ARTICLE
Most AI projects in utilities never make it past the pilot stage. That was the recurring message across the AI and digital sessions at this 2026’s Future of Utilities Summit, where speakers from energy retailers, water companies and technology vendors described a sector that has proved AI can work, but is still struggling to make it stick.
Jo Page, Data Director at E.ON Next, put a figure on the problem: only 26% of AI use cases reach production. E.ON Next’s own deployment is extensive – AI agents named Claire and David handle routine customer service queries, an internal tool called Wiki Whisperer streamlines knowledge access for staff, and the company processes 5,500 billion smart meter data points to power personalised recommendations. Page said companies with strong data foundations – structured governance, clear ownership, documented processes – were 2.2 times more likely to get AI projects to scale.
Marcus Gardiner, client director at Softwire, made a similar argument from a different angle, warning against what he called “zombie projects” stuck between proof of concept and proof of value. He said utilities should resist importing the “move fast and break things” mindset from consumer tech, given the consequences of getting it wrong on critical infrastructure. His phrase – “foundations make tech smooth, and smooth is fast” – was echoed by other speakers later in the day.
Stig Martin Fiska of Cognizant and Patrick Feast of Databricks framed AI scaling as a people problem rather than a technology one. They cited a study finding that 97% of leaders with unified data architectures beat their own AI business case targets, but argued the harder task is cultural: building what they called a “builder’s gene” among staff, empowering employees to become AI trainers and product owners rather than passive recipients of tools handed down by IT.
OVO‘s Andrew Mohammed described the company’s shift “from chatbot to operator,” with agentic AI now handling complaints triage, billing queries and refunds policy lookups, saving an estimated half a million hours of staff time and lifting customer satisfaction scores by 20 percentage points. But Mohammed also gave a less flattering example: a customer named Sarah who spent six months and twelve separate interactions trying to resolve a single issue, despite OVO’s extensive AI deployment. His conclusion was that resolution, not deflection, is what matters – a 33% cut in cost-per-contact only counts if the AI is actually fixing problems rather than routing them elsewhere.
Future of Utilities’ broadcast on AI in CX “Towards pre-emptive CX: the next phase of utilities customer engagement” pushed the conversation further, asking how many AI agents is too many. Agents are more effective when they are specialised – but Edward de Mas Latrie, Director AI & Agentic at NiCE, was asked at what point specialisation tips into a problem: “It’s at the point when you start to lose control.” He also pressed on a more basic measure of success – not whether an AI agent responded, but whether it actually closed the customer’s issue. Emma Fletcher, Innovation Director at Octopus Energy Homes, made the case that AI has to support staff rather than add to their workload: “If you have happy employees, you have happy customers. That AI needs to help and assist – not be another system or another app to use.” Abdul Khaled, Director of Transformation at Vandebron, argued utilities should work backwards from customer outcomes rather than starting with efficiency targets and dealing with the customer impact afterwards, pointing to Klarna’s reversal of its AI-driven staff cuts as a cautionary example of what happens when that order is flipped. Audience questions from the session ranged from how to balance efficiency against customer experience, to how CX should be benchmarked when a supplier’s customer base is a handful of large industrial accounts rather than millions of small ones. It’s important to remember that the “production” problem in AI isn’t only about getting agents live, but about agreeing what success actually looks like once they are.
Cybersecurity added urgency to the digital conversation. Alistair Wildman of Palo Alto Networks said attackers can now exfiltrate petabytes of data in 25 minutes, down from 48 hours a year ago, and that his company is responding to twice as many major incidents as it did twelve months ago. He referenced early access to Anthropic‘s Mythos and Fable models through the Glasswing partnership, used by 11 organisations including Palo Alto, which consumed a billion tokens in a single day. A separate Unit 42 briefing, referenced in the same session, described an agentic AI tool that deleted 26,000 customer records while attempting to improve efficiency. Wildman said identity, not malware, is now the primary attack vector, with roughly 100 non-human “agent identities” for every human one inside a typical enterprise.
Jennifer Rigby of Southern Water and Iain Dougan of SSEN Transmission discussed what digital transformation looks like in practice at board level. Rigby said her team’s roadmap was built on 250 stakeholder conversations; Dougan described SSEN Transmission’s shift from a Microsoft 365 baseline toward £5 billion in annual capex, en route to £29 billion of investment by 2032. In the panel that followed, Marzia Zafar of Ofgem rated the UK’s overall digital maturity at 7.5 out of 10, behind Australia and Estonia, while Matt Webb of South East Water noted the unevenness within the sector itself – Northumbrian Water and Anglian Water cited as ahead of the pack, with many smaller or more fragmented organisations lagging behind.
Across the sessions, the consistent thread was that data quality, governance and board-level ownership – not the AI models themselves – are what determine whether a pilot survives contact with a live operation.
For the epicentre of European energy digitalisation, join
Future of Utilities Smart Energy
(2-3 November, London).
Or join us next year at
Future of Utilities Summit 2027
(16-17 June, London).
Subscribe to our newsletter to receive news, insights and special offers.