If you run a mid-market company and your board, your peers, or your own gut has told you it's time to do something serious about AI, you are now sitting in front of a question that nobody trained you for. Do you hire someone? Do you retain a consultant? Do you put an agency on a monthly retainer? Do you skip humans entirely and just buy a stack of tools? The answer is almost never one of those four in isolation, and the right blend depends more on your company stage than on the technology itself. The framework I use with clients is straightforward. Match the staffing model to the maturity of your AI use case, not to the hype cycle. Spend the smallest amount of money that gets you to the next decision, not the biggest amount you can defend.
The four staffing models, ranked by cost and time-to-value
There are exactly four ways to add AI capacity to a company. Every other arrangement you have heard pitched is a combination of these four. Here is how they sort out on a 12-month spend basis.
Tools-only, no humans. You buy ChatGPT Team or ChatGPT Enterprise, Claude for Work, a Microsoft Copilot license per seat, maybe Notion AI, maybe a Zapier or Make subscription with AI steps wired in. Total 12-month cost lands between $5,000 for a small team on basic plans and $50,000 for a 50-seat company on enterprise tiers with a few specialty subscriptions. Time to value is measured in weeks. The ceiling is low because nobody on staff is paid to think about AI, so the tools end up being used the way the most curious person on each team happens to use them. Some of that is fine. Most of it leaves money on the table.
Consulting, project-based. A specialist or a small firm engages on a defined scope. Audit, strategy, vendor selection, build, training, handoff. Total 12-month cost lands between $25,000 for a single targeted engagement and $500,000 for a full transformation roadmap with multiple builds. Time to value is typically 90 days from kickoff. The ceiling is medium-to-high because a good consultant can compress two years of trial and error into a quarter, but the consultant leaves and you keep what was transferred. Best for one-time setups, vendor evaluation, training your existing people, and getting unstuck when an internal effort has stalled.
Agency, monthly retainer. An ongoing relationship with a firm that runs your AI-driven content, automation, or operations as a managed service. Typical retainers range from $5,000 to $25,000 per month, putting 12-month spend between $60,000 and $300,000. Time to value is ongoing rather than a milestone. The ceiling is medium. An agency is great at executing a defined playbook every month. They are not usually great at rethinking your business model. Best for content production, social presence, marketing automation, and anything where the work is repeatable and the value comes from consistency rather than novelty.
In-house hire. A full-time employee carrying the AI title. Could be a Director of AI, a Head of AI, an AI Engineer, an AI Product Manager, or an AI Operations Lead. Loaded cost in 2026 lands between $180,000 and $350,000 base salary, plus another 25-35 percent in benefits, equity, and overhead. Time to value is 6 to 12 months by the time you scope the role, run the search, hire, onboard, and let them ship something meaningful. The ceiling is the highest of the four. An in-house owner can build deep institutional understanding and deliver compounding value year over year. Best when AI is becoming a core part of your product or your operating model, not when AI is a project you want to run.
The matrix: which model fits which company stage
Here is the rough mapping I walk clients through.
Under $5M revenue. Tools-only plus a single consulting engagement to set up the stack and train the team. Total spend should not exceed $30,000 in year one. Hiring an AI engineer at this stage is almost always a mistake. You do not have enough surface area to keep them productive.
$5M to $25M revenue. Tools-only plus an agency retainer for the highest-volume function (usually marketing or customer support) plus a consultant on call for strategy and vendor selection. Annual spend lands between $80,000 and $200,000. This is the sweet spot for the hybrid pattern I describe below. A full-time AI hire still does not pay back here unless AI is core to the product itself.
$25M to $100M revenue. Now an in-house hire starts to make sense, but it should be an AI translator or an AI operations lead, not an engineer. Pair them with the agency retainer and a quarterly consulting check-in. Annual spend lands between $300,000 and $600,000 fully loaded.
$100M+ revenue. A small in-house team of two to four people, an agency for content scale, and a consulting firm for strategy. The in-house team starts to include actual engineers. Annual spend lands between $1M and $3M.
The single biggest mistake I see is mid-market CEOs benchmarking against $100M+ companies and concluding they need a Director of AI on payroll. The economics rarely work below $25M.
Real 2026 salary data and the illusion of the head-of-AI hire
Let me give you real numbers, because the salary inflation in this space has been disorienting for everyone trying to budget honestly. These are 2026 base salaries before benefits and equity.
AI Engineer, L4 (mid-level, 3-5 years experience). San Francisco and New York: $190,000 to $240,000 base. Major secondary metros (Boston, Seattle, Austin, LA): $160,000 to $210,000 base. Tertiary markets and remote: $130,000 to $180,000 base. Add 20-40 percent in equity at a venture-backed company, less at a private mid-market firm.
