Energy & UtilitiesLeadership Insights & Strategy
The AI roadshow: taking an energy and utilities AI strategy round the estate
An AI roadshow is the leadership programme of taking an approved AI strategy physically round a utility estate — control centres, generation sites, depots, field crews and customer operations — so a board decision becomes site-level ownership. It is a change instrument, and it is judged on what it captures, not what it broadcasts.
Key takeaways
- An AI roadshow is a capture instrument, not a broadcast one. A stop is measured by how many site-specific use cases left the building with a named proposer attached, not by how many people heard the strategy.
- What an executive presents changes completely by stop type: advisory versus authority in a control room, scheduling and access on a wind farm, handle time and the script in a contact centre, the work pack at a depot muster. One deck at all four is the signature of a tour that will not land.
- The question a 25-year control-room operator asks — "is this replacing my job?" — has one survivable answer: a specific description of what changes in their day, not a reassurance about headcount that nobody believes and no executive can guarantee.
- Roadshow energy dissipates in about six weeks without a funded intake process. A use case with no standing board, no seed budget and no published decision teaches the estate that leadership visits are theatre — which makes the next programme harder, not neutral.
- In utilities, "automation" carries specific historical weight — protection schemes, remote switching, meter reading, consolidated control rooms — so safety-culture framing has to be explicit. Without it, an AI programme is heard as the next round of the same thing.
Abbreviations used on this page
- TSO
- Transmission system operator
- DSO
- Distribution system operator
- SCADA
- Supervisory control and data acquisition
- EMS
- Energy management system (the control-centre application suite)
- ADMS
- Advanced distribution management system
- OMS
- Outage management system
- GIS
- Geographic information system (the network records asset)
- CMMS
- Computerised maintenance management system
- AMI
- Advanced metering infrastructure (smart meters and their head end)
- ETR
- Estimated time of restoration (the figure given to a customer during an outage)
- SAIDI
- System average interruption duration index — the headline reliability KPI
- EWC
- European Works Council (the statutory information and consultation body)
Free · 8 questions · ~3 minutes
Score your roadshow
Eight questions, one at a time, about three minutes. Answer them and we build your personalised roadshow report — where your programme sits on the five-stage ladder, your score on each of the four dimensions, and the specific thing standing between your tour and site-level ownership — and send it to your inbox.
0 of 8 answered
Pick an option to continue
Report ready
Your personalised roadshow report is ready
Tell us where to send it. Your stage appears on screen straight away, and the full report — dimension scores, the stop types you have not yet reached, and a route and cadence plan for your weakest dimension — arrives in your inbox.
Your result
Your full report is on its way to your inbox.
Stage 1 · Announced
Announced is the stage where the AI strategy exists as a board decision and an intranet post, and nobody outside head office has met it in person.
Your next moveBuild a route document — which sites, in which order, with which executive and on which shift — and a one-page stop dossier per site drawn from that site's own numbers.
Stage 2 · Toured
Toured is the stage where executives have physically presented at sites, but the traffic is one-way — the estate has heard the strategy and contributed nothing to it.
Your next movePut a scribe and an engineer in the room, run a one-page capture card per raised idea with a named proposer, and write back to the site inside forty-eight hours.
Stage 3 · Understood
Understood is the stage where people at each stop can explain, in their own words and their own units, what the AI does and does not do to their job.
Your next moveCreate a standing, quorate intake board with a seed budget and published decision dates, and give every captured item a disposition — funded, parked with a reason, or declined.
Stage 4 · Owned
Owned is the stage where use cases raised at stops have named site-level owners, funded delivery pairings and a visible place in the site's own KPI set.
Your next movePublish the roadshow as a standing operating calendar with named deputies, a written intake quorum and rotating executive fronting.
Stage 5 · Self-sustaining
Self-sustaining is the stage where the tour has become a standing rhythm: sites pull visits, a funded board disposes of what they raise, and the executive's stop is a review rather than a launch.
Your next moveName a cadence custodian, put site ownership in role descriptions, and review the pull ratio quarterly as a leading indicator of decay.
0 / 24
Narrative & framing
— / 6
Site engagement depth
— / 6
Use-case capture
— / 6
Follow-through & funding
— / 6
Your score maps to a stage on the roadshow ladder. The dimension breakdown matters more than the total: the lowest dimension is what caps the programme, and in most utilities it is follow-through rather than framing. Your lowest-scoring dimension is —, and that is where the next investment belongs.
Your score maps to a stage on the roadshow ladder. The dimension breakdown matters more than the total: the lowest dimension is what caps the programme, and in most utilities it is follow-through rather than framing.Your four dimensions score evenly, so there is no single weak link to attack — follow the stage’s next move above rather than picking a dimension.
Want the route designed with you?
We will walk your operations and communications leads through the dimension scores, sequence your estate into a stop order, draft the stop dossiers from your own SAIDI, ETR and work-order data, and leave you with a costed 90-day route. No obligation, and you keep the route either way.
How the score maps to a stage
- 0–4 — Stage 1, Announced. Announced is the stage where the AI strategy exists as a board decision and an intranet post, and nobody outside head office has met it in person.
- 5–9 — Stage 2, Toured. Toured is the stage where executives have physically presented at sites, but the traffic is one-way — the estate has heard the strategy and contributed nothing to it.
