AI energy strategic alignment: making a utility's AI portfolio answer to its strategy

Stage 1 is not the absence of AI — most stage-1 utilities have plenty. It is the absence of any connective tissue between that activity and the strategy the board actually steers by. The load-forecast pilot came from a vendor demo, the drone-inspection trial from an innovation allowance, the churn model from a retail team with budget left in Q4. Each is individually defensible; collectively they are a portfolio nobody chose.

The tell is the inventory question. Ask a stage-1 executive team to list every AI initiative currently consuming money or engineering time, and the list takes three weeks to assemble, disagrees between functions, and surprises everyone with its length. Ask which strategic objective each line serves and the answers are retrofitted on the spot — plausible narratives constructed backwards from what already exists, which is the opposite of alignment.

The cost of staying here is not the money the pilots burn; it is that the strategy's hardest commitments go unsupported while the portfolio grows. A utility can run nine AI experiments and still walk into a price-control review, a rate case or a storm season with no model evidence behind the commitments that actually carry regulatory or financial consequence. Activity is not coverage.

Stage 2 is the mode of the industry and the most comfortable place on the ladder, because it looks finished. The board asked for an AI strategy and received one. The document is competent: it names the energy transition, grid reliability, customer affordability, and commits to being 'AI-enabled' in each. Funding papers now carry a strategic-fit paragraph. Every artefact of alignment exists except the alignment.

The structural problem is that reference is one-directional and unfalsifiable. An initiative that 'supports our reliability ambition' has made a claim nobody can test, because no feeder, no KPI, no baseline and no target attaches to it. The strategic-fit paragraph becomes a genre of creative writing — every proposal passes, so the check selects for writing quality rather than strategic weight, and the portfolio composition remains what it was at stage 1: feasibility and enthusiasm, now better documented.

Stage 2 has a second, quieter cost: it inoculates the organisation against the real work. Because a strategy document exists, the executive team believes the alignment question is answered, and proposals to build a traceability register read as bureaucracy — 'we already have a strategy'. Utilities that sit at stage 2 for three years are harder to move than stage-1 utilities, because the document is evidence, to every internal audience, that the problem is solved.

Stage 3 is where alignment stops being a claim and becomes a data structure. The register is not a spreadsheet for its own sake; it is the enforcement point. A proposal that cannot complete its row — which objective, which decision, which KPI, measured how, against what baseline — is not fundable, however impressive the demo. That single rule changes the portfolio's composition within two funding cycles, because it filters at intake rather than auditing after the fact.

The register's first pass is always uncomfortable, and the discomfort is the point. A typical first pass across a mid-size utility finds a quarter to a third of live initiatives cannot complete a row. Some die, some merge, and some — often the most interesting outcome — re-scope onto an objective that was genuinely unserved: the anomaly-detection team that had been polishing a trading-floor demo discovers the storm-response commitment has no model behind it and becomes the most strategically load-bearing initiative in the portfolio.

What stage 3 does not yet have is motion. The register is accurate but static: it records alignment at the moment of funding and drifts thereafter. Strategy reviews still run without the portfolio in the room; a determination from the regulator or a revision to the demand outlook does not automatically re-open any row. Mapped is a photograph. The next stage is a film.

At stage 4 the interesting change is in what the utility is willing to promise. A stage-3 utility uses AI to help meet commitments it would have made anyway; a stage-4 utility makes commitments it could not responsibly make without the demonstrated capability — a reliability improvement priced on two years of fault-prediction performance, a totex position that assumes condition-based intervention intervals, a connections-processing commitment that assumes automated network analysis. The models have moved from supporting the strategy to underwriting it.

This is also the stage where the regulatory calendar stops being a constraint and becomes the alignment mechanism itself. Price controls, rate cases and IRPs force a utility to write its strategy down in auditable, quantified form on a fixed schedule — which is exactly the artefact a traceability register needs at its far end. Steered utilities draft regulatory commitments from portfolio evidence and re-plan the portfolio when determinations land, in both directions, on the same dates. The submission timetable becomes the portfolio's heartbeat.

The discipline that makes this safe is attribution. A commitment priced on model performance is a liability if the performance evidence would not survive scrutiny, so stage-4 utilities hold the evidence to regulatory standard: holdout feeders or plants, baselines agreed before go-live, benefit rules written down. This is why stage 4 cannot be skipped from stage 2 — the evidence habits are built at stage 3 or they do not exist when the submission needs them.

Stage 5 inverts the question the rest of the ladder asks. Below it, the problem is making the portfolio serve the strategy; here, the recognition is that the portfolio has become the utility's best instrument for discovering that the strategy is wrong. Load forecasts see electrification clusters before the planning cycle does. Asset-health models see a fleet ageing faster than the depreciation assumption. Flexibility dispatch sees value the connections queue is starving. A utility that only pushes strategy down is discarding the most current evidence it owns about the world its strategy describes.

The mechanism is deliberately boring: a watchlist of the strategy's ten or fifteen load-bearing assumptions — demand growth by segment, asset-failure trajectories, flexibility value, connections demand — each with a named model-evidence feed and a tripwire that defines how far reality may drift before the assumption is formally re-opened. When a tripwire fires, a decision forum convenes with a defined mandate. What that forum may decide, and how, is decision-playbook territory; what makes the utility stage 5 is that the trigger exists at all, and fires between refresh cycles rather than waiting for one.

Almost nobody should chase stage 5 directly, and the honest reading of the ladder is that its top is mostly built from its middle: a utility that cannot attribute value to a holdout (stage 3) or hold evidence to submission grade (stage 4) has nothing a strategy refresh should trust. The current demand shock makes the case concrete — a sector where data-centre connection requests are rewriting load outlooks year by year is a sector where the utilities that noticed first were the ones whose forecast evidence had a standing route to the strategy table.