Build vs. Buy vs. Embed. The real cost of an AI team for a $10M to $50M business.

We get asked the cost question on every discovery call. The honest answer takes ten minutes and a piece of paper. Here is the version with the paper.

Option one: build the team in-house

The textbook AI team for a mid-market company has five seats:

• Senior Applied AI engineer: $300,000 to $500,000 fully loaded
• ML engineer: $250,000 to $400,000
• Data engineer: $220,000 to $350,000
• MLOps / platform engineer: $250,000 to $380,000
• Product manager (AI experience): $180,000 to $280,000

Round numbers: $1.2M on the low end, $1.9M on the high end, before equity refresh, recruiting fees, training, hardware, vendor licenses, or any of the other line items HR will surface in month two.

A more realistic mid-market reality is three of those five seats, half of which you will hire below the senior tier because the senior people are not on the market. So your effective spend is $750K to $1.2M for a team that is one senior short of the team you actually want. The senior is at Anthropic. Or Stripe. Or fishing.

Total exposure for a real in-house build: $1.5M to $3M annually. Three years to break even on the recruiting and training cost. Two years before the senior person is shipping at full output, assuming they stay.

For a $20M business, this is not a hire. It is a strategy bet. The bet might be right. It is not the kind of bet you place if AI is one of three priorities.

Option two: buy a consulting engagement

The consulting market for Applied AI sorts into three rough tiers:

• Big 4 / tier-one (McKinsey QuantumBlack, BCG X, Accenture, Deloitte): $500K to $5M for six-to-twelve-month engagements. Strategy decks, proofs of concept, partner SI handoff.
• Mid-tier consultancies (Slalom, Pariveda, dozens of regionals): $200K to $1M for three-to-six-month engagements. More implementation, less strategy.
• Boutique AI shops: $50K to $500K for six-to-twelve-week MVPs. Builds and ships, then disappears.

All three tiers share a structural problem. The engagement is shaped like a project, not an operation. The AI ships, the team rotates off, and your operating environment is supposed to take over. In practice, the model needs retraining in month four, the prompts need adjustment in month six, the integration breaks in month eight, and the firm you paid is not on call.

The common workaround is a "managed services retainer" tacked on after the build. Read the fine print. The retainer team is offshore, junior, and not the people who built the thing.

Total exposure for a typical consulting engagement: $300K to $2M for the build, plus an annual retainer that is either too small to actually maintain the system or too big to justify.

Option three: embed a senior fractional team

The embed model is what Mozr does. The structural difference is that the senior engineer who scoped the work is the senior engineer who ships the work and is still the senior engineer two years later. There is no rotation, no handoff, no junior bench.

The pricing has three tiers, scaled to the work ahead:

• Single Project: one focused build, tightly scoped. Right when you know the workflow you want solved first.
• Three Projects: a portfolio of builds, sequential or parallel. Right when AI touches more than one workflow. Most common tier.
• Enterprise: custom scope for teams running AI as a permanent operating function. Multiple Mozr engineers embedded, on-call for production, co-owned roadmap.

Pricing is a flat monthly fee, scoped on the discovery call, no annual lock-in. The relationship is month-to-month with thirty days notice on either side. The first month is the trial.

Total exposure for an embed engagement at the typical Three Projects tier: a flat monthly fee for as long as the work is shipping value. No annual contract. No retainer that does not get used. The line item is recoverable inside the first workflow we replace.

The decision frame that actually matters

When founders compare line items, the in-house team always looks expensive and the consulting engagement always looks scoped. That framing misses the actual question.

The question is: how much can the engagement cost you in month nine, when the model needs retraining and the original team has moved on?

For an in-house build, month-nine cost is a senior engineer on payroll who is now working on something else because the AI is shipping. That is a real win.

For a consulting engagement, month-nine cost is a Slack message to a partner who is on a different client, a $40K change-order, a four-week wait, and a junior engineer who has never seen your codebase doing the actual work.

For an embed engagement, month-nine is a standing call on Tuesday with the same engineer who built the thing, looking at the same dashboard, shipping the next iteration on Friday.

That is what you are buying. The line item is downstream of that.

When each option is actually right

Build in-house when AI is becoming a defining product surface and you have $3M of operating budget that is not at risk if the senior hire takes eighteen months to land. Most $5M to $50M businesses do not.

Buy a consulting engagement when you need a board-level strategy artifact and a one-time build that runs for two years without needing to evolve. Rare in Applied AI. Most production AI needs to evolve continuously.

Embed when you want a senior practitioner shipping production AI inside your operation on a flat fee, with the relationship structured so the senior person is incentivized to stay. This fits most $5M to $50M founder-led businesses.

The honest part

Mozr is not the cheapest option on a single-month basis. A junior boutique with a fixed-price MVP is cheaper. The reason we charge what we charge is that the senior engineer is not subsidized by junior bench rates. Every hour of a Mozr engagement is a senior practitioner doing senior work.

And we are not the cheapest option on a three-year horizon if you can actually pull off a real in-house team. A team you fully own, that stays, with the senior person you actually wanted, will compound faster than any embed relationship. We tell founders who can do this to do it.

What we are is the right shape for the gap between those two cliffs.