- 10–14 — Stage 3, Understood. Understood is the stage where people at each stop can explain, in their own words and their own units, what the AI does and does not do to their job.
- 15–19 — Stage 4, Owned. Owned is the stage where use cases raised at stops have named site-level owners, funded delivery pairings and a visible place in the site's own KPI set.
- 20–24 — Stage 5, Self-sustaining. Self-sustaining is the stage where the tour has become a standing rhythm: sites pull visits, a funded board disposes of what they raise, and the executive's stop is a review rather than a launch.
What an AI roadshow is — and what it is actually for
A definition, the broadcast trap it exists to escape, and the loop that turns a board decision into site-level ownership.
An AI roadshow is the leadership programme of taking an approved AI strategy physically round an operator's estate — control centres, thermal and hydro stations, wind and solar sites, network depots, field crew musters, contact centres and the metering back office — so that a decision made in a board room becomes a commitment held at site level. It is a change and communication instrument first and a technology exercise a distant second: nothing on the route requires a model to exist yet.
Utilities need one more than most industries for structural reasons. A vertically integrated operator is a federation of buildings with different clocks: a control centre runs continuously on rotating shifts, a station runs on outage seasons, a depot on a daily work-pack cycle, a contact centre on half-hourly handle-time targets. There is no hour at which all of them are available and no channel all of them read. The estate has to be visited because it cannot be assembled.
The trap is that a roadshow is almost always scoped as broadcast when its value runs the other way. The most valuable sentence at a stop is not the executive's; it is the one naming an operational cost head office did not know existed. Research bodies listing barriers to AI in energy put data access, digital infrastructure and skills alongside, in the IEA's own phrasing (opens in a new tab), "a general resistance to change". A tour that only transmits does nothing about the last and learns nothing about the rest.
The roadshow loop: broadcast down, capture up
Two paths out of the same board decision. The top lane is what most utility programmes run — a deck, a town hall, questions taken but not recorded — and it decays within about six weeks. The middle and bottom lanes are the same tour with capture and a funded intake attached, which is the only version that produces site-level ownership.
- Human in the loop
- Data & feeds
- Where value leaks
- System-of-record action
- AI / model
The process, in words
- On the broadcast path, one deck is written centrally and shown at every site. The town hall runs forty-five minutes one-way, questions are answered verbally, and nothing site-specific is written down. Within about six weeks the visit has no operational trace, and the site has learned that leadership visits are a format rather than a conversation.
- On the capture path, the same board decision is broken into a one-page stop dossier per site — that site's reliability numbers, its open work orders, the last event everyone in the room remembers. The executive speaks for twenty minutes in those units and then listens. Every idea raised is written on a capture card with a named proposer and the system of record it would touch, and the site receives a verbatim write-back within forty-eight hours.
- Follow-through is where ownership is created or lost. Captured cards go to a standing, quorate intake board with its own seed budget, which publishes a disposition — funded, parked with a reason, or declined with a reason — on a stated date. Funded items get a named site owner alongside the delivery engineer, and the next executive visit reviews that work instead of relaunching the strategy.
Step-by-step insights
- The stop dossier is the whole difference in preparation
- One page, built from the site's own systems, and what allows an executive to speak in the room's units for twenty minutes without a script. For a depot it is open work orders by age, wasted journeys and the last storm's restoration performance; for a control centre, alarm counts per shift, current ETR accuracy on wind-driven faults and the two applications operators complain about. It takes an analyst about a day per site and is the highest-return hour in the programme.
- Twenty minutes, then listen — and mean it
- The common design error is a fifty-minute presentation with ten minutes of questions, which produces the questions people ask when they are being managed. Twenty minutes of framing followed by open floor changes what surfaces, particularly in control rooms where the room waits to see whether the first hard question gets a real answer. Everything after that answer depends on it, which is why the job question should be invited rather than avoided.
- The capture card — three fields that matter
- Named proposer, the operational cost in the site's own units, and the system of record the idea would touch — ADMS, OMS, GIS, CMMS, AMI head end or the billing platform. The name makes the item retrievable and gives the site something to follow. The cost lets the intake board rank without a business case. The system of record lets an engineer say whether this is a two-week integration or a three-year vendor negotiation.
- Forty-eight hours is the credibility window
- The write-back is the cheapest high-leverage artefact in the loop: a short document listing verbatim what the site raised, who raised it, and the date each item will be decided. It costs the scribe an hour. Beyond about a week the site has already concluded nothing is coming, and the reply arrives as an apology rather than as evidence of a process. Publish it on the muster room board, not only an intranet page nobody at the depot opens.
Site-level ownership released across the roadshow ladder
The curve is not linear. Ownership stays close to flat through Announced and Toured — where most programmes are — and inflects at Understood, when what the site says starts changing what the programme does. That flat opening is why tours measured in stops completed report activity without any change in the estate.
Site-level ownership of the AI strategy by stage
- Stage 1 · Announced — 22% of operators. Announced is the stage where the AI strategy exists as a board decision and an intranet post, and nobody outside head office has met it in person.
- Stage 2 · Toured — 34% of operators. Toured is the stage where executives have physically presented at sites, but the traffic is one-way — the estate has heard the strategy and contributed nothing to it.
- Stage 3 · Understood — 26% of operators. Understood is the stage where people at each stop can explain, in their own words and their own units, what the AI does and does not do to their job.
