Octopus Energy
Octopus Energy's Arlo AI answers routine customer emails
Octopus Energy built Arlo, an AI assistant that writes replies to routine customer emails on the Kraken platform. In a three-month trial it handled about 4% of UK customer emails, and the company reported a higher satisfaction score than comparable human replies.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
76%
Customer satisfaction, Arlo replies1
72%
Customer satisfaction, comparable human advisor replies1
About 8,000
Emails handled per week during the trial1
4%
Share of all UK customer emails handled1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Octopus Energy receives a large volume of customer emails, many of them routine questions with clear answers. The company wanted to answer these faster while keeping people available for harder cases.
What they built
Arlo is an AI customer service assistant that writes replies to routine emails, such as questions about tariff renewals, payment dates and account details. Octopus's tech team built it with Kraken, the company's software platform.
How they did it
Octopus set strict boundaries. Arlo never handles vulnerable customers, sensitive cases or complex complaints, which go to experienced staff. Every AI-written email is clearly labelled, and customers can ask for a human advisor at any stage. The company ran a three-month trial before planning a wider release. Deputy CTO Ashley Firth is quoted saying that too many chatbots exist to keep customers away from people.
Results
During the trial Arlo handled about 8,000 emails a week, 4% of all customer emails the UK supplier received. Octopus reported a 76% customer satisfaction score for Arlo against 72% for comparable human advisor responses.
Lessons for your project
Start with routine, low-risk enquiries such as tariff renewals and payment dates, and exclude vulnerable customers and complaints from the start.
Label every AI-written message and let customers ask for a person at any point. Octopus built both into the design.
Compare AI replies against human replies on the same satisfaction measure so the result means something.
Caveats
- Figures come from Octopus Energy's own press release; the Marketing Week article was paywalled and only its introduction was read.
- The underlying AI model is not named in the sources.
- The date shown is when the trial results were reported (July 2026). The three-month trial itself ran before that.
Sources
- Octopus Energy's AI trial wins customer approval
Octopus Energy · 2026-07-07 · Press release · octopus.energy
- How Octopus Energy uses AI to enable better customer experiences
Marketing Week · 2026-08-26 · News · marketingweek.com
Summarized in our own words from the sources above. Spotted an error? See corrections.