How Anthropic's Finance Team thinks about Claude and Workday

Anthropic's own finance team ran a webinar in 2026 called "How finance teams use Claude Cowork", and the session is worth a closer look for one specific reason: the presenters provide detailed and live demos on how they migrated their own ERP platform from NetSuite to Workday Financials. That is particularly relevant for all organizations that are already live with Workday Financials, described by the finance org that just went through it.

Two practitioners presented: Lisa, about a year into the finance team at the time, and Tim, the finance AI product lead, about seven months in. Tim ran two live demos. Lisa walked through what the finance systems side has been doing with the same tooling. This was more than just marketing, it is a glimpse of how generally available tools can help streamline financial process. Both are worth unpacking against a Workday Billing close.

The reconciliation demo, and what actually made it work

Tim's first demo was a month-end staple: reconcile the AR subledger against the GL. He dragged in two Excel files, an open invoice extract and a GL AR aging detail, and asked Claude Cowork to find what did not tie. No data connector, just two spreadsheets.

The result: the two sides differed by a certain amount. Cowork built an actual variance bridge, not just a number: two invoices sitting in the GL but missing from billing, three in billing but missing from the GL, and one with an amount mismatch. That is the same bridge structure any Workday Billing admin builds by hand during a close.

What made this reliable rather than a confident guess is the part finance people should pay attention to. Anthropic's team had written a skill ahead of time, a plain-language document describing the two data sources, the key columns, and the exact steps of the reconciliation. The skill is what turned a general-purpose model into something that reconciled the way their team actually does it. Tim was direct about this earlier in the session: finance teams are right to push back on LLMs making things up, and the answer is not blind trust, it is skills that encode the procedure, connectors that pull real data, and a human who reviews the output before anything counts.

The Workday Financials migration and master data governance

Lisa's section is the one that maps most directly onto Workday Billing work. Her team had just moved off NetSuite onto Workday Financials, which meant reconciling and validating years of historical data on a compressed timeline. She was careful to note this specific piece was not run inside Cowork itself. Instead, they built a purpose-built application on the same underlying approach: define a skill for the reconciliation and validation logic, then let it run against source and target files, either one at a time or in a batch.

Reopening a prior period used to mean hundreds of hours reconciling every line, every month, every year. With the tool built around this approach, a year of data now takes about 20 seconds, with as much manual spot-validation layered on top as a team wants for comfort. That is not a claim about eliminating the close. It is a claim about compressing the mechanical part of it so the time goes to review instead of data wrangling.

The second use case is master data governance, and it is one every Workday Billing admin will recognize by the pain it describes. Changes like a new cost center or a renamed account typically move through a ticketing system to satisfy ITGC change management controls. The evidence trail lives in the ticket, but the ticket is often a bad form filled out by a busy finance user who just wants the change made. Her team replaced that intake with a chat-based request process where Claude asks the follow-up questions needed to capture what the approval workflow actually requires, then routes it.

Questions to ask before you try this on your own close

• Where does the skill come from. Both demos worked because a domain expert, not an engineer, wrote the procedure down first. A tool with no skill behind it is back to a confident guess
• What is the human still checking. Cowork's own five-step process ends in verification and delivery, not automatic posting. Decide up front what a reviewer signs off on before anything counts
• Is the data synthetic or real. The webinar demos ran on fake data by design. Ask specifically how the same workflow performs against your actual customer master and your actual GL
• What happens to the audit trail. ITGC controls need evidence, not just a faster outcome. Confirm the chat history or skill log satisfies what your auditors already ask for

The takeaway

The detail that separates this from most AI-and-finance content is that a finance team described its own reconciliation and its own Workday Financials migration, not a demo built to look good on stage. The mechanism behind both, a written skill plus a connected data source plus a human check, is the same pattern any Workday Billing team would need to trust this kind of work on their own close.

Many AI tools are being made avaiable to finance teams and connections to ERP and other systems are becoming available. But just like the problems accountants face when Excel sheets become too complex and impossible to maintain, the same is true for AI tools. Before critical processes are being passed on to Agents, its creations need to consider the scalability, security, and compliance aspects.

If you are looking at where this pattern fits inside your own Workday Billing close, let's talk.

More information on Claude Cowork