- Stage 4 · Owned — 14% of operators. Owned is the stage where use cases raised at stops have named site-level owners, funded delivery pairings and a visible place in the site's own KPI set.
- Stage 5 · Self-sustaining — 4% of operators. Self-sustaining is the stage where the tour has become a standing rhythm: sites pull visits, a funded board disposes of what they raise, and the executive's stop is a review rather than a launch.
Curve shape: logistic, plotted from the stage data above. Distribution: Shape consistent with Prosci's change-management effectiveness research.
The five stages in detail
Announced, Toured, Understood, Owned, Self-sustaining — what each looks like inside a real building, the signals a reviewer can check in an afternoon, and the anti-pattern that traps programmes there.
Each stage below describes an observable condition of the estate rather than a phase of a plan. The hallmarks are things you can see in a building, the diagnostic signals are checks you can run this week against your own calendars and capture lists, and the anti-pattern is the mistake most often made trying to leave that stage — which is, in four cases out of five, doing more of what got you there.
Select a stage
Every stage's full detail is in the page source — the selector only changes which panel is visible, so nothing here depends on JavaScript to exist.
Stage 1
Announced
22% of operators sit here
Announced is the stage where the AI strategy exists as a board decision and an intranet post, and nobody outside head office has met it in person.
Stage 1 is not the absence of a strategy but the absence of any mechanism by which it reaches the people who would have to change their day. The board paper is often good: it names the domains, sets an ambition, attaches a number. What is missing is a route — which buildings, in what order, and what the executive says in each.
The tell is the vocabulary gap. At head office the programme is described in portfolio language: use cases, value pools, enablers. In the control room and at the depot it has no name at all, or one acquired from the corridor — "the AI thing". That gap is not a communications failure to be fixed by a better email; it is a strategy never translated into the units of a single site.
Announced is cheap to leave and expensive to sit in, because it does not stay neutral. Every month the strategy exists without an executive in the room, the estate builds its own explanation — usually the least generous one available, which in this industry is redundancy. The first stop is then not an introduction but a correction.
In practice
The intranet post nobody at the depot opened
A DSO published a well-written AI strategy article in January: five domains, a governance structure, a named sponsor. In April an operations director visiting a regional depot asked the supervisor what he made of it. He had not seen it — the shared PC is in the muster room and crews are on the road by 07:10. What he had heard, from a contractor, was that "they're bringing in AI to do the fault-finding".
What it looks like
- The strategy exists as a board paper, a press line and an intranet article
- No executive has presented it inside an operational building
- Site managers learned about it from the trade press or a recorded town hall
- There is no route, no dates and no named executive per stop
Diagnostic signals you can check this week
- Ask three site managers to name the programme. Three different answers, or none, means you are here
- Check whether any executive calendar entry in six months names an operational site and the AI programme together
- Ask what the strategy says about jobs. If it says nothing, the estate has already filled the gap
- Look for a route document — sites, dates, named executive per stop. Its absence is the definition of stage 1
Anti-pattern · Mistaking the all-hands for the roadshow
The instinctive fix is a company-wide webinar with a Q&A box: cheap, wide, and it changes nothing. Shift workers cannot attend, field crews watch none of it, and the questions that matter are not the ones people type under their own name into a moderated queue. A recorded all-hands is a prerequisite, never a substitute — it establishes that the programme exists, and the roadshow makes it real in a specific building.
What holds you here
There is no route and no stop-level content, so the strategy has no way of reaching an operational building.
Highest-leverage next move
Build a route document — which sites, in which order, with which executive and on which shift — and a one-page stop dossier per site drawn from that site's own numbers.
Cost of leaving
- Effort
- 4–6 weeks
- Team
- One executive sponsor, a programme lead, a scheduler with access to shift rosters
- Risk
- Low — nothing operational changes, but the route decisions made here set the tour's credibility
- To next stage
- 4–6 weeks
If this is you, the next step is
A two-week engagement: sequence the estate, build the stop dossiers, brief the executive.
Stage 2
Toured
34% of operators sit here
Toured is the stage where executives have physically presented at sites, but the traffic is one-way — the estate has heard the strategy and contributed nothing to it.
Stage 2 is the most common place to find a utility AI programme, and it looks like progress because the hard logistics were genuinely done. Somebody moved a chief operating officer through six sites in three weeks, arranged shift overlap so operators could attend on the clock, and got the safety brief right. What did not happen is any change in direction of travel: information left head office and nothing came back.
The cause is that the tour was scoped as communication, so it was staffed as communication — an executive, a communications manager and a deck. There is no scribe, no capture template, no engineer who can say on the spot whether an idea is plausible. So when a control engineer with nineteen years on the desk says the OMS restoration times are wrong on wind-driven faults, it is received as an anecdote rather than as the most valuable output of the day.
Time at stage 2 is not free. A site that receives a visit, asks a real question and hears nothing again has learned that leadership visits are a broadcast format. The second tour into that building is materially harder than the first. Programmes that have toured twice without capturing anything are usually further from stage 3 than programmes that have never toured at all.
In practice
The six-site tour that produced a photograph
A vertically integrated utility ran a six-week executive tour of two control centres, two generating stations and two depots. Attendance was strong and the internal comms team produced an excellent set of photographs. Four months later the programme team was asked to shortlist candidate use cases and started from the vendor's standard list, because nothing said in any of the six buildings had been written down in a form anyone could retrieve.