AI Engineer, L5 (senior, 5-8 years). SF/NY: $260,000 to $340,000 base. Secondary metros: $220,000 to $290,000 base. Tertiary and remote: $180,000 to $240,000 base.
Staff AI Engineer or ML Lead, L6/L7 (8+ years, FAANG-trained). SF/NY: $350,000 to $500,000 base, with total comp regularly exceeding $700,000 at top labs. Secondary metros: $280,000 to $400,000 base. Most mid-market companies cannot compete here and should not try.
Director of AI / Head of AI. $250,000 to $400,000 base in major markets, $200,000 to $320,000 in secondary markets. The variance is enormous because the title means radically different things at different companies. Some Heads of AI are individual contributors with a fancy title. Some manage teams of 20 and own a P&L.
AI Product Manager. $180,000 to $280,000 base, scaling with the technical depth required. A PM who can read model evaluations and write specs that engineers respect commands the high end. A PM who is essentially a translator commands the low end.
Here is the part that nobody wants to say out loud. If you are a $20M company and you hire a Director of AI at $300,000 base, that hire needs to generate roughly $1.5M in incremental value over their first 18 months just to clear the hurdle of being a sensible spend. Most do not, because the company has not yet defined the problem clearly enough for a senior person to attack it. The hire ends up running pilots, attending vendor demos, and writing strategy documents nobody reads. By month nine they are restless. By month fifteen they are gone.
The Head-of-AI hire pays back when one of three things is true. AI is a core feature of your product and competitors are shipping faster than you. You have a specific revenue line item (say, a $5M+ services business) that AI is going to eat or transform. Or you have already proven AI ROI through agency and consulting work and now need an internal owner to scale what's working. If none of those apply, defer the hire.
Honest disclosure: where consulting wins, and where it doesn't
I run a consulting firm. So you should treat this section with the appropriate amount of skepticism. I'll try to be straight about it anyway.
Consulting genuinely wins in five situations. First, one-off setup. Configuring an AI stack the right way the first time saves a year of false starts. Second, vendor evaluation. There are 200 AI tools claiming to do the same thing in any given category, and a consultant who has used 30 of them in the last 12 months can save you from a six-figure mistake. Third, training your team. A consultant teaches once and leaves. An agency keeps doing the work and your team never builds the muscle. Fourth, executive alignment. When the CEO, CFO, and head of operations disagree on what AI should be used for, an outside voice with no political stake can break the tie faster than another internal meeting. Fifth, getting unstuck. When an internal AI effort has been spinning for six months without shipping, a consultant can usually diagnose it in a week.
Consulting does not win for ongoing operational work. If you need a thousand pieces of AI-assisted content per quarter, a consultant is the wrong tool. You want an agency or in-house production. Consulting also does not win when the work is highly proprietary and confidential, because consultants by definition rotate across clients and you may not want your hardest problems being solved by someone whose next engagement is your competitor.
The honest test is this. If the work is bounded, episodic, strategic, or pedagogical, hire a consultant. If the work is repetitive, ongoing, and operational, do not.
The hybrid pattern that works for most mid-market companies
The configuration I see succeed most often at companies between $10M and $75M in revenue is a three-part stack. None of these parts in isolation is enough. Together they cover the gaps.
Part one: an agency retainer for content and automation. The agency runs the production line. Blog posts, social posts, email sequences, lead routing, customer support deflection, internal SOP automation. Whatever your highest-volume repetitive function is, the agency makes it cheaper and faster. Budget $8,000 to $15,000 per month.
Part two: a consultant on a quarterly cadence. Not a fractional CTO who shows up every Wednesday. A real consultant who comes in for a defined engagement once a quarter. Q1 is an audit and strategy. Q2 is a build. Q3 is a training rollout. Q4 is a vendor refresh and roadmap update. Budget $40,000 to $100,000 per year. The consultant's job is to make sure the agency is not just running the playbook from 18 months ago and to keep your strategy ahead of where the tooling is heading.
Part three: one in-house person who is not an engineer. This is the move most mid-market companies miss. You do not need an AI engineer. You need an AI translator. Someone whose full-time job is to identify which workflows in your company should be automated, brief the agency on what to build, evaluate what the consultant proposes, and train the rest of the team to actually use the tools. This person comes from your operations, marketing, or finance org, not from a computer science program. They are typically a high-performing operator with 5-10 years of experience who is curious about AI and willing to spend their nights tinkering. Pay them $120,000 to $180,000 base, give them a clear charter, and protect their calendar from getting eaten by every department head wanting an AI demo.
That stack runs you between $250,000 and $500,000 per year all-in, which is roughly the loaded cost of one mid-level AI engineer in San Francisco. The difference is that this stack actually delivers compounding value, because each part covers what the others cannot.
