aicost.ai
VC/PE Diligence Memo
2026-07-22
Pricing as of 2026-07-20
aicost.portfolio-ebitda-lift
Portfolio EBITDA-Lift Estimate
Verdict
MATERIAL LIFT
375 bps sits inside the 200-400 bps range PE operating teams target from AI and cloud cost work.
Key figures
| Margin expansion |
375 bps |
| EBITDA margin |
15% → 18.8% |
| Annual savings |
$1,500,000 |
| from AI |
$900,000 |
| from cloud |
$600,000 |
| Enterprise-value lift |
$18,000,000 |
| EV lift vs today |
25% |
| Payback |
2 months |
| Year-one net |
$1,250,000 |
Assessment
On $40,000,000 of revenue at a 15% EBITDA margin, the addressable AI and cloud cost base is $6,000,000. Applying 30% AI and 20% cloud savings releases $1,500,000 a year, lifting EBITDA from $6,000,000 to $7,500,000, an expansion of 375 bps to a 18.8% margin (material lift). At a 12x exit multiple that is worth about $18,000,000 of enterprise value, roughly 25% of today's implied EV. Implementation pays back in about 2 months, with $1,250,000 net in year one. The savings rates are the assumption to test: model the actual workloads before underwriting them.
Questions for the founder
- What share of the AI spend runs on the most expensive model tier, and could a routing policy move it without hurting quality?
- What is the current cache-hit rate, and what would a serious caching layer do to cost per request?
- Is any cloud capacity committed or reserved, and what is the utilization against it?
- Who owns cost in the org today, and is there a budget or guardrail enforced in the gateway?
- The model assumes 30% AI and 20% cloud savings. What evidence supports those rates for this specific workload mix?
- Does the plan survive if the company's AI usage doubles during the hold period?
Assumptions & method
- Cost savings are treated as flowing to EBITDA close to 1:1, which holds for opex-classified AI and cloud spend.
- Savings rates of 30% (AI) and 20% (cloud) are analyst inputs, not measured results. *
- Exit multiple of 12x is the fund's assumption. *
- Excludes revenue-side effects, headcount changes, and any capitalized spend.
What moves the answer
- aiSavingsPct: Model routing is usually the largest single lever; the price spread between tiers can exceed 5x.
- exitMultiple: Value creation scales linearly with the multiple, so the EV figure is only as good as that assumption.
- adoptionRatePct: Portfolio roll-ups fail on adoption, not on math. Haircut it honestly.