The "Don't Build This" List: 10 AI Projects That Waste Small-Business Money

Why you need a list like this

Anyone can build with AI now. That's the problem. Cool tools are easy to build, so people build them — and the business doesn't improve. The hard part isn't building. It's knowing what not to build.

Most AI money gets wasted on projects that were wrong on day one. Not because the tech failed. Because the project never should have started.

So here's the list. Ten projects small and mid-sized businesses pour money into, why each one is seductive, why it fails, and what to do instead. If this saves you one bad build, it paid for itself. And it was free.

1. The chatbot on a website nobody visits

Why it seduces: It's the most visible AI project there is. You can show it to your board, your spouse, your golf buddies. "We have AI on our website now."

Why it fails: Math. If your site gets 800 visitors a month and 2% would ever open a chat widget, you built a tool for 16 people. Most of them wanted your phone number. A chatbot doesn't create demand — it deflects questions from traffic you already have. No traffic, nothing to deflect.

Do instead: Look at where questions actually arrive. If your team answers the same 20 questions by phone and email every week, an AI-drafted email response system or a better FAQ page pays off. Fix the channel with volume, not the one that looks good in a screenshot.

2. The "AI everywhere" initiative

Why it seduces: It sounds strategic. "We're rolling AI out across the whole company." Nobody gets fired for having an AI strategy.

Why it fails: AI everywhere means accountability nowhere. Fifty shallow pilots, no owner, no number attached to any of them. Twelve months later you've spent six figures and can't name one metric that moved. The fix is the opposite approach: find the few places where AI actually pays off and build there.

Do instead: Pick one department, one process, one number. Improve it. Then move to the next. Depth beats coverage every single time. And to be clear — "the few places" might be one spot or it might be a dozen. The point isn't scarcity. The point is that each build exists because a specific number needed to change, not because a memo said AI.

3. The cool demo that doesn't change a number

Why it seduces: Demos are fun. An AI that summarizes your meetings, generates images for your deck, answers trivia about your industry — it feels like the future.

Why it fails: Ask one question: "If this works perfectly, which line on the P&L changes?" If nobody can answer, it's a toy. Toys are fine. Just don't put them in the budget as investments.

Do instead: Run every proposed project through that one question before anyone opens a laptop. If the honest answer is "none, but it's neat" — enjoy it as a hobby, not a project. One useful test: write the before-and-after sentence in advance. "Right now, quoting a job takes four hours. After this build, it takes twenty minutes." If you can't write that sentence, you don't have a project yet. You have a demo looking for an excuse.

4. AI for a problem a spreadsheet solves

Why it seduces: AI is the hammer of the moment, so everything looks like a nail. Vendors are happy to encourage this.

Why it fails: You pay AI prices — build cost, maintenance, error handling — for a problem that a pivot table, a shared spreadsheet, or a $30/month off-the-shelf tool handles fine. Some of the highest-payoff fixes in any business aren't AI at all. They're process fixes with a formula in a cell.

Do instead: Exhaust the boring options first. Spreadsheet, checklist, standing meeting, one clear owner. If the boring option works, you just saved five figures — and putting a real analyst inside your Excel shows how far an ordinary spreadsheet plus one good prompt actually gets you. AI earns its place when the problem genuinely needs pattern recognition, language, or scale a spreadsheet can't touch — forecasting demand across dozens of locations, reading thousands of documents, drafting at volume.

5. Replacing judgment calls that need a human

Why it seduces: Judgment calls are expensive. Your best people spend hours on pricing exceptions, hiring decisions, big-customer negotiations. Automating that sounds like the biggest win in the building.

Why it fails: High-stakes, low-frequency decisions with fuzzy context are the worst possible fit for AI. There isn't enough repetition to learn from, the cost of a wrong call is huge, and the context lives in someone's head. You'll either build something nobody trusts or — worse — something people trust and shouldn't.

Do instead: Automate everything around the judgment call. Pull the data, summarize the history, draft the options, flag the outliers. Let AI do the two hours of prep and let your person make the ten-minute decision. That's where the payoff is, and nobody gets burned.

6. Automating a broken process

Why it seduces: The process is painful, so speeding it up feels like relief.

Why it fails: Automate a broken process and you get faster garbage. If your ordering process double-counts inventory, an AI that executes it instantly just double-counts faster. Now the errors show up before anyone can catch them, and you've spent money making the problem worse.