The three hiring mistakes mid-market makes
If you only remember three things from this paper, remember these.
Mistake one: hiring engineers when you need operators. The first AI hire at a mid-market company should almost never be an engineer. It should be the AI translator I described above. You will be tempted to hire an engineer because that is what tech companies do, and because engineers are easier to interview (you can give them a coding test). Resist. An engineer with no defined problems to solve will build elegant infrastructure for use cases nobody asked for. An operator with a coding habit will look at your invoicing process, see that 40 percent of the work could be automated, and ship something useful in two weeks.
Mistake two: hiring before defining the problem. If you cannot write three sentences describing what success looks like for the AI hire 18 months after they start, do not hire yet. Spend $30,000 on consulting first to define the problem. The cost of a bad senior hire at this level is roughly $500,000 once you account for salary, opportunity cost, and the morale damage of a high-profile role that did not work out. Spending $30,000 to avoid that risk is one of the easiest decisions you will make this year.
Mistake three: hiring too senior, too early. The Head-of-AI title is intoxicating. Boards love hearing it. Recruiters love filling it. But a Head of AI without a team to lead and without a clearly defined mandate is a $300,000 individual contributor with a frustrated calendar. If you are going to make an in-house investment, start with the operator-translator at $150,000. Let them prove the business case for AI internally. Then, if and only if the volume of work justifies it, hire above them in year two or three.
Build vs buy vs rent: a decision tree
Every AI capability you might want falls into one of three buckets. Build it yourself, buy a tool, or rent the work as a service. Here is how to decide.
Buy when: the capability is commoditized, the leading tool has more than 1,000 customers, the integration cost is under 10 percent of the annual license, and your needs are within 80 percent of what the tool already does. Examples in 2026: meeting transcription, basic content generation, customer support chat, sales prospecting, document summarization. Do not build any of these. The category is settled and someone else is going to spend $50M on R&D you would never recover.
Rent when: the capability is repeatable but requires expertise to set up, your volume is high, and the work is not strategically differentiating. Examples: AI-driven content production, marketing automation builds, custom GPT configuration for your sales team, AI training programs for your staff. Hire an agency. Their margin is your discount versus building the same capacity in-house.
Build when: the capability is core to your product or your competitive advantage, no off-the-shelf tool comes within 50 percent of what you need, the data you would feed it is proprietary and gives you a real edge, and you have or can hire the engineering depth to maintain it for years. Examples: a domain-specific knowledge agent for a specialized industry, AI features inside your own SaaS product, an automation that touches data your vendors cannot legally access. Most mid-market companies have one or two of these in their entire business. Maybe.
The trap is building things you should buy. The opposite trap (buying things you should build) is rarer at the mid-market level because the strategic depth required to build well is usually missing. Default to buy and rent. Build only when you can articulate why an off-the-shelf solution actively damages your competitive position.
The decision framework by company stage
To pull all of this together, here is the decision I would walk a mid-market CEO through, by stage.
If you have not started. Buy ChatGPT Team or Claude for Work for everyone, run a 30-day usage audit, and book a single consulting engagement to identify your top three use cases. Total spend year one: $15,000 to $40,000. Do not hire anyone yet.
If you have tools but no traction. Add an agency retainer for your highest-volume function. Layer in a quarterly consultant. Total spend year one: $100,000 to $200,000. Still do not hire yet.
If the agency is delivering and you can see the next 18 months of work. Hire your in-house translator at $130,000 to $170,000. Keep the agency. Reduce consulting to twice a year. Total spend year one: $300,000 to $450,000.
If AI is now generating measurable revenue or saving measurable cost in the seven figures. Now you can start to build a small in-house team. Add an AI engineer. Maybe an AI product manager. Keep the agency for content and automation. Use consulting for the hard strategic questions. Total spend year one: $700,000 to $1.5M.
Most mid-market companies should be in stage two or three for the next 18 to 24 months. The temptation to skip ahead to stage four is what gets people in trouble. Do the cheap experiments before the expensive ones.
Where to start this quarter
If you are at the beginning of this decision, the highest-leverage thing you can do in the next 60 days is define the problem before you commit any capital. Run a workflow audit across your top three departments. List the five most repetitive, time-consuming tasks in each. Estimate the hours and the dollar cost. That single document will tell you whether you need an agency, a consultant, a hire, or just better tools, and it will save you from spending six figures on the wrong move.
If you want help running that audit, I do them on a fixed-fee basis with a written deliverable inside three weeks. I will tell you honestly whether you need to hire me again after that or whether you should retain an agency, make a hire, or do nothing for another quarter. I am a consultant, but I would rather lose the next engagement than waste your money on it. Book a scoping call and we will figure out which of the four staffing models fits your stage.