The first bill after the funeral

A widow opens the first energy bill since her husband died. The account is in credit — yet the Direct Debit is going up. She doesn't understand how that can be right. She is grieving, confused, and frightened of the winter ahead. She calls her energy supplier.

The agent who answers must get this call right twice over: procedurally right, under Ofgem's rules on vulnerability and Direct Debit handling — and humanly right, for a customer at one of the lowest moments of her life.

Calls like this reach OVO's contact centres every day. Until this programme, the first time a new agent faced one, the customer was real.

The cost-of-living crisis has turned routine energy calls into high-stakes conversations, and the regulatory bar has hardened with them. Ofgem's 2024 enforcement action against suppliers explicitly cited staff capability and complaint handling as root causes of consumer harm — moving the standard from training completion to demonstrable capability.

Inside OVO, the gap was structural. Onboarding moved new agents through theory, systems and shadowing, but offered no way to practise a real conversation before the first live call. That drained tenured agents' time, exposed customers to unpractised agents, and carried compliance risk.

The traditional alternatives could not close it. E-learning proves attendance, not readiness. Facilitated role-play was actively unpopular with agents, inconsistent between facilitators, and unscalable across continuous, high-volume hiring. Only simulation lets an agent feel the pressure of a distressed, confused customer — the hesitation, the emotion, the pushback — with zero risk to a real one. That was the rationale: move the first hard conversation off the customer and onto an AI.

Real Talk Studio built OVO a library of browser-based AI voice simulations replicating its highest-volume, highest-risk calls: Direct Debit increases, vulnerability disclosures, affordability distress. The agent speaks; a lifelike AI customer listens and responds in real time — hesitating, pushing back, softening only when handled well. In the flagship scenario, Direct Debit Increase Confusion, the agent faces Helen Carter: a customer whose payment has jumped while her account is in credit, and who wants answers.

Every scenario is mapped to four explicit, individually scored objectives drawn from OVO's own policies and QA framework — each tied to the business measure it was designed to move.

1. Acknowledge and validate the customer's concern before moving to process. → CSAT up

2. Explain the Direct Debit calculation accurately, clearly, and in plain language. → DD adherence up

3. Recognise and respond to vulnerability cues — bereavement, distress, confusion. → Transfers down

4. Agree appropriate, compliant next steps the customer actually understands. → Reopened cases down

A live mood meter shows how the conversation is landing, and objectives tick off as they are met — no script, no multiple choice, no "next" button. Each session ends with a score out of 100, the full timestamped transcript, and line-by-line coach notes filtered to what to fix, with one concrete behavioural tip for the next attempt. Feedback lands in minutes, not weeks, and agents can practise again immediately.

Plenty of tools digitise role-play. This simulates the thing role-play can never reach: the emotional weather of the call itself.

Emotion and pressure, simulated. Every character carries a mood, composure level and backstory — tearful, apprehensive, frustrated, quietly composed — and shifts state live in response to how the agent handles them. The same call plays out differently against different people, stress-testing empathy and human skill, not just process recall.

Unfakeable practice. You cannot click through a live conversation. Every session records what an agent actually said under pressure — competence data, not completion data — and scenarios can be built from real calls in hours, not weeks.

Compassion, scored. Feedback assesses policy adherence and human quality together: a missed acknowledgment of distress is flagged as precisely as a mis-explained calculation. No traditional method does both, consistently, for every agent.

OVO designed the pilot as a controlled benchmark. One defined cohort — the Direct Debit Tiger Team — practised on a single, high-complaint call type over a 12-week window while the rest of the floor continued business as usual. That design meant movement in complaints, QA effort and confidence could be read directly against normal performance on the very same calls: a like-for-like baseline, not a before-and-after guess.

On the business measures reported by OVO across March → April, Direct Debit process adherence rose from 65.2% to 96.6%, CSAT from 75.9% to 79.9%, and reopened cases fell from 5.3% to 2.0% — more than halved. Every supporting metric moved the right way too: same-day repeat calls fell 5.4% → 4.4%, transfers fell 6% → 4.4%, and process compliance rose 77.7% → 79.6%.

The platform's own scored session data pointed the same way. Average scores climbed 45 points from first attempt to best attempt (37 → 82/100), the flagship scenario improved by 172% (32 → 87/100), and all four scenarios showed the same climb. Practice quality averaged 4.5/5, and 100% of pilot users rated the tool highly.

The return is measured in outcomes, not just cost saved. The real question is not what practice costs — it's what an unpractised conversation costs: a distressed customer mishandled, a case reopened, a complaint raised, a regulatory exposure created.

Against that, the pilot's value lands where OVO's customers feel it. Vulnerable customers are met by agents who have already practised the hardest version of the call. The experience is measurably better — CSAT up 4 points, reopened cases more than halved, fewer transfers and fewer repeat calls. And regulatory exposure is reduced, with Direct Debit adherence at 97%.

The comparison with the alternatives is then stark: e-learning could not have produced these outcomes at any price, and facilitated role-play could not have produced them at scale. Simulation delivered them with no facilitators, no scheduling, and no agents pulled off the phones.

For OVO's agents: confidence with evidence behind it. Every practice call ends with a score, a transcript and one concrete thing to change — so skill builds visibly, attempt over attempt, before a real customer is ever on the line. Agents step onto the phones having already survived the hardest version of the call.

For OVO's customers: the bereaved customer confused by a rising bill no longer meets an agent learning on her call. She meets someone who has handled it before — who acknowledges her loss before explaining the calculation, and gets the compliant answer right first time. Compassion and compliance stop being a trade-off.

On these results, simulation is moving from pilot to an embedded part of how OVO makes agents call-ready, with impact tracked against attrition, first-contact resolution, handle time and customer satisfaction.