Short answer

ROI compares project-attributable net gains with full cost. Measure before and after on comparable volume and include design, integration, licenses, usage, review, training, maintenance and change.

01 · What you need to understand

How to calculate the ROI of an AI project

Full costs, baseline, observed gains and scale-up decisions. Reliable results require connecting technology to a workflow, data, an owner and a measure. The following principles structure that decision.

01 — Choose one primary workflow metric: minutes, delay, errors, cost, conversion or capacity

Choose one primary workflow metric: minutes, delay, errors, cost, conversion or capacity.

02 — Calculate current cost over a representative volume and period

Calculate current cost over a representative volume and period.

03 — Include human correction and failed cases in solution cost

Include human correction and failed cases in solution cost.

04 — Separate theoretical, observed and actually captured gains

Separate theoretical, observed and actually captured gains.

05 — Set a continuation threshold before the pilot to avoid attachment bias

Set a continuation threshold before the pilot to avoid attachment bias.

02 · Action plan

Action plan

Use this sequence as a starting point. Each step should produce a decision or verifiable output before the next.

  1. Set metric
  2. Measure baseline
  3. Estimate all costs
  4. Run pilot group
  5. Compare equal volume
  6. Decide and document
03 · Mistakes to avoid

Mistakes to avoid

  • Valuing every saved minute without checking reallocation
  • Ignoring operating costs
  • Attributing all improvement to AI
04 · FAQ

Frequently asked questions

Use proxy indicators such as satisfaction, compliance, timing, consistency or capacity and document the method.

Before the pilot, at regular points during it and after a stable adoption period.

05 · Key takeaway

Key takeaway

ROI compares project-attributable net gains with full cost. Measure before and after on comparable volume and include design, integration, licenses, usage, review, training, maintenance and change.

The important point is to progress through evidence: a precise use case, representative test, documented limits and an outcome-based decision.