What it looks like
- Executives have presented in person at control centres, sites and depots
- The same deck was shown at every stop, with the site name changed
- Questions were taken from the floor but not recorded
- No use case raised at a stop has a name, a number or an owner attached
Diagnostic signals you can check this week
- Ask for the list of use cases raised at stops. No list means the tour was a broadcast
- Compare the decks used at the control centre and at the wind farm. The same file means the framing was not stop-specific
- Count the people in the executive party. No scribe and no engineer means capture was never designed in
- Ask a site that has been visited what happened next. Silence is the stage-2 answer
Anti-pattern · Adding more stops to fix a flat tour
When a tour lands flatly the reflex is coverage: more sites, a second lap. It is the wrong lever. A flat stop is rarely caused by too few people hearing the message; it is caused by the message arriving in portfolio language and leaving with nothing. Doubling the stops doubles the number of buildings that have learned the format is one-way. Fix one stop's design — dossier, framing, scribe, capture card, write-back — and re-run it at the same site.
What holds you here
The stop is designed to transmit, so nothing site-specific is recorded and the programme's use-case list stays generic.
Highest-leverage next move
Put a scribe and an engineer in the room, run a one-page capture card per raised idea with a named proposer, and write back to the site inside forty-eight hours.
Cost of leaving
- Effort
- 1–2 months
- Team
- Add a scribe and a delivery engineer to the executive party; a capture template; a triage owner
- Risk
- Medium — the first captured list creates an expectation of a reply, and there is not yet a board to give one
- To next stage
- 1–2 months
If this is you, the next step is
We design the dossier, the framing and the capture card, and sit in on the first run.
Stage 3
Understood
26% of operators sit here
Understood is the stage where people at each stop can explain, in their own words and their own units, what the AI does and does not do to their job.
Stage 3 is where the roadshow starts doing what it is for, and the test is linguistic. Stand in the mess room a fortnight after a stop and ask what the programme is. At stage 2 you get the strategy's own words back, or nothing. At stage 3 you get "they're trying to make the restoration times we give customers less rubbish on wind days" — understanding, because it is expressed in the speaker's own operational units.
Getting there requires the executive to drop portfolio framing entirely for their twenty minutes. In a control centre the subject is advisory versus authority, alarm load and who is accountable at 03:00. At a station it is outage windows and the maintenance backlog. In a contact centre it is handle time and what an agent may say. A director holding the site's own numbers converts more of the room than one with a better deck, and the dossier is what makes that possible.
The constraint that emerges is asymmetry between hearing and answering. Every stop now produces a real list, and the lists arrive faster than anyone can dispose of them. Sites that raised something well and got a warm acknowledgement start asking what happened. Without a standing body with a budget and decision rights, the programme is generating a promise it cannot keep.
In practice
The control-room question that reframed the programme
At a transmission control centre a shift manager asked one question: "when your model and my engineer disagree at three in the morning, who is on the incident report?" The executive answered directly — the engineer decides, the model is advisory, the recommendation and its inputs are logged so the review has both. Repeated by the shift manager to two other shifts, that answer did more for adoption than the preceding forty minutes.
What it looks like
- Site teams describe the programme in site vocabulary, not corporate vocabulary
- The job question is asked out loud at most stops and answered specifically
- Each stop produces a written capture list with named proposers
- The site receives a verbatim write-back within days, not weeks
Diagnostic signals you can check this week
- Ask a site team two weeks after a stop to describe the programme. Site vocabulary means stage 3; corporate vocabulary means stage 2
- Check whether the job question was asked out loud at each stop, and read the answer given
- Look at the capture lists for named proposer, site and system of record. Anonymous lists do not convert
- Measure days from stop to write-back. Beyond a week the site has concluded nothing is coming
Anti-pattern · Promising at the stop to secure the room
Understanding creates warmth, and warmth tempts an executive into commitment: "leave that with me, we'll get it funded". In a utility that promise is remembered precisely, repeated across shifts and dated. When it is not delivered — usually because capital planning runs on an annual cycle the executive did not have in mind — the site concludes it was managed. Commit to a decision date, never to an outcome, and publish the disposition whichever way it goes.
What holds you here
Capture now outruns disposition: sites raise more than the programme can decide on, and unanswered lists read as broken promises.
Highest-leverage next move
Create a standing, quorate intake board with a seed budget and published decision dates, and give every captured item a disposition — funded, parked with a reason, or declined.
Cost of leaving
- Effort
- 2–4 months
- Team
- Executive plus a delivery engineer per stop, a triage lead, and a finance partner who can define a seed budget
- Risk
- Medium — the expectation created by good capture becomes a liability without a funded route to disposition
- To next stage
- 2–4 months
If this is you, the next step is
Decision rights, quorum, seed budget and a published disposition format — in three weeks.
Stage 4
Owned
14% of operators sit here
Owned is the stage where use cases raised at stops have named site-level owners, funded delivery pairings and a visible place in the site's own KPI set.
Stage 4 is where the roadshow stops being a programme activity and becomes an operating relationship. The observable difference is accountability: a use case raised at the Northern depot has a name from the Northern depot on it, alongside the delivery engineer, and that person's own performance conversation references it. Ownership stated in a steering pack is not ownership; ownership in a site manager's objectives is.
