If you run a small business and you're asking what business process should I automate first with AI, you're already ahead of most owners — because the pitch landing in your inbox right now is designed to make you skip that question entirely. It's not a workflow. It's a platform. Finance, HR, procurement, customer service, all automated end to end, one big rollout, one big number.
Salim Ismail argued this week on LinkedIn that the enterprise AI vendor pitch — agent catalogs that run every core process at once, backed by partner deployment funds and nine-figure price tags — is structurally broken. Not overpriced. Structurally wrong. The math looks clean on a slide. It doesn't survive contact with how organizations actually work, how data actually sits in silos, and how change actually happens inside a company. Big vendors are selling certainty about outcomes they can't actually guarantee, because no single deployment can rewire five departments at once and have all five work.
The Six-Figure Suite Pitch Is Now Landing in Small Inbox Sizes
Here's what nobody tells you: that same pitch has been resized for you. You won't get a nine-figure number. You'll get a scaled-down version — a monthly platform fee, a demo showing AI handling your billing, your scheduling, your client intake, and your reporting all in one dashboard. It looks like a bargain compared to the enterprise version. It is still the same structural mistake, just smaller.
The vendor's incentive hasn't changed. They want you locked into the platform before you've proven any single piece of it works inside your business. A 15-person law firm doesn't have a change management team. A dental clinic doesn't have an IT department to troubleshoot when the AI mishandles insurance codes in week two. When the all-in-one system stumbles on one process, you don't get to fix that process in isolation — you're stuck debugging a platform you don't fully understand, while paying for four other modules nobody has touched yet.
Why "Automate Everything" Breaks Down at 15 People
Enterprises can absorb a failed rollout. They have redundant staff, dedicated project managers, and budget to write off a bad quarter. A 15-person company doesn't have that slack. If the AI agent suite mishandles a client's invoice, or double-books a surgery slot, or sends a wrong HR notice, there's no buffer team catching the error before the client notices. The failure is visible immediately, and it's visible to the exact people whose trust you need most.
The other problem is simpler: at 15 people, you don't have five broken processes needing simultaneous fixing. You usually have one. Maybe your intake process eats six hours a week of a partner's time. Maybe your billing cycle slips two weeks every month because nobody chases invoices consistently. Buying a platform built to fix five problems, when you have one real bottleneck, means you're paying for capability you'll never use and complexity you'll never fully control.
Pick One Workflow: Intake, Billing, or Scheduling
This is the actual mirror of Salim Ismail's point. Instead of asking which suite to buy, ask which single workflow, done badly today, costs you the most real hours or real money. For most small businesses it's one of three things.
A real-estate brokerage: lead intake and follow-up. Agents forget to respond to inbound inquiries within the golden hour, and deals go cold. One narrow AI workflow — qualify the lead, schedule the showing, send the follow-up — beats a platform that also tries to automate commission splits and compliance filing.
A boutique hotel: reservation and guest-request scheduling. Front desk staff juggling phone bookings, walk-ins, and housekeeping requests loses coordination during busy weekends. Automating that one scheduling loop pays for itself in fewer double-bookings, long before you touch payroll or vendor procurement.
A small manufacturer: purchase order and inventory reconciliation. If someone spends four hours a week manually matching supplier invoices to received goods, that's the process to fix — not the entire procurement-to-payment chain the vendor wants to sell you.
What Proving One Use Case Actually Looks Like
Proving a use case means three things happen, in order. First, you can measure the before state in plain numbers — hours spent, error rate, dollars lost to delay. Second, the AI tool runs on that one workflow for 60 to 90 days with a real person checking its output weekly, not blindly trusting it. Third, you can show the after numbers went down or output went up, and you can explain why, in one sentence, to someone who wasn't in the room.
If you can't do that third step, you haven't proven anything — you've just turned on a tool and hoped.
The Real Cost of Buying the Big Suite Too Early
The real cost isn't the subscription fee. It's the eighteen months of your team half-trusting a system nobody fully understands, quietly working around it, and you not finding out until the annual review that three of the five modules were never used. It's the cost of training staff on a platform's full interface when they only ever touch one screen. It's the opportunity cost of not fixing the one process that was actually bleeding money, because attention went to configuring modules for processes that were fine as they were.
Small businesses rarely fail an AI project because the technology didn't work. They fail because they bought scope they didn't need before they'd proven scope they did.
How to Expand Once the First Automation Is Actually Working
Once one workflow is proven — real numbers, real months of use, real staff comfort with it — expansion gets much cheaper and much safer. You know what "working" looks like in your own operation, so you can judge the next tool against that bar instead of against a vendor's demo. You also know which internal person actually owns AI oversight now, because someone had to check that first workflow every week for three months. That person becomes your filter for the next pitch that lands in your inbox promising to run everything at once.
The sequence matters more than the ambition. One workflow, proven, before the next one. That's the whole method.
