Octopus Energy
genAI customer agent (drafting assistance + autonomous agent)
At Octopus Energy, the genAI tool Magic Ink assists the drafting of around 35% of customer emails and has summarized more than 6.2 million calls, with the autonomous agent Arlo handling 8,000 emails a week at 76% satisfaction versus 72% for agents.
Key points
- GenAI assistance for drafting email replies, under human supervision.
- Magic Ink tool integrated into the Kraken platform, plus the autonomous agent Arlo.
- About 35% of emails drafted with assistance; satisfaction 80% versus 65%.
- Evidence B, confirmed status in production at group scale.
Objective
Absorb the volume of customer emails without scaling headcount, holding a satisfaction level at least equal to that of human agents and reducing the cost per case.
The deployment
Magic Ink is the genAI tool integrated into Octopus's Kraken platform. When an agent opens a customer email, the tool reads the full account history and product data, then drafts a reply in the agent's tone. It annotates verified facts with their source and highlights what it could not verify. The agent reviews, adjusts if needed, and sends: the human remains the sender. About 35% of emails are written with this assistance, and nearly one in three messages generated by Magic Ink is sent with little or no editing. Octopus then tested Arlo, an agent that answers simple requests on its own (tariff renewal, direct-debit date, account details), excluding vulnerable customers and complex cases. During the trial, Arlo handled about 8,000 emails a week, or 4% of the customer emails received in the UK.
Results Proof B
Quantified platform case study (techUK/Kraken) with precise volumes, confirmed by an official brand release and by the national press (The Times relayed by City AM) naming the CEO. Consistent sources.
How it works
Documented architectureThe stack in detail
- plateforme Kraken Octopus Energy Group's customer operations platform (account, billing, interactions), the base that provides the unified history the genAI relies on.
- outil Magic Ink GenAI tool integrated into Kraken: it drafts replies in the agent's tone, annotates verified facts with their source, and highlights unverified text; it also summarizes calls.
- llm LLM type GPT (modele exact non publie) GPT-type text-generation models used by Magic Ink; Octopus does not publish the exact version or provider.
- outil Arlo Autonomous agent tested by Octopus on simple requests (tariffs, direct debits, account details), with exclusion of vulnerable customers and human escalation.
How it runs, concretely
For ops teams-
1Reception and contextualization AI
The email arrives in Kraken, which gathers the account history and associated product data.
-
2Draft writing AI
Magic Ink generates a reply in the agent's tone, annotates verified facts with their source, and highlights unverified text.
-
3Review and send customer service
The agent checks, corrects if necessary, and sends. The agent is the sender, not the machine.
-
4Autonomous handling of simple requests AI
Within a framed scope (tariffs, direct debits, account details), the Arlo agent answers on its own and hands off to a human for any sensitive or vulnerable case.
The full account history (contract, bills, past exchanges) and the post-reply satisfaction score. Without access to this unified history, the draft loses relevance and the share of messages sent without editing collapses.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- unified customer history (contract, billing, exchanges)
- corpus of past emails to calibrate tone
- product reference accessible in real time
Org prerequisites
- customer service willing to supervise rather than write from scratch
- quality-control process on AI replies
- framework for excluding vulnerable cases
Possible stack
- enterprise LLM (GPT-type) grounded on in-house data
- CRM/unified service platform
- fact-verification layer with source citation
The plan, step by step
- Step 1Unify the customer history (contract, billing, past exchanges) in a single view accessible to the system and to agents.Deliverable: 360 customer record queryable in real time
- Step 2Connect an LLM in supervised-draft mode on the email channel, with annotation of verified facts and highlighting of unverified text.Deliverable: Drafting assistant piloted on an email queue
- Step 3Measure the usage rate, the share of drafts sent without editing, and post-reply satisfaction; define exclusion rules (vulnerable cases, disputes).Deliverable: Quality dashboard + validated exclusion scope
- Step 4Extend the assistance to all agents and add automatic call summarization.Deliverable: General rollout with continuous quality control
- Step 5Test an autonomous agent on a bounded scope of simple requests, with systematic human escalation and a satisfaction comparison vs agents.Deliverable: Documented controlled trial, extension decision
First step: Unify the customer history in a single view and connect an LLM in supervised-draft mode on one channel (email).
The same pattern, proven in other industries
ADT
genAI customer agent
Air India
genAI customer agent (automated customer service)
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Sources
- S1 AI Adoption Case Study: Kraken's generative AI tool for customer service helping Octopus Energy Interested party archive pending
- S2 Octopus Energy's AI trial wins customer approval Primary archive pending
- S3 AI doing the work of over 200 people at Octopus, chief executive says Established press archive pending
- S4 AI 'now doing work of 250 people three months after launch', Octopus Energy boss reveals Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.