The mechanics that get a programme here are financial and unglamorous. A seed budget large enough to start without a capital case, a delivery pairing that puts an engineer alongside the proposer within weeks, and a published disposition list naming what was declined and why. The declines matter most: a site that sees three of its four ideas declined with reasons trusts the process, where a site hearing only about the approved one cannot tell a process from favouritism.
The remaining constraint is that the rhythm still depends on the executive. Stops happen because a director's office schedules them; intake meets because the programme convenes it. It works, and only as long as that individual stays in post. Utilities see regular executive movement, and a change of chief operating officer is the most common single cause of a stage-4 programme falling back to stage 2 inside two quarters.
In practice
The depot supervisor whose idea became her objective
A supervisor raised, at a stop, that crews were re-driving to sites because the GIS record of pole-mounted assets did not match the CMMS work pack. The intake board funded a four-week reconciliation against AMI and inspection imagery. She was named site owner, the delivery engineer worked from her depot two days a week, and wasted journeys per crew went on the depot board next to the safety metrics. Two quarters later the depot proposed the follow-on itself.
What it looks like
- Every funded use case has a site owner as well as a delivery owner
- Captured items receive a published disposition on a stated date
- Site KPIs — SAIDI, ETR accuracy, work-order completion — carry the AI contribution explicitly
- Sites start asking for the visit rather than being scheduled for one
Diagnostic signals you can check this week
- Open a funded use case and look for two names — a delivery owner and a site owner. One name means stage 3 with a budget
- Ask to see the published disposition list, including declines with reasons
- Check whether any AI-linked measure appears on a site's own performance board, not only in the programme pack
- Count inbound requests for a stop. Sites pulling rather than being pushed is the stage-4 signature
Anti-pattern · Letting the executive's diary be the operating system
Because stage 4 works, the temptation is to leave the rhythm attached to the person who created it. Every stop, every intake meeting and every escalation routes through one director's office, and programme health tracks that individual's availability. The fix is deliberately boring: publish the cadence as a calendar with named deputies, give the intake board a written quorum that does not require the sponsor, and rotate which executive fronts a stop.
What holds you here
The rhythm still depends on one executive's diary, so a sponsor change resets the programme to a broadcast tour.
Highest-leverage next move
Publish the roadshow as a standing operating calendar with named deputies, a written intake quorum and rotating executive fronting.
Cost of leaving
- Effort
- 4–8 months
- Team
- Intake board with finance and HR representation, delivery pairings per funded item, named site owners
- Risk
- Higher — funded commitments now exist at multiple sites, and a withdrawal is visible across the estate
- To next stage
- 4–8 months
If this is you, the next step is
We convert an executive-led rhythm into a published operating calendar with deputies and quorum.
Stage 5
Self-sustaining
4% of operators sit here
Self-sustaining is the stage where the tour has become a standing rhythm: sites pull visits, a funded board disposes of what they raise, and the executive's stop is a review rather than a launch.
Stage 5 is narrower than it sounds and is not a state of universal enthusiasm. It is an institutional arrangement: a published stop calendar, a quorate funded board that meets whether or not the sponsor attends, a disposition record anyone can read, and enough site-level owners that the centre of gravity has left head office. The executive still tours, but the visit's purpose has inverted — they arrive to review what the site did.
The most reliable indicator is co-presentation. At stage 5 the twenty minutes are delivered by an operator or supervisor from another site who ran the thing being discussed, with the director present for accountability and funding questions. Peer delivery survives scepticism that executive delivery does not, particularly in control rooms, where the credibility currency is time on the desk.
Sustaining stage 5 is a governance discipline and it regresses easily. Reorganisations dissolve intake boards, a bad year moves the seed budget, and a new operating model quietly removes the site-owner role from a job description. The failure is never announced; it shows up two quarters later as a falling pull ratio, which is why that number belongs on the dashboard from the day it is first positive.
In practice
The stop that the site ran
By its third year one network business had a published quarterly stop calendar per region. At a depot stop the forty-minute slot was run by a control engineer from a neighbouring region who had built the ETR reconciliation work, with the operations director present for the funding and accountability questions. Of six ideas captured that morning, two were already on the intake board's forward list because the depot had submitted them the previous month.
What it looks like
- The intake board meets and funds on a published calendar without the sponsor convening it
- Sites request stops and co-present their own results at them
- Union and works-council engagement is a scheduled standing item, not an escalation
- New executives are inducted into the cadence rather than restarting it
Diagnostic signals you can check this week
- Check whether the intake board met and funded during a month the sponsor was on leave
- Count the share of stops fronted by a site person rather than an executive
- Track the pull ratio — site-initiated requests per quarter — over four quarters, not one
- Ask whether the last executive appointment was inducted into the cadence or announced a new approach
Anti-pattern · Declaring the roadshow finished
Once the rhythm is self-sustaining it reads as overhead, and the calendar is the first thing a cost programme removes: the estate understands the strategy now, so why keep touring? Within a year the pull ratio decays, capture dries up, and the next AI wave arrives to an estate that must be re-toured from stage 1 — this time with a memory of a cadence that was withdrawn. Treat the calendar as a permanent operating cost, like the safety stand-down.
What holds you here
Sustaining the rhythm is a governance problem — reorganisations, budget cycles and role redesigns quietly remove the parts that make it standing.
