How do you estimate and present ROI for an AI project to a customer?

30-second answerSay your answer out loud first, then reveal.

Build a simple, transparent model: volume × time per task × cost per hour × the percentage of time saved (from pilot measurements), plus quality benefits (fewer errors, faster SLAs, revenue effects), minus costs (platform, model usage, integration, change management, ongoing ops). Use conservative, measured inputs, show ranges and assumptions, and let the customer adjust them. Tie it back to the metrics they already report.

Example

text

Tickets per month:           60,000
Avg handle time today:       12 min
Measured reduction (pilot):  30% → 3.6 min saved per ticket
Hours saved per month:       60,000 × 3.6 / 60 = 3,600 hours
Loaded cost per agent hour:  ₹600
Monthly value:               3,600 × ₹600 = ₹21.6 lakh
Monthly costs:               platform + tokens + support ≈ ₹6 lakh
Net monthly benefit:         ≈ ₹15.6 lakh  (payback on ₹40 lakh setup ≈ 2.6 months)

Make it credible

Use pilot data, not vendor benchmarks.
Sensitivity analysis: show results if the time saved is 15% vs 30%.
Separate "capacity freed" from "cost cut". Saved hours may go to backlog reduction or better service, not headcount reduction. Be careful with the language.
Include qualitative value: consistency, compliance, employee satisfaction, faster onboarding.
Include all costs: customer staff time, maintenance, model costs at scale.

PreviousDesign a workflow for customer SMEs to label data and review AI outputs during an engagement.

Every expert started right here.

How do you estimate and present ROI for an AI project to a customer?

Example

Make it credible

Related