HPE's first AI app started with a Monday meeting

People ask me what they should automate first, and they usually expect a big answer, something with "agent" in the name or a system that runs the whole back office. I want to give them something they can actually ship, so what I tell them is a lot less exciting... open the calendar and find the meeting someone spends two days getting ready for.

HPE started in pretty much that spot. Speaking to CFO Dive, HPE's CFO Marie Myers described the first process they gave Alfred -- the agentic AI tool they built with engineers at Deloitte -- and it was the weekly ops call, which she called "the heartbeat of the company." It brings together 40 to 50 people from finance and sales every Monday, and before Alfred the team would spend days beforehand pulling together about a hundred pages of PowerPoint for it.

I really like that pick. Myers said the idea was to choose one of the most strategic processes in the company, so instead of a side experiment tucked away where nobody would notice if it went sideways, they went for the meeting everyone sees.

Your weekly review is already a workflow

Most teams don't think of a recurring review as a workflow, because it feels like a meeting, so it lives in the calendar and not on a process map. Look at what actually goes into one, though, and it's the same sources every week (usually an export from one system, a report from another and a spreadsheet someone keeps by hand), the same questions asked of them, and the same deadline every Monday morning. That's as clean a workflow as you'll find -- the inputs are known and the output has a fixed shape... and meanwhile someone, often someone fairly senior, is spending a real chunk of their week copying numbers into slides. I don't want that person doing it.

Myers described the old version as a manual, rear-vision mirror look back, and I think that works as a test for your own meetings too. If most of the time goes on reporting what already happened, the reporting part probably doesn't need a person putting it together by hand.

A lot of AI use is quietly heading this way. Anthropic's Economic Index report found that the share of API transcripts tied to office and administrative support tasks grew about 3 percentage points from August to reach 13% in November 2025, and the report reads that as businesses using Claude to automate routine back-office workflows like email management, document processing and scheduling. It's the ordinary work around the real work, which is exactly what a Monday deck is.

So go through every recurring meeting on your team's calendar and ask a few things. Who builds the material, and how long does it take? If it's one person and most of a Friday, you've probably found a good candidate. Which documents and systems does it pull from? If you can write them down as a list, an app can read them too. How much changes week to week? The ideal case is one where the structure stays put and only the numbers move. What happens in the first twenty minutes? If people spend it agreeing on what the numbers are, the prep isn't doing its job. And who needs to see it? A clear audience makes it easy to decide who gets access to whatever you build. None of this needs a strategy offsite -- it's usually an afternoon, and my guess is you'll find two or three meetings that fit.

I'd slow down here though... an app won't save a review nobody reads. If the deck runs to a hundred pages because no one ever agreed on what matters, automating it just gets you a hundred pages faster. So before you build anything, ask the room which ten pages they'd actually miss if they disappeared. Sometimes the honest answer is to shrink the meeting, which is a better outcome than any app... I'd much rather someone cancel a review than spend a month automating one nobody wanted in the first place.

My read is that HPE saw it this way too. Myers framed the change as a chance for finance to help lead a broader transformation of the company, not as a way to make slides faster, so the slides were just the means and a better Monday conversation was the point.

Start small, then earn the bigger jobs

After the ops call, HPE's plan (as CFO Dive reported it) was to take Alfred into the transactional side of finance, from credit and collections to accounts payable and receivable processing. Myers said they were already in production with Deloitte on many of those initiatives, and that forecasting could be the next major focus after that.

That's a much bigger set of processes, more money moving through them and more ways to get things wrong, and I don't think it's an accident that it came second. A weekly review is visible, it repeats, and everyone in the room can tell whether it got better, which is how a team earns the trust it needs before the harder things.

So my advice for a first app is boring on purpose. Pick the thing that happens every week, that someone dreads, and that runs on the same documents every time, build that one first, then see whether the following Monday feels any different to the people sitting in it.

A weekly review built from the same documents every week is exactly the kind of single-workflow app we built Workbricks for, but whatever tool you use, I'd still start there.

Start with Monday.