Highest-leverage next move
Name a cadence custodian, put site ownership in role descriptions, and review the pull ratio quarterly as a leading indicator of decay.
Cost of leaving
- Effort
- Continuous
- Team
- Standing intake board, published calendar, site owners in role descriptions, a named cadence custodian
- Risk
- Concentrated — low frequency, high consequence: reorganisation and budget cycles are what break it
If this is you, the next step is
We run the calendar, quorum and ownership model against a real restructure scenario.
Where utility AI roadshows actually land
The distribution across the ladder, and why the Toured → Understood step is the largest single loss.
Most utility AI roadshows stall at Toured. The distribution below is weighted heavily toward programmes that have done the hard logistics — moved an executive through operational buildings, arranged shift overlap, got the safety brief right — and have no mechanism for anything to travel the other way. A small minority have reached the point where sites request the visit rather than being scheduled for one.
Distribution of utility AI programmes across the roadshow ladder
Illustrative distribution. Toured is both the mode and the plateau: the drop from Toured to Understood is the largest single transition loss on the ladder, and it is caused by capture design rather than by executive quality.
Share of programmes
- 22% — 1 · Announced
- 34% — 2 · Toured (the plateau)
- 26% — 3 · Understood
- 14% — 4 · Owned
- 4% — 5 · Self-sustaining
Source: Illustrative distribution, synthesised from IEA, Eurelectric and Prosci change-management research
The plateau is not a failure of nerve; it is what happens when a change programme is resourced as communications. Prosci's benchmarking research (opens in a new tab) finds that individual adoption, not organisational announcement, determines whether an initiative meets its objectives; the change-management literature (opens in a new tab) has said the same for three decades, and McKinsey's organisational-performance research (opens in a new tab) reaches the same conclusion about transformations generally. What is utility-specific is the blocker's shape: the estate is spread across buildings with incompatible clocks, so the only channel with real reach is a person in the room — and that channel is expensive enough that programmes use it to transmit rather than to listen.
The sector's own bodies describe the same pressure from the workforce side. Eurelectric's skills work (opens in a new tab) records its president — E.ON's chief executive — saying that baby-boomer retirement means losing a third of his most experienced technicians over a decade, while the Edison Electric Institute's workforce programme (opens in a new tab) tracks the equivalent replacement problem across US investor-owned utilities. A roadshow arrives into the middle of an argument the workforce is already having about who will still be here in ten years.
Designing the route: what to present at each kind of stop
The route card. Every stop type in a utility estate — who is in the room, what the executive actually says there, the question that will come, and what must be captured before the car leaves.
What an executive presents changes completely by stop type, and the clearest tell of a tour that will not land is one deck shown in all of them. A control room, a wind farm and a contact centre have different accountability structures, clocks, measures of competence and histories with automation. The route card below is how we build a tour with operators: one row per stop type, every column filled in before the first date is booked.
| Stop | Who is in the room | What the executive actually presents | The question you will be asked | What to capture | Cadence |
|---|---|---|---|---|---|
| National / regional control centre | TSO or DSO shift engineers on the clock, shift manager, EMS or ADMS support, often a union rep | Advisory versus authority. Where a recommendation appears in the EMS or ADMS, what it may do, what it may never do, and how alarm load changes | "When your model and my engineer disagree at 03:00, whose name is on the incident report?" | Which decisions are already effectively automated; the two applications operators fight with; where ETR and switching advice go wrong | Twice a year, per shift pattern |
| Thermal, hydro or nuclear station | Station manager, O&M supervisors, control-room staff, outage planning | Condition data and the maintenance backlog. What AI does to outage scoping and spares, and what it does not touch inside the safety case | "Does this change the safety case, and who re-approves it?" | Condition signals that exist but are unused; backlog items that recur; where CMMS data quality blocks anything useful | Annually, aligned to the outage cycle |
| Wind or solar site / renewables O&M | Site technicians, regional O&M lead, contracted service partners | Scheduling and access. Forecasting, turbine or inverter fault triage, and how a recommendation changes what the crew is dispatched to do | "Whose data is it — ours or the OEM's — and can we act on it?" | OEM contract constraints on data; the false-callout rate; access and weather-window losses | Annually, plus after any OEM contract change |
| Network depot / regional operations | Supervisors, planners, control-room liaison, crews between work packs | Work packs and driving. How AI changes what is on the pack, how faults are prioritised, and what happens to GIS and CMMS record quality | "Are you going to use this to time us?" | Wasted journeys; where GIS records disagree with reality; repeat visits and the reasons behind them | Quarterly, at muster |
| Field crew muster / linesmen briefing | Craft staff at start of shift, first-line supervisor, safety representative | Fifteen minutes, standing, no slides. What changes on the device in their hand, what does not change about switching and isolation, and who they raise a problem to | "Is this replacing my job?" | Every workaround they describe; the tools they already ignore and why | Quarterly, in the muster slot, never as an extra meeting |
| Customer operations / contact centre | Agents on the floor, team leaders, workforce management, quality assurance | Handle time and the script. What is drafted for the agent, what the agent must still decide, and what happens to quality assurance and vulnerable-customer handling | "If the AI writes the answer, what am I being measured on?" | The top three call drivers; where agents override the suggested answer; the complaints that come back | Quarterly, across shift patterns |
| Metering, billing and back office | AMI operations, revenue protection, billing exceptions, data teams | Exception volume. Which exception queues shrink, which grow, and what the AMI data quality problem is costing today | "Whose number is right when the model and the meter disagree?" | Exception backlog by type; manual reconciliation hours; the AMI estate's known gaps | Twice a year |
Two structural rules make the card work. The executive presents for twenty minutes and listens for forty — reverse those and the stop produces the questions people ask when they are being managed. And every stop needs three people from the programme, not one: the executive, a scribe whose only job is the capture cards, and a delivery engineer who can say on the spot whether an idea touches a system the operator controls or one a vendor does.
