Most guides to choosing an AI consultant for a small business are written by AI consultants, rank themselves first, and do not tell you what anything costs. This one is also written by an AI consultancy, which you should factor in, so it is built the other way round: costs first with sources, then the criteria, then where different kinds of firm actually fit. OpSpring appears once, near the end, in the category we genuinely belong to.
What AI consulting actually costs a small business
This is the question buyers ask first and most published guides answer last, if at all. The figures below are the most granular public breakdown we found, from Layer3Labs:
| Engagement | Typical range |
|---|---|
| Freelance AI developer, hourly | $100 to $200 |
| Boutique firm of 5 to 20 people, hourly | $150 to $300 |
| Enterprise consultancy, hourly | $300 to $600 |
| Strategy assessment | $5,000 to $12,000 |
| Proof of concept | $8,000 to $20,000 |
| Full implementation | $15,000 to $75,000 |
| Monthly advisory retainer | $2,000 to $8,000 |
By individual workflow, again from the same source: support triage $12,000 to $28,000, document extraction $15,000 to $35,000, lead intake and CRM $10,000 to $25,000.
Other published estimates differ, and the spread is worth seeing rather than smoothing over. AI Essentials puts consultants at $150 to $350 an hour or $5,000 to $25,000 per project. The AI Consulting Network starts its range at $2,000.
Do not average these. The spread is not noise, it is the answer: the same request costs very different amounts depending on one variable that almost no pricing page mentions.
The variable that actually moves the number
Whether the software you already run exposes a usable API.
Automating document intake for a business on a modern platform with an open API is a fraction of the work of the identical automation on a system with no integration path. That single difference is usually the largest line in a quote, and it is why a firm that gives you a number before looking at your systems is guessing. The guess is revised upward later.
Four other things move it, in rough order of weight: how many workflows are in scope, how much human judgement the system has to replicate, your monthly volume for the ongoing fee, and whether regulated data is involved.
The mismatch nobody in these lists mentions
Search "best AI consulting firms for small businesses" and you will meet the same names repeatedly. Several of them are, by their own published client lists, enterprise vendors.
LeewayHertz names Siemens, Budweiser and Pearson. Addepto names Porsche, Hertz and MSG. IDEO is a global design firm. These are genuinely capable companies and none of that is a criticism of their work.
But a team whose delivery model is built around enterprise programmes will tend to scope, staff and price your project as a smaller version of one. You get a discovery phase, a deck, a statement of work and a team you meet on a weekly call. For a fifteen-person business trying to stop rekeying invoices, that shape is expensive relative to the outcome.
Read a firm's client list before you read its ranking. It tells you more about fit than any roundup position, including this one.
The criteria, stated before the list
Any guide that ranks firms without publishing its criteria first is asking you to trust the order. Here is ours, and you should disagree with it where your situation differs.
- Do they scope before quoting? A fixed price against a defined scope means the number does not drift while they work. Hourly billing with no cap moves all the risk of scope growth onto you.
- Do they name a specific workflow? "AI for your operations" cannot be built, tested or judged. "Cut policy comparison from three hours to twenty minutes per account" can.
- Do they quote the ongoing cost? Every AI system has a running cost: hosting, model and API usage, monitoring, and keeping it correct as your process changes. A quote covering only the build describes half the commitment.
- Can they answer what happens when the AI is wrong? This is the question that separates people who have shipped AI systems from people who have demoed them. "The model is very accurate" is not an answer. A human review step, a checkable citation back to the source, and an audit trail are answers.
- Do they own the outcome after launch? If the engagement ends at delivery, the risk of the thing quietly going unused sits entirely with you.
- Will they talk you out of work? Bring three ideas. If you hear enthusiasm for all three, you are being sold to rather than assessed.
Which kind of firm fits which job
| If your situation is | The right shape is | What it usually costs | Watch out for |
|---|---|---|---|
| One clear workflow eating hours weekly | Boutique firm or specialist, fixed scope | $10,000 to $35,000 build, plus monthly | Scope growing mid-build |
| You are not sure AI is the answer yet | Paid assessment, or a free audit | $0 to $12,000 | Assessments that only ever conclude "yes, hire us" |
| Standard workflow, modern single system | No-code tools, possibly nobody | Tool subscription only | Paying a firm to configure Zapier |
| Regulated data, healthcare or insurance | Specialist in that vertical | Higher, and correctly so | Generalists underquoting compliance work |
| Multi-department programme, 200+ staff | Enterprise consultancy | $75,000 and up | This guide is not aimed at you |
| You need a product built, not a process fixed | Product studio or dev shop | Varies widely | Buying a build when you needed a decision |
The most common expensive mistake is the third row: hiring a firm for something an off-the-shelf integration already does. The second most common is the last: commissioning a build before anyone has decided what should be built.
When you should not hire anyone
The useful half of an assessment is the part that says no. Four cases where automation is not worth it yet:
- Volume is too low. If a task happens twice a month, the build will not pay back regardless of how annoying it is.
- The process changes constantly. Anything built against a process that shifts quarterly is stale before it earns out.
- There is no way into the system. Where the workaround costs more than the problem, wait for the vendor's API rather than paying someone to fight the interface.
- Nobody agrees what the process is. Automating an undefined process makes it faster and no clearer. Write it down first, by hand. That exercise alone frequently removes the need for the project.
Where we fit
OpSpring is a boutique firm, which puts us in the first row of that table and firmly outside the last two. We work with small and mid-sized businesses on specific workflows, and our deepest vertical is independent P&C insurance, where we build policy comparison and coverage gap analysis and run our own product with paying agencies.
Being straight about the limits: we are two people plus contractors. If you need a multi-department transformation programme across a 500-person organisation, one of the enterprise firms above is a better call than us. If your workflow is standard and lives in one modern system, you may need a Zapier subscription rather than a consultancy, and we will say so.
Our AI audit is free and produces a written plan whether or not you work with us, including the workflows we think are not worth automating yet. How we structure and price engagements is published here.
How to run the evaluation
Talk to three firms. Give each the same specific workflow rather than a general brief. Then compare four things:
- Who narrowed the scope, and who took the brief as given.
- Who asked about your systems before pricing anything.
- Who quoted the ongoing cost without being asked twice.
- Who told you something you did not want to hear.
The last one correlates with delivery more reliably than any case study, because a firm willing to lose the sale by being accurate is a firm that will be accurate once you have paid them.
Key Takeaways
- Published costs range from $100 an hour freelance to $600 enterprise, and $5,000 for a strategy assessment to $75,000 for a full implementation. The spread is real, not noise.
- The largest single cost variable is whether your existing systems expose a usable API. A quote given before anyone looked at your stack will be revised upward.
- Several firms topping best-for-small-business lists publish enterprise client lists. Read the client list, not the ranking.
- Insist on a specific workflow, a fixed scope, the ongoing monthly cost, and an answer to what happens when the AI is wrong.
- A firm that talks you out of some of your ideas is doing the job you hired it for.
- Sometimes the right answer is a no-code tool, or nothing at all. Low volume, a constantly changing process, no integration path, or an undefined process are all reasons to wait.