The Report Used to Take a Week. Now It Takes a Conversation.

Every operations leader has a Friday report.

The one someone exports, pastes into a spreadsheet, cleans up, and reformats before a Monday meeting. It started as a favor. Then it became a habit. Now it's load-bearing, and the person who built it is the only one who fully understands it.

Last time, I wrote about what's newly possible: general-purpose AI tools can now build the infrastructure side of operational problems in a single conversation. This is issue two of three. It goes deeper on the part most teams underestimate, the data infrastructure itself: the dashboards, the Excel builds, and the point where a spreadsheet quietly becomes a database.

A quick grounding fact. In MGMA's 2026 Regulatory Burden Report, 52 percent of respondents were in practices with 20 or fewer physicians, and 60 percent were in independent practices. These are the organizations least likely to have a dedicated analyst, and most likely to be running critical reporting out of a spreadsheet somebody maintains between other tasks. The reporting need is real. The staffing to meet it usually isn't.

That gap is what's changing.

The hardest part of building a dashboard was never the formulas. It was starting.

You know what you want to see. Denial rate by payer. Authorizations still pending. Average days to decision. AR by aging bucket. What you didn't have was the time to sit down and structure it from nothing.

AI is good at exactly that first draft. Not connecting to your systems. Not pulling live data. Structuring the thing.

Describe the dashboard you want and ask for the shell:

"Build me an Excel dashboard structure to track prior authorization performance. Include the raw-data columns I'd need, a summary section with approval rate, denial rate, average days-to-decision, and AR by aging bucket, and tell me which chart type fits each metric. I'll populate it with my own exports."

What comes back isn't finished. You rename columns to match your data, adjust the calculations, and connect it to your exports. But the blank page, the part that kept the project on a someday list for a year, is gone. You're editing a draft instead of staring at nothing.

Most operational spreadsheets are databases wearing a costume. Rows of records, columns of fields, relationships between tabs. The work that makes them painful is the moving and cleaning: three monthly exports that don't line up, duplicate claim numbers, denial reasons written five different ways.

That's the work AI handles well now, and it's worth being specific about how well. In an independent May 2026 test of five AI tools across 30 real Excel and Google Sheets problems, spanning array formulas, pivot logic, and Power Query M code, Claude scored highest at 76 out of 90, ahead of Gemini, ChatGPT, and Microsoft's own Copilot. It scored 9 out of 10 on the hardest category, array formulas with LAMBDA functions. The point isn't the leaderboard. It's that the code these tools generate for data cleanup and consolidation is now good enough to start from.

You describe the data. It writes the Power Query M code. You paste that into Excel under Data, Get Data, Launch Power Query Editor. No programming background required:

"Write Power Query M code that consolidates three monthly denial report tabs (each with columns for claim number, date of service, CPT code, payer, denial reason, and amount) into one table. Remove duplicates on claim number, and categorize each denial reason as eligibility, authorization, coding, timely filing, or other."

This is also arriving inside the tool your team already uses. Microsoft made Agent Mode generally available in Excel in April 2026, added a =COPILOT() function that generates, classifies, and summarizes data directly in a cell, and by its June 2026 Build conference had made Agent Mode the default in Microsoft 365 Copilot. You don't have to adopt a new platform to see this. It's showing up in Excel itself.

There's a moment every operational spreadsheet reaches. Too many tabs. Too many people editing. A formula that breaks every time someone adds a column. The spreadsheet outgrew the job, but rebuilding it properly always required someone technical, so it limped along.

AI narrows that gap too. It can help you design the structure of a simple database, the tables, the fields, the relationships, before you ever ask IT to build one. Not to replace your IT team, but to walk into that conversation with a clear specification instead of a vague request:

"I'm tracking prior authorizations in a spreadsheet that's become hard to maintain. Help me design a simple relational structure for this: what tables I'd need, what fields belong in each, and how they relate. I want to hand this to IT as a starting spec."

The value isn't that AI builds the database. It's that the operations leader who understands the workflow can now describe it in the language the build requires. The person closest to the problem gets to shape the solution.

Three things hold true no matter how capable the tools get.

Don't put patient data in. General-purpose AI tools are not HIPAA-covered by default. Work with de-identified data, column headers, and structure-only descriptions. PHI requires an enterprise-tier tool with a signed Business Associate Agreement.

The output is a draft. Every formula, every dashboard shell, every database spec needs review by someone who knows your environment. AI doesn't know your EHR's export format or how your team categorizes denials. That context lives with you.

And decide who owns it. This is the quieter risk. The reason that Friday spreadsheet became fragile wasn't the spreadsheet. It was that no one was ever assigned to own it. Building the next tool faster doesn't fix that. If anything, building faster makes ownership matter more, because you'll build more.

For years, the constraint on operational reporting wasn't ideas. Every leader has a list of things they'd build if they could. The constraint was the distance between knowing what you need and having a starting point to work from.

That distance just got a lot shorter. A dashboard that needed an analyst and a budget line is now something you can draft the structure of in an afternoon. It still needs your data, your review, and your judgment about who maintains it. But the starting point exists, and for a lean operation without a bench of analysts, that's the part that was always missing.

Issue three closes the series with the revenue cycle intelligence side: coding lookups, denial pattern research, and prior authorization documentation.

Which report on your team is one person's spreadsheet that the whole department quietly depends on?

If the blank page weren't the barrier, what's the first piece of operational visibility you'd build?

And when you build it, who owns it?

Sources: MGMA 2026 Regulatory Burden Report; Talkory AI, "Best AI for Excel Formulas 2026" (May 2026 test of 5 models across 30 tasks); Microsoft Tech Community, "Agent Mode in Excel is now generally available on desktop" (2026).