Which stop goes first
Plot each candidate site on data readiness against workforce trust. Sequencing is the most consequential decision in the route, because the first stop's story travels through the estate faster than any internal communication ever will.
The credibility test — go second
- Data is fine; the relationship is not
- Take an operator from the opening site into the room with you
- Expect the job question first, not last
Open the tour here
- Good data and a workforce that expects to be consulted
- Produces the first real reference the rest of the estate believes
- Ask this site to co-present at the next two stops
Not a launch site
- Neither the data nor the relationship supports a commitment
- A stop here manufactures a story about broken promises
- Fix one axis before booking a date
A listening stop
- Willing people, unwired systems
- Capture heavily, promise nothing on timing
- The highest-yield stop for finding data problems nobody logged
Book against the shift roster, not the executive diary
A stop at 10:00 on a Tuesday reaches the day shift and nobody else — in a control centre, roughly a fifth of the people who will live with the outcome. Run the same stop twice across a shift overlap, or at handover. That scheduling cost is the price of not having to explain later why the night shift heard about the programme from the day shift.
Fifteen minutes at a muster, standing, no slides
Field crews are the hardest audience to reach and the easiest to lose. The muster slot already exists, already has a rhythm, and is where safety information is delivered — exactly the frame the message needs. Adding a separate meeting for the AI programme signals that it is not real work.
Never let a vendor front a stop
A supplier presenting at a control centre converts the session into a demonstration and the room correctly reads it as a sale. Vendors can attend and answer technical questions; the framing, the accountability answer and the job question belong to the executive, because those are the things only the employer can commit to.
Put the return date in the room, out loud
Ending a stop with a stated write-back date and a stated intake decision date converts the visit from an event into the first step of a process. It is also the only commitment an executive can make at a stop and guarantee to keep, which is why it should replace every commitment about outcomes.
"Is this replacing my job?" — answering the question honestly
The question a 25-year control-room operator asks, the four answers that fail, and the role-by-role account that is the only thing that survives contact with the room.
The only survivable answer to "is this replacing my job?" is a specific description of what changes in that person's day, given by someone who has actually looked at their day. Every general reassurance fails, and it fails in a particular way: the room does not conclude that the executive is lying, it concludes that the executive does not know — which is worse, because it means the decision was taken without anyone examining the work.
"Nobody is going to lose their job" — the promise you cannot keep
It is heard as a corporate formula, remembered exactly, and falsified by the first unrelated restructure — at which point every other statement made on the tour is reclassified as spin. A promise about employment five years out is not within an operations director's gift, and everyone in the room knows it.
"AI will free you up for higher-value work" — the phrase that ends the conversation
It is heard as a euphemism, because in most people's experience the higher-value work was never defined and the freed-up time became a headcount reduction. If there genuinely is higher-value work, name it: which task, whose backlog, starting when. If you cannot name it, do not use the phrase.
"It's just a tool, like any other system" — under-promising and still failing
Operators have watched a decade of systems arrive with that description and change the job substantially. It also concedes the wrong point: if it changes nothing, why did an executive travel here to describe it? The honest version — it does change the work, and here is exactly how — is stronger and more credible.
Silence — declining to raise it at all
Programmes often avoid the subject in case it creates anxiety. The anxiety already exists and the vacuum is being filled from outside — trade press, contractors, social media. Not raising it means the answer is written by somebody else. Invite it in the first five minutes and answer it before the operational content.
What works instead is a role-by-role account: for each job in the room, what the AI actually does, what remains a human decision, what genuinely changes about the day, and what the person should say to a colleague who was not there. That last column matters more than it looks — most of the estate hears about this stop second-hand, and giving the attendee a sentence they can repeat accurately is the highest-leverage minute of the session.
