Pictet
Pictet gives about 700 staff Claude Code and Cowork, cutting a compliance review from weeks to hours
Pictet, a Swiss private bank, rolled out Anthropic's Claude Code and Claude Cowork to about 700 people, starting with its technology division. In Anthropic's account, a check of over 50 internal directives against current regulation took hours instead of the two weeks planned for a small team.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
around 700
People with access to Claude Code and Claude Cowork to date1
more than 500
People trained through hands-on Claude Code workshops run with Artefact1
over 50, completed in just hours
Internal directives checked against current regulatory standards in one compliance gap analysis (originally projected at two weeks for a three-person team)1
two hours instead of two weeks
Time to a working prototype (the story contrasts hours with a traditional cycle of weeks)1
The problem
Pictet is a 221-year-old Swiss private bank that serves a small number of very large clients. Its stated approach is a very personal service at the front and as much automation as possible behind the scenes. After an internal chat assistant for the whole group, most staff were still limited to question-and-answer tools, and coding assistants gave engineers only a small lift. Some back-office work sat out of reach. Checking a large set of internal directives against the latest regulations in several countries was sized at two weeks for a team of two to three, and existing tools could not do it.
What they built
Early in 2026 Pictet began giving staff two Anthropic tools. Claude Code is a coding tool that works on its own across many steps, reading code, editing files and running commands. Claude Cowork extends similar abilities to colleagues who are not engineers. Access started with the 1,500-person technology division and now covers about 700 people in engineering, product and business teams.
How they did it
Winning approval meant working closely with the bank's risk, security and data-protection functions. The bank wrote an AI policy on top of its risk framework. Business uses must run in the EU, and some only in Switzerland, with a purpose-built gateway controlling access. The policy also says data must never be kept to train models. Use was kept internal, aimed at operations rather than client-facing products.
Rollout went by type of user. Developers went first because residency limits are loosest for them. A hackathon with consultancy Artefact came next, where every team had to build a working prototype. Artefact then ran a half-day workshop 25 times for groups of around 20, and more than 500 people attended. Some product managers also got early connections to tools such as GitHub, Jira, Figma and the internal wiki.
Results
According to Anthropic's account, the standout result came from compliance. A gap analysis comparing over 50 internal directives with current regulation was finished in hours, where the plan had been two weeks for a dedicated three-person team. Product managers produce working prototypes in minutes or hours instead of one to two weeks. An infrastructure team built a central alerting system in hours rather than weeks, and another team automated a weekly newsletter that used to take three hours.
The story also reports a team that added a disaster recovery feature after other AI tools had failed. Asset management and investment teams are next in line, and some investment teams already use Claude Code to manage machine learning pipelines.
Lessons for your project
Involve risk, information security and data protection teams early, and write an AI policy on top of your existing risk framework before asking for tool approval.
Roll out by group of users and type of use. Start where data residency rules are loosest, such as developers, then move to teams that touch sensitive business data.
Pay for training, not only licences. Pictet ran 25 hands-on workshops so people learned how to use the tool on real work.
Run a hackathon where every team must produce a working prototype, to turn curiosity into concrete use cases.
Look for back-office tasks that were never worth staffing, such as cross-checking many policies against regulation, and test the tool there.
Caveats
- The only source is Anthropic's own customer story. No independent outlet confirms the figures, and Anthropic has a commercial interest in presenting them well.
- The story is undated, so the publication date is unknown. The rollout start is given only as the beginning of 2026.
- The story names the products, Claude Code and Claude Cowork, but not the underlying model version.
- The compliance result compares actual time (hours) with a projection (two weeks for three people), not with a measured earlier run. The story also states the prototype speed-up differently in two places (two hours versus two weeks, and minutes versus one to two weeks).
- Business data processing is limited to the EU and Switzerland, and for some uses only Switzerland, through a purpose-built gateway. Results may not transfer to firms with looser residency rules.
Sources
- Pictet turns weeks of work into hours with Claude Code
Anthropic · Undated · Vendor case study · claude.com
Summarized in our own words from the sources above. Spotted an error? See corrections.