a new team a self-service view of application access: which apps and roles their colleagues have, what they're missing for their own role, and a pre-filled IT request to close the gap. "Employee access rights are something that we haven't been able to crack for quite some time, but now we are tackling it with Claude Code," Blanchet noted.

Turning a launch event into a working capability fell to Artefact. The consultancy built a half-day workshop for Pictet's own developers, ranging from newcomers to agentic coding to engineers already using Claude Code personally, and delivered it 25 times to groups of around 20 on Pictet's internal training platform; the first six to eight weeks booked out as soon as the schedule went live. Sessions started with agentic fundamentals, what an agent is and how the loop inside a tool like Claude Code works, then moved through skills, sub-agents, and Model Context Protocol (MCP) integrations into hands-on use. More than 500 people went through it. "Adoption and enablement is often how you see value actually materialize," said Zachary Schillaci from Artefact. "Not just taking the license, but taking the training and putting it in the hands of users."

Work that wasn't feasible now takes hours

Around 700 people across the group now have access to Claude Code and Claude Cowork: mostly engineers, plus product and business colleagues. "We are using Claude Code and Claude Cowork for problems where our other tools fall short," Blanchet said. One example came from compliance. One standout application emerged from compliance, where a complex gap analysis—comparing over 50 internal directives against current regulatory standards—was completed in just hours using Claude Code. This specific initiative had been originally projected as a two-week undertaking for a dedicated three-person team.

Outside of compliance, product managers now generate working prototypes in minutes instead of the one to two weeks a traditional cycle took. One infrastructure team built a centralized alerting system in hours rather than weeks, while another quickly delivered a customized UI for thin clients, the locked-down terminals used across the bank. A separate team successfully implemented a disaster recovery feature with Claude Code where other AI tools had failed. Another team fully automated its weekly marketwatch newsletter on major AI developments, a task that previously took three hours each week. "I'm surprised by the sheer range of problems that can be tackled," Blanchet said of Claude Code. "It is not only about developing code and building apps, but also about cracking tough problems."

Claude also reaches into the business's own systems, so its rollout moves more carefully. Some product managers went first, working with early MCP connections to GitHub, Jira, Figma, and the internal wiki. It also covers Pictet Alternative Advisors, the group's alternatives arm, where heavy Excel and PowerPoint users report the strongest results. Asset management, where demand runs high, is queued next. The group's investment teams aren't waiting for their turn: some have begun using Claude Code to manage machine learning pipelines for portfolio asset allocation. "Our investment teams are especially eager to use this force multiplier," Blanchet added. The next phase is accompanying business teams as they reinvent their processes, which requires staying close to how they work day to day.

His advice to other firms is the rollout's own lesson: "Broad impact doesn't just magically happen,” Blanchet said. “It's not only about rolling out a tool: it is about building capabilities and bringing people along."