A working picture of what Lorikeet-powered support looks like inside Fuse: in the app, on the web, on voice, and reaching out before the bill lands.
Plugs stays as the brand. The team keeps its inbox. What changes is the agent behind it: authored in natural language by ops, guardrailed by default, and covered by simulation.
Voice, simulations, guardrails, and proactive outbound. The four things Fuse is looking at building next, shipped from day one, without adding engineering headcount.
Plugs is already at 70–80% containment. The value is what's above that: proactive reach-out on bill day, meter-read voice calls, onboarding recovery. Support that becomes part of the Fuse product.
Same entry point customers use today, contextual to the account (tariff, meter, EV, solar). Runs the switching and refund tools directly. Hands off to a human cleanly, with a full internal note.
Handles switching questions on the marketing site. Qualifies leads before signup. Answers "what's my next bill" for signed-in customers. Same concierge, different context.
Bill-day preview. Meter-read reminders by voice for repeat non-responders. Onboarding recovery. Complaint follow-up. Triggered from Fuse's system of record, driven by Lorikeet.
24 hours before bill day, Plugs reaches customers flagged by the context engine: unusual usage, first bill on a new tariff, big EV month. Explains the number before it lands.
The meters queue has higher escalation because a missed input compounds into a wrong bill. Plugs reaches out on schedule, walks the customer through submission, retries if it fails. Voice for repeat non-responders.
"Sign up in under 3 mins" leaks people. Plugs reaches abandoned signups with the specific step they got stuck on, answers the question, recovers the conversion.
Streaming ASR tuned for tariff and meter vocabulary. Barge-in for interruptions. Voicemail hits a deterministic subworkflow instead of the LLM freestyling. A guardrail suppresses filler so the call sounds like an agent, not a bot narrating.
Every workflow change runs against a bank of representative tickets before it ships. Regression coverage that Fuse would otherwise build in Vertex from scratch.
Pre-processing catches vulnerable customers, sensitive intents, PII. Post-processing catches unauthorised promises (refunds, discounts, policy statements) before the customer sees them.
Workflows are natural-language documents in the web app, not code. Support and ops edit them directly. Engineers only get involved when new tools need building.
Tickets that take five days (complaints, backend malfunctions, change of address) get a goal-oriented concierge. It holds the goal, re-engages after silence, closes only when the goal is met.
Voice + proactive outbound, live from a Lorikeet sandbox. Amber energy customer story alongside. Same day: commercial framing sent to Evan.
Anonymised sample of ~200 tickets across switching, billing, and meters. Volume breakdown by inbox. Sent from Evan same week.
NDA already in place. DPA and Cloud Services agreement kicked off with legal on both sides so they don't gate the sandbox build.
Half-day session to confirm scope, success criteria, and DRIs. Sandbox build begins Week 1. Decision by end of Week 4.
Ticket-level audit and build-vs-buy write-up delivered to COO, CTO, CEO. Written decision to proceed to production, extend, or stop.
If the decision needs to land inside a month, we can be in a sandbox by Week 1 and in front of your leadership team by Week 4. The only thing waiting on Fuse is the tickets and the sign-off.