Do instead: Fix the process on paper first. Walk it end to end with the people who run it. Cut the steps that shouldn't exist. Then automate what's left. Half the time, the walkthrough alone fixes it and the automation gets a lot smaller and cheaper. The manual-work inventory is the method for doing that walkthrough properly.

7. Tools built without the person who'll use them

Why it seduces: It's faster. The owner and the developer design it in a conference room, and the team gets it when it's done. Fewer meetings, fewer opinions.

Why it fails: The tool lands and the parts person, the dispatcher, the bookkeeper — whoever was supposed to live in it — quietly goes back to the old way within a month. Not out of stubbornness. Because the tool missed the messy exceptions that make up half their day, and nobody asked.

Do instead: Put the end user in the room from day one. They name the annoyances, they test the drafts, they get veto power on the workflow. Tools shaped by the person who uses them daily stick. Tools designed around that person die, quietly, within a month.

8. The big-bang platform project

Why it seduces: "One system that does everything" sounds efficient. One vendor, one login, one place to call when it breaks.

Why it fails: Big-bang projects take 9–18 months, and by month six the requirements have changed, the champion has left, or the budget got cut. You get nothing until you get everything, and most businesses never get everything. Meanwhile, small targeted builds ship in weeks and start paying rent immediately.

Do instead: Break it up. Build the highest-payoff piece first, get it earning, then build the next piece. Small builds that talk to each other beat one giant build that never finishes. If a piece flops, you're out one small build — not the whole platform.

9. Chasing every new model release

Why it seduces: The headlines are relentless. Every few months a new model drops, and it feels like whatever you built last quarter is obsolete.

Why it fails: Model quality is almost never the reason a business AI project succeeds or fails. The reason is the workflow, the data, and whether anyone uses the thing. Rebuilding to chase releases means you're perpetually migrating and never operating. The gap between models shrinks; the gap between businesses that ship and businesses that tinker grows.

Do instead: Build so the model is a swappable part — which any competent builder does by default — then ignore the news. Upgrade when there's a specific, measured reason: cheaper, faster, or noticeably better at your task. Once or twice a year is plenty.

10. AI content that produces slop

Why it seduces: Content is a grind, and AI makes volume free. Fifty blog posts for the cost of a subscription.

Why it fails: Everyone can smell it now. Generic AI slop with your logo on it tells customers exactly one thing: you don't care enough to write to them yourself. It doesn't rank, it doesn't convert, and it quietly spends down the trust you built over years. Free content that costs you your reputation is expensive content.

Do instead: Use AI to capture and shape your knowledge, not to replace it. Interview your best people — the answers they give customers on the phone every day — and use AI to turn that raw expertise into drafts a human edits and signs. The knowledge is the asset. AI is just the transcriptionist and the editor's first pass. The interview agent is the step-by-step version of exactly that.

The pattern behind all ten

Read the list back and it's one mistake wearing ten costumes: starting from the tool instead of the number.

Every project that wastes money starts with "what can we do with AI?" Every project that pays off starts with "this number is wrong — inventory is too high, quotes take too long, we're paying $100K a year for software we barely use — what's the cheapest reliable way to fix it?" Sometimes the answer is AI. Sometimes it's a spreadsheet. Sometimes it's a process fix and an awkward conversation.

Here's a worksheet you can run on any proposed project before spending a dollar. Copy it, fill it out, and be honest:

• The number: Which specific metric changes if this works? (Revenue, hours, error rate, dollars of spend — pick one.)
• The size: By roughly how much? What's that worth per year?
• The boring alternative: Could a spreadsheet, a process change, or an off-the-shelf tool get 80% of it? What would that cost?
• The user: Who touches this daily, and have they been in the room?
• The process check: Is the underlying process sound, or are we about to automate garbage?
• The kill criteria: What result, by what date, means we stop?

If a project survives that worksheet, build it small, ship it fast, measure it. If it doesn't, congratulations — you just made money by doing nothing, which is the most underrated move in business.

Where Hoven fits

If you'd rather have a second set of eyes on this, most clients start with a $999/day audit — most take one to five days. You get a prioritized plan for where AI actually pays off in your business, and yes, a written "don't build this" list, because half the value of an audit is what you avoid. If it isn't worth more than you paid, you don't pay. Details on the pricing page, or get in touch.