Ask Your Data a Question: Forecasting, Margin, and AI for DME Operations

Most DME owners already have the data to answer their biggest operating questions, what rents out the least, whether hospital beds are actually worth owning, when they'll run out of wheelchairs. The data just lives in three different places (a scan history, an accounting system, and whatever's in someone's head), and pulling it together usually means a spreadsheet nobody has time to build.

What inventory forecasting actually gives you

A forecast isn't a prediction about demand you don't know yet. It's a projection built from commitments you've already made: units scheduled to go out, units already expected back, run forward against what's sitting in the warehouse today. If fourteen wheelchairs are already promised over the next three weeks and only nine are on hand, that's not a guess, it's arithmetic, and it's worth knowing before a new referral calls and there's nothing left to send.

The only thing standing between "we found out too late" and "we saw it coming two weeks out" is whether your checkout and return dates are actually captured as part of the normal workflow, not filled in after the fact when someone remembers.

Average rental duration and utilization tell you where the money is

Once you can see how long each equipment type actually stays out, on average, two things become obvious that usually aren't: which categories turn over fast enough to serve a lot of patients from a small fleet, and which categories tie up capital for months at a time. Combined with utilization rate, the share of a fleet actually out earning money at any given point, you get a real answer to "should we buy more of this or stop replacing it when it wears out," instead of a hunch.

None of this is exotic. It's the same math a good operations person has always done by hand for their busiest categories. The difference is doing it for every category, continuously, instead of only when something goes visibly wrong.

Then there's just asking

The newer piece is Model Context Protocol, usually shortened to MCP, a standard that lets an AI assistant like Claude connect directly to a business's own software and use its real data in a conversation. For a DME operation, that means you can ask an assistant something like "what's our margin on hospital beds this quarter" or "what's about to run low" and get an answer pulled from your actual numbers, not a generic explanation of how DME margins usually work.

A connection like this should be scoped the same way any login is: authorized once by an admin, limited to that one business's data, revocable at any time. It should also be conservative by design. A well-built assistant connection can look things up and make a narrow set of specific edits, add a product, register a unit, adjust a rate, but it should never be able to delete a customer, an order, or anything else on its own. If you ask it to do something like that, the right behavior is to tell you to go do it in the app yourself, not to guess at what you meant.

The combination is the point

Forecasting, duration, utilization, and margin are useful on their own. What changes the day-to-day is not having to go build a report to see any of them, and increasingly, not even having to open a dashboard, you can just ask. The data was always there. The work was always making it visible without someone having to reconstruct it by hand every time a question came up.