Why AI ROI conversations feel unsatisfying
Most AI business cases stop at "hours saved". Finance sees through it, boards discount it, and the programme loses funding. The organisations winning at AI have moved on to a four-layer ROI model that reflects how AI actually creates value.
The four layers of AI ROI
1. Cost savings
Direct reductions in run cost: licences retired, vendor spend cut, contractor hours removed. Measurable in the ledger within a quarter but rarely the largest component.
2. Capacity release
Hours returned to employees, but only counted when redeployed to specified higher-value work with a named owner. Capacity that leaks into meetings is not ROI.
3. Quality and risk
Error rate reduction, rework reduction, compliance improvements, customer complaint reductions. Often the largest component and the one most enterprises fail to quantify because the baseline was never captured.
4. Revenue impact
Faster cycle time enabling more deals, higher conversion from better follow-up, improved retention from faster resolution, new product capability. This is where AI stops being a cost play and becomes a growth engine.
The metrics that matter
- Cycle time per transaction, before and after.
- Cost per transaction, fully loaded.
- Error rate and rework rate.
- First-contact resolution for service workloads.
- Capacity redeployment tracked to a named use case.
- Revenue attribution for customer-facing processes.
- Payback period and NPV on conservative assumptions.
Building an AI investment governance model
Stage 1: Portfolio design
Every AI initiative is scored on strategic fit, expected value, data readiness, and risk. Weak initiatives are killed before they consume budget.
Stage 2: Stage-gated funding
Fund in tranches: discovery, pilot, production, scale. Each gate requires the previous stage's results to meet pre-declared criteria.
Stage 3: Live ROI dashboard
Publish the KPIs weekly for the first quarter after go-live, then monthly. Finance signs off the baseline and the ongoing measurement.
Stage 4: Portfolio review
Quarterly. Kill what is not working, double down on what is, reallocate capacity to the next wave.
A worked example
An insurer's claims triage automation delivered £480k of direct cost saving. Adding capacity redeployment (adjusters moved to complex claims), quality (62% reduction in mis-routing), and revenue impact (retention lift from faster resolution) took defensible 18-month ROI to 6.4x. Cost saving alone was 22% of the total.
What separates winners from expensive experiments
- They capture the baseline before any change goes live.
- They measure all four ROI layers, not just cost.
- They track capacity redeployment to named use cases.
- They kill weak initiatives quickly.
- They defend the numbers with finance from day one.
See how this plugs into the wider automation ROI framework and the OpenGaps three-step method.