| Role | What the AI does | What stays a human decision | What actually changes in the day | What they can tell a colleague |
|---|---|---|---|---|
| Control-room operator (20+ years) | Ranks alarms, proposes switching sequences, drafts an ETR from OMS and weather inputs | Every switching instruction, every safety document, every decision to accept or reject the proposal | Fewer nuisance alarms per shift; the ETR arrives pre-drafted; more time on the events that matter | "It sorts the alarm list and drafts the restoration time. I still make the call and sign the paperwork." |
| Field technician / linesman | Prioritises the work pack, flags likely fault causes, matches asset records to what is actually on the pole | The isolation, the test, the repair method, and whether the job is safe to start | Fewer wasted journeys; better information on the device before leaving the depot; more accurate as-found records | "It gives the pack a better order and better information. It does not touch how we isolate or how we work." |
| Outage planner / scheduler | Proposes outage windows against condition data, weather and resource availability | The plan that is issued, the risk acceptance and the customer commitments | The first-pass schedule takes an hour instead of two days; the argument moves from building the plan to judging it | "It builds the first draft of the schedule. Choosing the plan is still ours." |
| Contact-centre agent | Drafts responses, retrieves account and outage context, suggests the next best action | What is actually said to the customer, and every vulnerable-customer judgement | Less typing and searching; more calls where the hard part is the conversation rather than the system | "It writes a first draft and pulls the account detail up. I decide what to send." |
| Revenue protection / metering analyst | Scores AMI anomalies and ranks the investigation queue | Whether a case is opened, what is alleged and what evidence is used | The queue is ranked by likelihood rather than by age; fewer dead-end visits | "It sorts the queue. The case is still built by a person." |
| First-line supervisor | Summarises shift performance, surfaces recurring issues across crews and sites | Every people decision, every allocation, every conversation about performance | Less time assembling the daily picture; recurring problems visible earlier | "It puts the daily picture together. It has nothing to do with how anyone is managed." |
Two commitments make the account credible rather than merely well-phrased. First, a written statement that AI outputs will not be used as an input to individual performance management — it costs nothing operationally and removes the fear doing most of the work behind the question. Second, naming the reskilling route and who pays for it, because where Eurelectric's president describes losing a third of his most experienced technicians to retirement within a decade (opens in a new tab), the honest framing is that AI is arriving into a workforce gap the sector already cannot fill.
Baby boomers are reaching retirement age so over the next decade I'm losing one-third of my most experienced technicians.
Unions, works councils and the weight the word 'automation' carries
Why employee representatives belong in the route plan rather than in the escalation path — and why safety-culture framing is not optional in this industry.
Employee representatives belong in the route plan because in a collectively represented workforce they are the channel through which the estate will interpret the programme whether or not you involve them. A union or works-council body that first hears about an AI programme from a member who attended a stop has been placed in exactly the position consultation frameworks exist to prevent. The sequencing rule is simple and rarely followed: representatives are briefed before the first stop, not after the first complaint.
The engagement sequence that keeps the route intact
Brief representatives before the route is published
Share the strategy, the route and the role-by-role account with national and site representatives — and with the European Works Council (opens in a new tab) where one exists — before any date is announced internally. The ask is not agreement; it is that nobody learns about the programme from a member. Expect the first meeting to be about scope and measurement rather than technology.
Agree what AI output will never be used for, in writing
The two commitments that matter are that AI-derived measures will not feed individual performance management and that no safety rule is changed by a recommendation. Both are cheap to give and both remove a specific, reasonable fear. Having them in writing before the tour lets the executive state them at every stop as agreed policy rather than as personal assurance.
Invite a representative into the room at each stop
A representative present at the stop hears the same answer as the members, which prevents the second-hand version diverging. It also means the hard question gets asked properly rather than in a corridor afterwards, and it puts the representative in a position to confirm rather than to investigate.
Route captured items about work organisation through consultation
Some ideas raised at a stop change how work is allocated or measured — crew dispatch, shift patterns, quality assurance sampling. Those need a consultation path, not just an intake slot. Marking them at capture avoids the situation where a funded use case has to be stopped six weeks in.
Publish dispositions to representatives and sites on the same day
Same document, same day. A representative who receives the disposition list at the same moment as the depot can answer a member's question immediately, which turns the relationship from adversarial to procedural. Delay by a week and every disposition becomes a request for information.
Review the arrangement annually, on the record
An annual joint review of what was deployed, what changed about the work and whether the written commitments held is a small standing cost that makes the arrangement durable. It is also the artefact that survives a change of executive, which is the most common cause of an engagement approach quietly reverting.
The second thing utilities cannot skip is safety-culture framing, because here "automation" is not an abstract word. The workforce's frame of reference includes protection schemes that mis-operated, remote switching that removed a person from a decision, meter reading that disappeared as a role, and control rooms that were consolidated and closed. It also includes a safety regime — permits, isolation, lock-out and tag-out, competence sign-off — that exists because automated systems have historically been trusted where they should not have been. "AI will make operations safer" is heard against that history.
Name the boundary before you name the benefit
Start with what the system will never do: it does not issue switching instructions, authorise a permit, change an isolation standard or sign competence. Then describe what it does. Reversing that order costs the rest of the session, because the room spends it waiting for the boundary statement instead of listening.
Use the safety regime's own vocabulary
Advisory versus authority, defence in depth, the hierarchy of control, competence and authorisation. These are precise terms with agreed meanings, and using them correctly signals that the programme has been examined by someone who understands the regime. Using them incorrectly does the opposite, so brief the executive properly or leave them out.
Treat the reliability standards as part of the framing
Where reliability and protection standards apply — NERC's standards regime (opens in a new tab) in North America, national safety regulation such as GB's Health and Safety Executive (opens in a new tab) frameworks elsewhere — say which standard governs the decision being discussed and confirm whether the AI sits outside the compliance boundary or inside it with evidence. Control-room staff know these regimes better than most head-office teams.
Do not present AI as a response to the workforce gap alone
"We need this because we cannot recruit" is true in many utilities and is heard by a 25-year technician as an argument for not replacing them when they leave. Pair it explicitly with the reskilling and recruitment commitment. Sector workforce programmes (opens in a new tab) and trade-union bodies (opens in a new tab) are both making the recruitment argument publicly, which makes it easy to align with rather than appear to exploit.
