To answer one tricky question, an agent put the customer on hold and dug through four or five disconnected systems by hand. The awkward silences piled up, the answers didn't match, and satisfaction was sliding. We put the answer on the agent's screen instead.
In a centre handling thousands of calls a day, a customer's time on hold is everything — and here satisfaction was dropping and wait times were climbing.
The cause was that agents worked against isolated islands of information. To answer one complex question, an agent had to put the caller on hold, open four or five different systems, and hunt by hand through procedures, price lists and old policies. The result was long waits and awkward silences on the line.
Because of the pressure and the scattered information, different agents gave contradictory answers or leaned on procedures that had expired — which bred repeat complaints and churn. And onboarding a new agent took weeks, just to learn how to navigate between the organisation's old systems.
Grounded only on the company's own knowledge, surfaced live, with a guardrail against guesses.
We connected the knowledge bases, policies, procedures and CRM into a single secure retrieval layer that answers only from the organisation's own, current information — not the open internet.
A smart assistant sits right in the agent's workspace. It reads or listens to the customer's request in real time, pulls the exact answer from the up-to-date procedure in seconds, and shows it on screen — no hold, no hunting.
Strict logic stops the model from “hallucinating” an answer. It is bound to surface only information backed by an organisational source, with access to sensitive data scoped per permission.
Because every agent draws from the same current source, callers stop getting contradictory answers or expired policies — and the repeat calls that used to follow simply stop coming.
A new hire no longer has to memorise how to navigate a maze of legacy systems before they can help — the assistant carries the knowledge, so ramp time drops from weeks.
The assistant proposes the answer and its source; the agent reads, judges and speaks. The relationship — and the final word — stays with a person.
The hold music stopped, and the numbers moved.
Customer satisfaction jumped from 56% to about 77% in the first quarter of the rollout, as agents cut the time spent searching for information mid-call by 42% and stopped putting people on hold to look things up.
Total time to resolve a query fell by about 30%, with a sharp drop in the number of customers who had to call back a second time about the same problem — the repeat-contact tax, largely gone.
The speed came with trust, because the assistant is bound to the company's own current sources and refuses to guess — the same grounded, human-in-control approach behind our airline support-agent work, where anything uncertain is held for a person. And it pairs with our after-call automation: one assistant helps during the call, the other handles the write-up after it.
Reef TRH