The proposal always has the same shape. An "AI readiness assessment," a "maturity matrix," a "use-case prioritization workshop," and a roadmap, billed somewhere in the $10,000 to $50,000 range depending on how many logos the firm puts on the cover. What it never contains, anywhere in the deliverables list, is working software. You pay five figures and you receive a deck that tells you what you already suspected: that you should probably be using AI somewhere, that your data is messy, and that you'll need a phased approach. This is the dominant product in AI consulting for small businesses, and it is, with very few exceptions, a deliverable that changes nothing.
The lazy consensus: every small business needs an AI strategy
The narrative being sold to a twelve-person company is the one that was written for a twelve-thousand-person one. It goes: AI is transformative, transformation requires strategy, strategy requires a consultant, therefore you, owner of a regional logistics firm or a dental group or a niche e-commerce brand, need an AI strategy engagement before you touch anything. It sounds responsible. It is mostly a way to sell a Fortune 500 process to a customer who has neither the headcount nor the time horizon to ever execute it.
Here is the thing the consensus gets backwards. A small business does not have an AI knowledge gap. The knowledge is free and abundant — the model vendors publish it, every other newsletter explains it, and the tools themselves are now usable by a non-engineer on a Tuesday afternoon. What a small business has is an adoption gap. The distance between "I understand that an LLM could answer our customer FAQs" and "our customers' FAQs are actually being answered by one in production" is not closed by a roadmap. It's closed by reps and by someone building the thing. A strategy deck is an expensive way to formalize the gap instead of crossing it.
The deck is a real product. It's just built for the wrong customer.
I'm not claiming readiness assessments are a scam in the abstract. For a bank with forty systems of record, three regulators, and a change-control board, mapping where AI can and can't legally touch the workflow is genuine, hard, necessary work. The deliverable there is the analysis, because the execution is someone else's eighteen-month problem. That's the engagement the methodology was designed for, and it's fine.
Drop that same engagement onto a small business and the economics invert. The analysis phase that's a rounding error for an enterprise is most of the budget for a company whose entire annual software spend might be what the consultancy charges for the discovery phase alone. You buy the part you don't need — the diagnosis — and run out of money before the part you do need: the build. A small business that can describe its problem in two sentences over coffee does not need a six-week workshop to discover that problem. It needs the two sentences turned into running code.
What a small business actually needs is one shipped thing
The most useful AI deployment I've ever been part of did not begin with a strategy. It began with an ugly afternoon. At EltexSoft — my roughly 40-person software studio, building since 2015 — the first thing we put into real use was a Slack bot wired to an LLM that answered the team's questions about our own documentation: how we report to a given client, what the process is when a client goes dark, how to upgrade Django on staging. It hallucinated our vacation policy at one point and we added guardrails. But it answered 200 to 300 questions a week, and within a month nobody could imagine working without it. No deck preceded it. No maturity matrix. One person, one afternoon, one workflow that paid for itself immediately.
That is the actual unit of AI value for a small business: one workflow, in production, that someone uses every day. Not a portfolio of prioritized initiatives. One. The script that formats your invoices. The bot that drafts replies to the ten support emails you get every morning. The tool that reads an intake form and pre-fills the quote. The reason these win is that they're small enough to ship before anyone loses interest and concrete enough that you can tell, within a week, whether they work. A strategy engagement produces none of these. It produces a plan to someday produce them.
The test that separates real help from theater
There's a single question that tells you whether an AI engagement is worth a small business's money: at the end, do you own working software, or do you own a PDF? If the answer is a PDF, walk. The good version of this work looks nothing like consulting and everything like building. When we take on a startup or a small company's first product, the deliverable is the running system, the code is full work-for-hire that the client owns outright, and we take no equity. The output is the thing, not advice about the thing. That's the bar. An AI engagement that can't clear it is selling you the wrap, not the gift.
This is also why the much-quoted 2025 finding — that the overwhelming majority of corporate generative-AI pilots showed no measurable return — surprises no one who ships for a living. The pilots didn't fail because the models were bad. They failed because a pilot is structurally a thing that doesn't reach production, and nobody on the engagement owned getting it into the daily workflow. A deck is a pilot you can frame on the wall. The failure was baked in at the proposal stage, the moment the deliverable became a recommendation instead of a deployment.
When paying someone is actually the right call
I'm not going to land this on "so never hire anyone." There's a narrow, real case for outside help, and it's worth naming precisely so you don't confuse it with the deck. You pay for a builder, or a fractional CTO who builds, when the thing you need crosses a line your team can't cross alone: it touches your production database, it has to handle payments or health data correctly, it needs to not fall over when it's wrong, or it has to integrate with the system that actually runs your business. That's engineering, and bad engineering with an AI label on it is just a new way to leak customer data.
The honest version of this is priced like engineering, not like strategy. Real build work runs in the range of a few thousand to low tens of thousands a month for an embedded team, and fractional-CTO oversight — the person who decides what to build, what to buy off the shelf, and what to refuse — sits in roughly the $4,000 to $16,000 a month band, the figure moving with how much hands-on architecture versus light advisory you need. Compare that to a $30,000 deck. One of those leaves you with a running system; the other leaves you with homework.
Refusal is a feature, not a gap in the offering
The quiet tell of good AI help is how much it tells you not to build. We run a quarterly review of new tools internally and reject 60 to 80 percent of what we evaluate — because the adoption cost is too high or because it doesn't beat what we already use. A consultant whose business model depends on you adopting AI everywhere has no incentive to ever say "this part of your business is fine without it, leave it alone." A builder who has to live with the maintenance burden says it constantly. The most valuable sentence in any AI engagement is often "you don't need that," and the strategy-deck model is the one financial arrangement under which nobody will ever say it to you.
Ruthless scope discipline is the actual skill, and it's the opposite of the expansive roadmap. The best small-company work I've watched us do was a stalled build we picked up after a previous team failed to deliver — and the first week was spent deconstructing scope and re-estimating it something like ten times, cutting thirty to fifty percent each pass until what was left was the smallest thing that could actually ship. That's the discipline a small business is paying for, if it's paying for anything: someone who shrinks the problem until it's buildable, not someone who inflates it until it's billable.
If you're evaluating the space broadly, what AI consulting really is maps it. Our AI engineering services show what production AI work looks like when engineers run it.
The position, stated plainly
Do not buy AI consulting for your small business if the deliverable is advice. The strategy deck, the readiness score, the prioritized roadmap — these are enterprise instruments sold to you because the methodology is easy to repackage and the margins are excellent. You don't have the gap they diagnose. Either start yourself — open the tool, give it a messy prompt, get a mediocre answer, and build one ugly thing that solves a problem you actually have — or pay an engineer to ship a single working system you own at the end. Both of those produce something real. The third option, the one with the maturity matrix on slide nine, produces a slightly better-informed version of exactly where you started, and an invoice. Spend the money on the build, or spend an afternoon and build it yourself. Don't spend it on the map.