Cuban’s right. The AI bill is for the workflow you never rebuilt.

We get more insight from Mark Cuban than we care to admit to anyone.

Generally, he comes across as abrasive and arrogant and minority stake or not, he still does own a big chunk of the team that let Luka Donic walk to the Los Angeles Lakers. The guy is worth $6 billion plus though, so foolish it would be to just dismiss what he has to say because you don’t like the style that comes across on television.

All that said, we were intrigued by something he said on the Big Technology Podcast a couple weeks ago.

Alex Kantrowitz asked him why giant public companies keep lighting cash on fire trying to “do AI” without seeing returns. Cuban’s answer was as clean a diagnosis we’ve heard from anyone with a checkbook that size.

“We hear a lot of stories about huge public companies that try, they spend a lot of money to try to implement AI, and it doesn’t really get them a return on investment, and people say, ‘Well, why if AI is so great?’ Because you already are running your business the way you’ve always run it. And in order to take advantage of AI, to answer your question, you have to reformulate your business completely to build it on AI. It’s the difference between running a business before PCs and computers, and running a business after PCs and computers. You would do it completely differently.” — Mark Cuban, Big Technology Podcast, May 2026

Fortune 500 procurement teams are buying AI like a copier. Sign the MSA, run a pilot in one department, measure the productivity bump, expand the footprint. That works for copiers. It doesn’t work for a technology whose value comes from rebuilding the workflow it touches.

Cuban’s PC analogy lands. Companies in 1985 didn’t get value from PCs by handing one to every accountant and asking them to type faster. Companies got value when the close process (ledgers, approvals, inter-office mail, printed reports) got rebuilt around the assumption that a spreadsheet now exists. Companies that didn’t rebuild paid for the PCs and kept doing manual reconciliation on top of them. A decade later they were gone, acquired, or left servicing your great-grandparents in an office that looked like it belonged in a black-and-white movie until they were gone and acquired. Either way, they ended up in the same place.

That’s where the big-company AI ROI problem comes from. A bolted-on pilot can only overperform the workflow underneath it so much.

Two SMB workflows that prove the point

A 28-person regional HVAC company

The controller runs a four-day month-end close. Three of those days are reconciling field-tech timesheets against dispatched jobs, invoiced amounts, and parts pulled from the truck inventory system. None of those systems talk to each other. She lives in Excel and a printed dispatch log.

The wrong way to add AI: buy a “smart” bookkeeping plugin for QuickBooks, watch it suggest categorizations she has to verify anyway, and call it AI adoption. She still spends three days reconciling.

The right way: replace the reconciliation workflow itself. An agent pulls the dispatch records, timesheet entries, parts log, and invoiced jobs every night. It produces a one-page exception report by 6 AM. These 11 jobs don’t tie out, here’s the likely cause, here’s the suggested fix. The controller now spends 30 minutes a day during the month instead of three days at the end of it.

Conservative back-of-napkin: controller’s fully-loaded cost is $95K. She used to spend 240 hours a year on reconciliation. Now it’s 60. That’s 180 hours back, about $8K of labor. Plus the close shifts from four days to one, so the owner sees real financials on the 3rd of the month instead of the 8th. Most owners care more about that than the labor savings.

A 45-person specialty contractor

The project manager spends six hours a week reading subcontractor emails. Schedule changes, RFIs, material delays, photo updates. He’s not doing project management during those six hours. He’s doing inbox triage so he can do project management on Friday.

Bolt-on version: ChatGPT browser extension. He pastes emails in, asks for summaries, gets summaries, still has to act on them. Same six hours, slightly better summaries.

Rebuilt version: every subcontractor email flows through an inbox agent that classifies the message (schedule change, RFI, delay, FYI), extracts the structured fields (which job, which trade, what date, what dollar amount), updates the project tracker automatically, and only pings the PM when something needs a human decision. The PM gets a morning briefing: three things he actually has to look at today. The other 90% of inbound never crosses his desk.

Conservative numbers on a $12M revenue specialty contractor: 250 hours recovered. At a $115K fully-loaded cost, that’s about $14K. Bigger number: projects he’s actually managing finish on schedule 8% more often. On a 14% gross margin business, that schedule discipline beats the labor savings by an order of magnitude.

Why SMBs have an edge big companies don’t

Cuban frames this as a big-company problem. It is. The corollary he doesn’t say: small and mid-sized businesses are uniquely positioned to do what he’s prescribing.

A 30-person HVAC company has one controller, one dispatcher, and one owner. To rebuild the reconciliation workflow you need three conversations and a Saturday afternoon. To rebuild it at a Fortune 500 you need a steering committee, two consulting firms, an SAP integration, and 18 months. Then the workflow you rebuilt is obsolete because the team that wrote the RFP rotated out.

At a big company, the legacy drag is the org chart. Nobody owns the whole workflow, so nobody can rebuild the whole workflow. The pilots stay siloed because the org is siloed.

SMBs have one or two people who actually understand how the work flows. The smaller the company, the closer the rebuild is to a single coherent decision.

This isn’t that much different than other insights we’ve shared: “You can just install Claude Code tonight,” “Why process beats raw capability,” and “You can just decide to be AI native.” The Cuban point is a really clear distillation of the same idea, and the idea is an opportunity waiting-to-happen for executives, founders, and entrepreneurs with the guts to do it.

Three quick application ideas

• Don’t buy AI. Rebuild a workflow. If you can describe what you bought as a feature, you bought a feature. If you can describe what changed as “we don’t do that step anymore,” you rebuilt a workflow.
• Pick a workflow with a number on it. Hours, dollars, days of close, jobs per month. If you can’t put a number on the current state, you can’t put a number on the rebuild, and you’ll have the same “we spent money on AI and can’t tell if it worked” problem the F500 has.
• Start where the legacy drag is smallest. Reconciliation, inbox triage, scheduling, intake. Not pricing, not customer service, not anything that touches a regulated process. Get the muscle memory for rebuilding before you go after the load-bearing wall.

The part Cuban didn’t have to say

The companies that did well after PCs asked which process, if rebuilt around the assumption of a spreadsheet, would compound. Buying the most PCs wasn’t the answer.

Same question, new tool. The companies who answer it first own the next decade of their industry. The ones who answer it last get bought by the ones who answered it first.