The business case for AI at a distribution utility rests on four levers: reducing non-revenue water and gas loss, cutting O and M cost, improving reliability metrics like SAIDI and SAIFI, and sharpening capital prioritization so replacement dollars go to the highest-risk assets. Because non-revenue water can run 15 to 30 percent of supply and capital budgets are large and rate-regulated, even modest improvements produce material returns. This playbook shows how to build a defensible ROI model for utility AI, quantify each lever, and estimate realistic payback for the regulated operator.
Small percentage gains move large regulated numbers
Distribution utilities run on large capital and operating budgets set through the rate case, which changes the ROI math. A water utility losing 15 to 30 percent of treated water as non-revenue water is paying to treat and pump water that generates no revenue; recovering even a few points of that loss frees real money and defers the need for new supply capacity. A gas utility running fixed-schedule leak surveys spends O and M on low-risk miles while risk-based targeting could redirect the same crews to where leaks actually are. An electric utility measured on SAIDI and SAIFI faces regulatory penalties and customer dissatisfaction when reliability slips, so minutes of avoided outage carry both financial and regulatory value.
The largest lever is often capital prioritization. Utilities replace mains, cables, and transformers on aging-asset programs worth tens of millions annually. If AI-based risk scoring lets the utility replace the same number of high-consequence assets while deferring low-risk ones, the effective productivity of the capital program rises without spending more. That deferral, properly documented, is both an ROI story and a prudence story for the regulator.
Four ROI levers with drivers and payback
Build the business case lever by lever. Each has a distinct driver, a distinct measurement, and a distinct payback profile.
| ROI lever | Value driver | Typical payback |
|---|---|---|
| Non-revenue water / gas loss | Recovered volume that was treated, pumped, or purchased but never billed | 6 to 18 months where AMI or district metering supports fast leak targeting |
| O and M cost reduction | Fewer emergency repairs, risk-based leak surveys, fewer confirmatory truck rolls | 12 to 24 months as risk-based workflows replace fixed schedules |
| Reliability (SAIDI / SAIFI) | Avoided outage minutes, faster restoration, avoided regulatory penalties | 12 to 36 months; value partly in penalty avoidance and customer satisfaction |
| Capital prioritization | Deferring low-risk replacement while targeting high-consequence assets | Multi-year; large absolute value from a more productive capital program |
| Cost to serve | Contact deflection, billing-anomaly catch, lower bad debt | 6 to 12 months on clean CIS and contact-center data |
Build an ROI case a CFO and a regulator both accept
- Quantify your current non-revenue water or unaccounted-for gas in both volume and dollars, and set a recovery target so the leak-detection business case has a hard baseline.
- Separate hard savings from soft benefits: put deferred capital, recovered water, and avoided repairs in the hard column, and reserve customer-satisfaction gains as supporting evidence.
- Model capital prioritization as productivity, not cut: show that the same or fewer dollars replace more high-consequence risk, which reads as prudence to a regulator.
- Price reliability improvement using both avoided penalties and the cost of the truck rolls and overtime you no longer spend confirming and chasing outages.
- Attach the ROI evidence to the governance record for each model so the same documentation supports both the internal case and rate recovery.
ROI claims that fall apart under scrutiny
- Counting recovered water twice or claiming loss reductions the field never actually closed out, so the model looks better on paper than the meters show.
- Treating capital prioritization as a budget cut, which invites the regulator to lower the capital allowance instead of crediting a more productive program.
- Ignoring the cost of the AI program itself, including data integration, sensors, licensing, and change management, and overstating net payback.
- Leaning on soft benefits like satisfaction as the primary justification, which will not survive a CFO review or a rate-case cross-examination.
Track the numbers that prove the return
- Non-revenue water or unaccounted-for gas percentage, and dollars of recovered volume attributable to AI-directed detection each quarter.
- O and M cost per mile or per asset, and the reduction in emergency versus planned work as risk-based targeting takes hold.
- SAIDI and SAIFI trends and avoided regulatory penalties tied to AI-assisted outage prediction and dispatch.
- Capital productivity: high-consequence risk retired per capital dollar, and documented deferral of low-risk replacement.
Frequently asked questions
What is a realistic first-year ROI for a utility AI project?
For a well-scoped leak-detection or demand-forecasting pilot with existing AMI or district metering, payback in the 6 to 18 month range is realistic because the savings are direct and measurable. Asset-health and capital-prioritization programs pay back over multiple years, since the benefit is deferred capital rather than an immediate cash saving. Be wary of vendor claims of instant transformational ROI; the credible cases are specific and measured.
How do we justify AI spending to a regulator who scrutinizes every dollar?
Frame it as prudence and productivity. Show that AI directs the same or fewer capital and O and M dollars to higher-risk work, recovers water that was already treated and paid for, or reduces outage minutes customers experience. Attach the model documentation and outcome measurement to the filing. Regulators respond to defensible, evidence-backed spending far better than to promises of innovation.
Does the ROI justify replacing our fixed leak-survey schedule?
Often yes, particularly in gas and water. Fixed schedules spend crew time uniformly across the network regardless of risk. Risk-based targeting sends the same crews first to the segments most likely to leak, which recovers loss faster and can defer replacement. The ROI shows up as recovered non-revenue volume plus more productive O and M, and the change is usually defensible to a regulator as a safety and efficiency improvement.
Related reading
Go deeper on this sector and topic.