AI for utilities
AI for utilities
A utility already knows where its money leaks, in truck rolls, repeat contacts and asset records nobody trusts, and each of those is a different kind of AI problem.
Utilities carry three burdens at once: a physical network with real safety consequences, a customer base that cannot leave easily and complains accordingly, and a regulator that wants numbers reconciled to the source. Useful AI work in this sector respects which of the three a project sits in.
Field work is an optimisation problem with human constraints
Scheduling is where the largest hard-currency saving sits, and where most projects overreach. A model that produces a mathematically optimal route ignoring skills, certifications, van stock, union agreements and appointment windows produces a schedule the dispatcher overrides by nine in the morning. The version that works treats those constraints as hard rules encoded outside the model, uses prediction for the genuinely uncertain parts such as job duration and no-show likelihood, and shows the dispatcher why a sequence was chosen. Trust is the deliverable. A schedule nobody follows saves nothing.
Asset records are the retrieval problem
Most utilities hold decades of drawings, inspection reports, maintenance histories and network diagrams across systems that were acquired rather than designed. An engineer answering a question about a specific asset often reads four sources of different vintage. Retrieval over that estate with strict versioning and a cited source is high value, and the failure mode is the same one every legacy estate has: a confident answer from a superseded drawing. The control is that applicability and date are part of retrieval, and that the system declines rather than guesses.
Contact volume, in the right order
- First, classify and route accurately. Most utility contact cost comes from misrouting and repeat contact, not from handling time on the first call.
- Second, answer agents rather than customers. An assistant that retrieves the right tariff, policy or outage detail for an agent removes hold time with no external exposure.
- Third, and only after the first two are stable, self-service on the narrow topics where a wrong answer is recoverable. Vulnerable customer, safety and supply interruption topics should route to a human by design, not by fallback.
Regulated reporting has to reconcile
Anything feeding a regulatory return needs a number that traces to a source system, not a model estimate. Use automation to assemble, cross-check and explain the variance, and keep the figure itself deterministic. Digiton builds and operates production AI agents and workflow automation across 8 countries. An AI audit maps which of these four your first project should be.
Frequently asked questions
How do utilities use AI in practice?
Four distinct ways: field scheduling that predicts job duration and no-show risk inside hard constraints a dispatcher trusts, retrieval over legacy asset records and drawings, contact classification and agent assistance to cut repeat contact, and assembly and cross-checking of regulated reporting where the figures stay deterministic.
Can AI schedule field engineers?
It can predict the uncertain parts, job duration and no-show likelihood, and sequence work within hard constraints such as skills, certifications, van stock and appointment windows encoded outside the model. Those constraints must be rules, not learned preferences. A schedule the dispatcher overrides before nine in the morning saves nothing.
Should AI answer utility customers directly?
Only on narrow topics where a wrong answer is recoverable, and only after classification, routing and agent assistance are stable. Vulnerable customer, safety and supply interruption topics should route to a human by design rather than as a fallback, because those are exactly the contacts where an approximate answer causes harm.
Related
Ready to put AI to work?
Book a discovery audit and we will map the highest-ROI AI agents and automations for your business.
Book a discovery audit →