AI Voice Agent Development in Paris

Energy / Residential Utilities

A major French energy utility (28 million residential customers, 35 million annual inbound calls, 8,500 contact centre agents across 12 centres in France) faced customer service and energy transition communication challenges. (1) Volume pressure: 35 million annual calls at 6.5-minute average handle time consumed 3.8 million agent-hours. 78% of calls were routine (relevé de compteur, facturation enquiries, changement de tarif, mensualisation setup, déménagement/emménagement). Despite digitisation efforts (Mon Espace Client online portal), call volume remained high — many customers preferred telephone for energy matters, particularly older customers and those with complex situations. (2) Energy transition enquiries: the French energy transition (transition énergétique) generated millions of additional enquiries — MaPrimeRénov' subsidy eligibility, pompe à chaleur installation guidance, panneaux solaires autoconsommation, and isolation thermique options. These questions were complex and evolving (subsidy rules changed frequently) — agents required regular retraining, and information accuracy was inconsistent. (3) Seasonal peaks: winter months (November-February) saw 50-70% volume increases — heating-related enquiries, high-consumption billing disputes, and payment difficulty discussions. The utility hired 2,000 seasonal agents at EUR 22M annual cost — but training quality for seasonal agents was lower, and customer satisfaction during winter dropped significantly. (4) Vulnerable customer management: French energy regulation required identification and support of customers in précarité énergétique (energy poverty). 3.4 million of the utility's customers qualified for social tariffs or payment support — but identifying vulnerability during phone interactions required training and attention that was difficult to maintain during peak periods. (5) Payment difficulty conversations: 1.8 million calls annually involved payment difficulty — customers unable to pay their energy bills. These conversations required empathy, knowledge of social support options (chèque énergie, FSL — Fonds de solidarité pour le logement, échéancier de paiement), and compliance with French consumer protection rules for energy debt management. These were emotionally difficult calls that contributed to agent burnout.

We built a residential energy service voice agent over 14 weeks, integrated with the utility's billing, metering, and customer systems. (1) Routine enquiry handling: the voice agent handled the 78% of routine calls — compteur readings (Linky smart meter data retrieval and manual reading submission), facturation enquiries (bill explanation, consumption analysis, payment status), tarif changes (heures creuses/heures pleines optimisation, switching between tariff options), mensualisation (monthly payment plan setup and adjustment based on consumption forecasts), and déménagement (address changes, contract transfer, and new customer enrollment). Natural French: the agent communicated in clear, professional French — using energy-specific vocabulary naturally (puissance souscrite, index de consommation, tarif réglementé) while remaining accessible to non-technical customers. (2) Energy transition advisory: the agent provided personalised energy transition guidance — assessing the customer's housing type, current energy consumption, and geographic location to recommend relevant measures (pompe à chaleur, isolation, panneaux solaires), explaining MaPrimeRénov' eligibility and estimated subsidy amounts, connecting customers with certified artisans (RGE — Reconnu Garant de l'Environnement), and scheduling free energy diagnostic visits (bilan thermique). Information accuracy: the agent was updated whenever subsidy rules changed — ensuring customers always received current information. (3) Vulnerability detection and support: the voice agent incorporated continuous vulnerability assessment — monitoring for indicators of précarité énergétique (payment difficulty mentions, expressed anxiety about bills, references to choosing between heating and eating), cognitive vulnerability (confusion, difficulty understanding information), and accessibility needs (hearing difficulty, language barriers). When vulnerability was detected: the agent adapted its communication style (simpler language, slower pace, more empathetic tone), proactively offered relevant social support information (chèque énergie application, FSL referral, échéancier de paiement options), and flagged the account for follow-up by the utility's social support team. (4) Payment difficulty management: for customers expressing difficulty paying their bills, the agent: assessed the situation with empathy (acknowledging the difficulty without judgment), explained available options (échéancier de paiement, chèque énergie, FSL, médiation — CNIL mediation services), set up payment plans within the agent's authority (plans up to 12 months), and escalated complex cases to social support specialists with full context. The agent followed French consumer protection rules for energy debt — including the trêve hivernale (winter moratorium on disconnection from November to March). (5) Winter surge management: the voice agent scaled automatically during winter peaks — handling 50-70% more volume without degradation. Winter-specific capabilities: heating troubleshooting guidance (for customers with electric heating issues), high-consumption bill explanation (contextualising winter bills against temperature data and historical consumption), and emergency heating assistance (connecting vulnerable customers with emergency social services when appropriate).

Call handling: voice agent handled 71% of calls without human intervention — 24.9 million of 35 million annual calls. Hold times: eliminated for voice-agent-handled calls. Peak-period agent queue: winter wait from 35 minutes to under 5 minutes. Seasonal staffing: 2,000 seasonal positions eliminated — EUR 22M annual savings. Energy transition: 1.2 million energy transition advisory conversations in the first year — generating 180,000 qualified leads for energy renovation services (pompe à chaleur, isolation, panneaux solaires). Revenue from referral commissions and associated energy contract upgrades: EUR 14.5M. Vulnerability detection: identification rate improved from 28% (human agents) to 68% (voice agent continuous monitoring). 340,000 additional vulnerable customers identified and connected to support programmes — reducing energy debt write-offs by EUR 8.2M. Payment difficulty resolution: 72% of payment difficulty calls resolved by the voice agent (payment plan setup, chèque énergie guidance) — reducing escalations to social support specialists by 45%. Payment plan compliance rate: 78% (from 62% with human-arranged plans — the agent's systematic approach produced better-structured plans). Customer satisfaction: NPS improved from +8 to +28 — driven by eliminated hold times, accurate energy transition guidance, and empathetic payment difficulty handling. Agent redeployment: 3,200 agents redeployed from routine handling to energy advisory, vulnerability support, complex case management, and commercial activities. Annual value: EUR 22M (seasonal staffing elimination), EUR 14.5M (energy transition revenue), EUR 12.8M (routine call agent redeployment to higher-value roles), EUR 8.2M (reduced energy debt from vulnerability detection), EUR 4.5M (payment difficulty — improved plan compliance and reduced write-offs) = EUR 62.0M. Development cost: EUR 380,000. Payback: under 3 days.