Most AI budgets you’ll see are theoretical or vendor-flavored. Here’s the actual money math for a $20M B2B service business in our priority industries, based on what we’ve seen work across hundreds of projects.
The starting assumptions
A typical $20M B2B service business in our priority industries looks like this:
- 75–150 employees
- $2–4M EBITDA
- One owner/CEO, one COO or operations director, one CFO or finance leader
- One or two senior people whose judgment is irreplaceable (an estimator, a lead engineer, a head of compliance)
- A small marketing/sales team (3–6 people)
- Field, service, or delivery teams that scale linearly with revenue
For this profile, a realistic AI Year 1 budget breaks into three components.
Component 1: The partner ($30K–$120K/year)
Not optional. If you’re trying to build AI capability without a senior outside partner, your timeline is 18 months instead of 12, and the failure rate is 3–4x higher. That’s not a scare stat — according to RAND, more than 80% of AI projects fail, twice the rate of IT projects that don’t involve AI. The thing that moves that number is senior execution, not more software.
- Sherpa ($7,500 per 3-month cycle): lightest touch. A guided program that gets your leadership team fluent and scopes the first build. No engineering — that starts with AI Office.
- Operator ($60K/year): strategist plus dedicated engineering capacity. One major build per quarter, or multiple lightweight builds. This is the most common starting tier.
- Embedded ($120K/year): standing senior team. Multi-stream build queue. Right for PE-backed operators or multi-entity buyers.
For a $20M service business, plan on ~$60K for the retainer in year 1 unless you have a specific reason to go bigger.
Component 2: Larger builds (expect 2–4 in year 1)
Some builds are too large to fit inside the retainer’s bandwidth. They flow out of the AI Office relationship and get scoped after a Deep Dive — never self-selected up front.
A typical year for a $20M service business:
- Build 1 (months 3–4): quoting acceleration, document processing, or field-to-office reporting. ~$25K–$50K.
- Build 2 (months 6–7): knowledge capture, internal knowledge search, or a sales/service copilot. ~$25K–$50K.
- Build 3 (months 9–10, if needed): a more complex integration. ~$50K–$80K.
Year 1 spend on larger builds: $75K–$150K depending on appetite.
Each build is tied to one agreed business metric, baselined before the build and reported against after it ships.
Component 3: Tools, infrastructure, internal time ($15K–$40K)
The smaller bucket, but still real:
- Frontier model API costs: ~$200–$2,000/month at production scale for a typical mid-market workflow set. As workflows scale this can grow, but it stays modest.
- AI tool licenses your team uses individually (Claude for Work, ChatGPT Enterprise, Copilot, and the like): $50–$500/user/month depending on usage. For a 100-person company with 30 active AI users, ~$50K/year is in range.
- Infrastructure (hosting, monitoring, deployment): typically $5K–$20K/year for early-stage deployments. Larger as you scale.
- Internal team time: 10–20% of one director’s time plus 5–10% of several operational owners. This is real cost — call it $30K–$60K of loaded labor allocated to AI work in year 1.
Putting it together
A defensible Year 1 budget for a $20M service business:
| Component | Year 1 spend |
|---|---|
| AI Office retainer (Operator tier) | $60,000 |
| 2–3 larger builds | $75,000–$150,000 |
| Tools + licenses + infrastructure | $20,000–$60,000 |
| Internal team time (loaded labor) | $30,000–$60,000 |
| Total Year 1 | $185,000–$330,000 |
Cash out the door, excluding internal labor allocation: $155,000–$270,000.
The ROI math that justifies it
For this profile of business, realistic year-1 outcomes from a well-run AI program:
- Quoting acceleration: an 8–15 point win-rate improvement on $5M of inbound bid volume = $400K–$750K in additional bookings.
- Field/office time recovered: one FTE-equivalent saved at ~$80K loaded cost.
- Senior estimator/expert leverage: 5–10 hours/week recovered for the senior person. At a ~$150K loaded rate, that’s $40K–$80K of expert capacity recovered.
- Customer service automation: one FTE-equivalent of routine call handling at ~$60K loaded.
- Indirect: faster decisions, better visibility, faster M&A integration capability, retention improvement.
A well-run program is designed to pay for itself inside year one — measured against the KPI agreed for each build, not asserted.
When the budget doesn’t work
Two situations where this math doesn’t apply:
You’re significantly under $20M revenue. The retainer pricing is the same, but the relative scale of the investment is bigger. Smaller businesses should start lighter — the free and low-cost on-ramps, or Sherpa ($7,500 per 3-month cycle) — to prove the partnership before scaling up.
You’re significantly over $20M revenue. The retainer scales — you’d typically be at Operator or Embedded. The build volume scales too: year 1 might be 4–6 larger builds, not 2–3. The total budget can run $400K–$800K, still a fraction of what a national consultancy or internal team would cost.
Want this run against your numbers?
If you want a budget walked through your specific business — your size, your industry, your current AI maturity — that’s a 30-minute conversation. Book a 30-minute intro, or take the AI Readiness Assessment first.