AI automation ROI

Most AI automation ROI claims you read online are either fabricated or cherry-picked. We are an agency that ships this work for a living, and we are going to give you something different — a model you can plug your real numbers into, ranges by vertical that we have observed across our client base, and a clean account of when this work pays back and when it does not. The numbers are illustrative, not guaranteed. Your business will land somewhere in the ranges, not on a single point.

The basic ROI formula

The simplest honest version:

ROI = (Annual Benefit − Annual Cost) / Annual Cost

Stated as a multiple: a 3x ROI means the project returned three times its cost over the year. ROI of 0 means break-even. Negative ROI means the work cost more than it returned, which happens more often than vendors admit.

What goes into "annual benefit"

• Labor saved. Hours per week × hourly cost (loaded — salary plus benefits plus overhead, usually 1.3x base) × 52. Be conservative; only count hours actually redeployed to other useful work.
• Revenue gained. Lead-to-meeting rate lift, meeting-to-close lift, average order value lift, retention lift. Each tied back to a dollar number.
• Cost avoided. Headcount you did not have to hire, contractor work you did not have to commission, overage you did not pay.
• Quality lift, where dollarizable. Fewer customer-service escalations, fewer compliance incidents, fewer broken handoffs. Translate to dollars or leave out.
• Speed lift, where dollarizable. Faster response times tied to conversion rates; faster cycle times tied to inventory turn or sales velocity.

What does not count:

• "Freed up time" that nobody can show was redeployed to a higher use.
• Strategic optionality with no concrete dollar pathway.
• Brand benefits, employee morale, and other qualitative goods. They are real, but they belong in a different column, not in the ROI calc.

What goes into "annual cost"

• Build cost. Agency project fee, internal engineering time, scoping work. Amortize over expected useful life — usually 24 months.
• Operate cost. Monthly retainer × 12, or in-house ops time at fully loaded rate.
• External API spend. LLMs, scrapers, enrichment, vector DB, observability. Often 20–60% on top of agency fee.
• Internal management overhead. Hours your team spends reviewing, escalating, approving. Easy to under-count; budget at minimum 2–4 hours per week per system.
• Failure cost. The expected value of mistakes the system will make. For low-stakes workflows, near zero. For revenue or compliance-critical workflows, can be the dominant line item.

Sample math by vertical (illustrative ranges)

These are observed ranges across our client base, with outliers trimmed. Treat them as a sanity check, not a guarantee.

Real estate brokerage (10–50 agents, dormant pipeline reactivation)

• Annual cost: $30,000 (agency) + $8,000 (API + tooling) = $38,000
• Annual benefit: 1 FTE coordinator avoided ($55K loaded) + 3–8 incremental closings × ~$8K commission contribution = $79,000–$119,000
• Year-one ROI: ~1.1x to 2.1x
• Year-two ROI: typically 2x–4x because build cost is amortized away

Solo coach / consultant ($150–$400/hr billable rate)

• Annual cost: $9,000 (agency) + $2,500 (API + tooling) = $11,500
• Annual benefit: 4 hours/week of admin redirected to billable work × 50 weeks × $200 effective rate + content production saved $9,000/yr = ~$49,000
• Year-one ROI: ~3.3x
• Risk: only real if those reclaimed hours are actually billed

E-commerce ($1–5M GMV)

• Annual cost: $48,000 (agency) + $12,000 (API + tooling) = $60,000
• Annual benefit: 1 support FTE avoided ($55K) + 1 copywriter freelance avoided ($24K) + ~1.5% conversion rate lift on $2M revenue ($30K margin contribution) = ~$109,000
• Year-one ROI: ~0.8x (basically breakeven year one)
• Year-two-plus ROI: 2x–4x because build investment amortizes

B2B SaaS (Series A/B, outbound + lifecycle)

• Annual cost: $120,000 (agency) + $36,000 (API + tooling) = $156,000
• Annual benefit: 1.5 SDR FTE avoided ($150K loaded) + lifecycle marketer half-time ($55K) + 25 incremental SQLs/mo × $5K MRR LTV contribution = ~$355,000+
• Year-one ROI: ~1.3x to 2.3x
• Range widens dramatically based on close rates and ACV

Law firm or accounting practice (10–40 staff)

• Annual cost: $36,000 (agency) + $9,000 (API + tooling) = $45,000
• Annual benefit: paralegal / admin time saved ($45K) + faster client onboarding tied to retention lift ($30K margin) = ~$75,000
• Year-one ROI: ~0.7x to 1.5x
• Watch out: regulatory review can add $10K–$30K to year one

When ROI goes negative

• Volume is too low. Below ~10 hours/week of replaceable work, the overhead of automation eats the savings.
• Process is unstable. If the workflow changes every two months, you spend more on rebuilds than you save in operation.
• Quality bar is too high. Some work needs slow and exactly-right; AI is fast and roughly-right. The mismatch costs more than it saves.
• Internal adoption fails. A perfectly built system nobody uses returns zero. The agency cannot fix this for you.
• Scope creep. The build that was quoted at 6 weeks lands at 14, doubling the amortized cost.
• Wrong agency. Inexperienced shops over-promise and under-build. The remedy is rigorous diligence — see our 25-question evaluation checklist.

Common mistakes in ROI modeling

1. Counting unloaded labor cost. A "$25/hr VA" actually costs ~$35/hr loaded. Use the loaded number.
2. Ignoring API spend. The agency fee is not the whole bill. Add 20–60% on top.
3. Counting hypothetical revenue. "We could capture 100 more leads/month" only matters if those leads convert at observed rates. Use historicals.
4. Skipping management overhead. Every system needs human review time. 2–4 hours/week minimum.
5. Forgetting model deprecation. Once or twice a year, you will pay to migrate. Build it into the model.
6. Modeling only year one. The build cost gets amortized; the operate cost is steady. Year two and three usually look much better — model them all.
7. No downside scenario. Always model pessimistic, base, optimistic. If pessimistic is negative, plan accordingly.

A simple modeling template

Build a spreadsheet with three columns: pessimistic, base, optimistic. Populate the rows:

• Hours/week of work replaced (be honest)
• Loaded hourly cost (× 52 for annual labor saved)
• Revenue lift (units × value, with stated assumptions)
• Cost avoided (headcount, contractors)
• Subtotal: annual benefit
• Build cost (amortized over 24 months)
• Annual retainer or operate cost
• API and tooling spend
• Management overhead (2–4 hrs/wk × loaded rate × 52)
• Subtotal: annual cost
• ROI = (benefit − cost) / cost, expressed as a multiple
• Payback months = (build + 12 × monthly recurring) / monthly net benefit

If pessimistic ROI is negative and base case ROI is below 1.5x, you should not sign. The math is too tight to survive a single bad quarter of execution.

A note on intangible value

AI automation also delivers things that are real but do not belong in the ROI math: speed of decision, optionality, founder time freed for high-leverage work. We argue for these openly with clients but never pretend they are dollars. Keep the ROI honest. Treat the qualitative benefits as a tiebreaker, not a justification.

Further reading

• How much does an AI agency cost — the cost side, in detail.
• What is an AI agency — the foundational definition.
• Build vs buy vs agency — three paths, with cost ranges.
• AI agency evaluation checklist — diligence before signing.
• Our methodology — how we structure engagements.