The most common question I get on a first sales call isn't what does it do. It's what does it actually cost. Operators have been quoted $5,000 by a freelance prompt engineer, $50,000 by a Big Four consultancy, and $500,000 by a system integrator — for what looks like the same Statement of Work. They want to know what's real.
Here's an honest breakdown. I'm going to ignore the prompt-engineer quote (that's a Loom video, not an install) and the system-integrator quote (that's procurement theater). The interesting band is the middle — what production AI implementation actually costs for a real operator at a real organization.
The four real cost buckets
Every AI install has four cost buckets. Most quotes name two and hope you forget about the other two.
1. Vendor subscriptions
The obvious one. OpenAI, Anthropic, a vector database, an orchestration layer, a CRM that the AI writes to. For a 25-person organization running production AI on real workflows, this is usually $1,500–$3,500 per month, depending on usage. Annualized: $18K–$42K a year. Most teams forget that they need a model and a model gateway and a vector store and a way to log what happened, and they're back at the Big Four's $200K number because nobody told them the stack has four line items.
2. Engineering time
The biggest hidden cost. A "two-week prototype" doesn't ship to production. It gets killed when the founder sees the error rate, or it limps along forever and nobody trusts it. Real production installs take 6–10 weeks for the first workflow, plus another 2–4 weeks each for additional ones. If you have a senior engineer on staff who can do it, that's $30K–$80K of their time you redirected from something else. If you don't, that's $50K–$150K to bring in an external team that knows the failure modes.
3. Integration debt
Nobody quotes this one. Your AI has to read from somewhere and write to somewhere — that means Gmail, Calendar, Slack, your CRM, your DMS, your billing system, whatever. Each integration has its own auth, rate limit, edge cases, and "we changed our API in a quiet email last Tuesday" moments. A meaningful install touches 5–10 systems. Each one is 1–3 days of work to do right (with audit logging, retries, idempotency, the works). Cumulatively: another $20K–$60K nobody put in the quote.
4. Outcome eval + handoff
This is the cost most installs skip — and it's why most installs fail. You need a way to know whether the AI is actually moving the metric you wanted moved. That's measurement infrastructure (event tracking, before/after windows, holdout groups), it's a quarterly review cadence, and it's a handoff package so the operator who inherits this in six months understands what was decided and why. Done right: $10K–$25K. Done wrong: $0 — and the install gets quietly replaced when someone new joins the team.
The honest total
Add it up: $75K–$250K to ship one meaningful AI install to production at a real organization, the first time. The next install is cheaper because the integration debt and outcome-eval scaffolding carries over. But the first one — the one that proves AI works in your operations — is between $75K and $250K of total ownership cost in year one.
That's where our pricing comes from. Our engagement bands are $75K, $150K, and $250K. Not because the labor costs us $75K — it doesn't — but because that's the floor at which we can ship the four buckets above and have anything left to fund the Institute (10% of every engagement).
What about the $99/month plan?
Real question, separate answer. Our Pro tier at $99/month handles bucket #1 (vendor subscriptions) and gives you a ready-to-use interface for the workflows that don't need custom integration. It's the right product if you're a solo operator or small team and your workflows are mostly chat + RAG + scheduling. It is not a substitute for an engagement. Operators who try to scale Pro into an organizational layer run into the integration-debt wall by week four.
The freelancer-and-prompt-engineer route
Worth naming directly: a $5K freelancer can write good prompts and hand you a working notebook. That's real value if your workflow is simple and you only need it to work for one person. The freelancer doesn't handle bucket #3 (integration debt) or bucket #4 (outcome eval). Operators come to us when the freelancer's notebook can't scale past one user, or when it breaks the first time a system on the integration path changes.
The Big-Four-and-SI route
Worth naming too: a $200K–$500K SI engagement gets you a polished deck, a project manager, three offshore engineers, and a 9-month timeline. You might get to production — but you'll pay for the theater. Operators come to us when they've watched the SI engagement burn $300K and ship nothing they can demo at the next board meeting.
How to spend less than $75K
Honestly: don't hire anyone yet. Run a discovery first. We sell Atlas Discovery at $5K–$15K — two weeks, no PowerPoint, written deliverable: which workflows would actually move your metric, ranked by leverage, with an outcome-eval scope and a contract framework your CFO can sign. If the discovery says "you don't need an engagement," we'll tell you that. We'd rather refer you out than sell you something you don't need.
The bottom line
Production AI at an organizational layer costs $75K–$250K in year one, regardless of who you hire. The variable is who pockets the margin: us, an SI, a freelancer, or your existing team's time. Pick the route that matches how you want to spend the next 90 days.
Talk to us if you want a numbers-backed proposal. Or just subscribe and read the next dispatch when it lands.