The Wrong Anchors
Most year-one AI budgets are anchored on one of two wrong references: the enterprise consulting proposal ($500k+) or the “ChatGPT for the team” number ($2k). Neither maps to real AI deployment for an SMB or mid-market company. The honest range for a 20–200 person business: $30k–$250k fully loaded for year one.
By Company Size
- Under 10 people: $5k–$30k. Tools and a part-time builder.
- 10–30 people: $30k–$80k. Tools, a project, internal owner part-time.
- 30–100 people: $80k–$200k. Tools, 2–3 projects, internal owner full-time, possibly an external partner.
- 100–500 people: $200k–$1M. Tools, a portfolio of projects, a small team, governance.
The Line Items
- AI provider costs. OpenAI / Anthropic / etc.
- SaaS AI tools. Per-seat or per-use.
- Integration tools. Zapier, Make, n8n.
- Data infrastructure. Postgres, vector DB, monitoring.
- Internal talent. Owner, builder, possibly data person.
- External partners. Consultancies, agencies, specialists.
- Training. Workshops, documentation, change mgmt.
- Governance. Policies, audits, legal review.
- • 40% — talent (internal and external).
- • 30% — tooling and infrastructure.
- • 20% — AI provider costs.
- • 10% — training, governance, miscellany.
Build vs Buy Math
Buy decisions cost 30–50% of equivalent build, but lock you into vendor terms. Build decisions cost more up-front and need maintenance. For most year-one use cases, buy. Build only when the use case is core to differentiation.
Talent Costs
The internal AI owner: typically a senior product or ops person, often at 0.5–1.0 FTE for year one. Fully loaded $80k–$200k depending on role and geography. The single biggest line item for most companies.
Tooling Costs
Modern stack for a 50-person company: $1.5k–$3k/month for the AI-adjacent tools. Add $0.5k–$2k/month in direct AI provider spend depending on volume.
Hidden Costs
- Data cleanup before AI deployment (often 30%+ of project cost).
- Change management and training time.
- Productivity dip during transition.
- Legal and compliance review.
- The second tool you need after the first one falls short.
The honest year-one AI budget for a company that takes this seriously: 0.5–2% of revenue. Less than that and you're half-pregnant. More than that and you're likely over-engineering.
See hidden costs of AI implementation.
FAQ
What if we have no budget? Start with $5k of tooling and a clear champion. Prove ROI, then ask for more.
Year two? Usually 1.5–2x year one as you scale wins.
How to defend the budget to the board? Tie every dollar to a measurable outcome.