AI Voice Agent Development in Munich

Insurance / German Composite Insurer

A Munich-based composite insurer (EUR 15B GWP, 22 million policyholders across Kfz, Hausrat, Haftpflicht, Leben, Kranken, and Gewerbe lines, 12,000 employees including 3,200 in customer service and claims) faced customer service challenges. (1) Call volume: 28 million annual inbound calls across all lines — Schadensmeldung (claims notification), Vertragsservice (policy management), Beitragsanfragen (premium enquiries), and general Kundenberatung. 2,400 agents handled these calls. Average handle time: 7.2 minutes. Peak periods (winter — Kfz-Unfälle, storm damage; January — policy renewals) created 30-minute hold times. (2) Multi-line complexity: the insurer offered 45 product lines — each with specific data requirements for claims, different coverage structures, and different service processes. Agents needed to master all product lines — requiring 4 months of initial training and ongoing product education. Even experienced agents occasionally directed customers to wrong departments or collected incomplete claims data for unfamiliar product lines. (3) Schaden notification data quality: for Kfz-Schäden (motor claims), the FNOL required: Schadentag, Schadenort, Unfallhergang (accident description), Unfallgegner (third party details), Polizeiliche Aufnahme (police report status), Verletzungen (injuries), Werkstatt-Wünsche (repair shop preferences), and Mietwagen-Bedarf (rental car needs). Human agents captured complete data for 82% of claims — the remaining 18% required follow-up calls (Nachfassgespräche) consuming agent time and delaying claims processing. (4) Kündigungswelle during renewal: the German insurance market saw increasing switching behaviour. During January renewal season, the insurer received 180,000 Kündigung calls — retention conversations competing with routine service calls for limited agent capacity. Retention rate: 35% (of saveable customers). (5) Elderly customer segment: 34% of policyholders were over 60 — a demographic that called more frequently, preferred phone over digital, and required more patient service. Agent performance metrics (calls per hour) created implicit pressure to handle calls quickly, which conflicted with the slower pace elderly customers needed.

We built an insurance claims and service voice agent over 14 weeks, integrated with the insurer's policy administration and claims management systems. (1) Comprehensive claims notification: the voice agent handled Schadensmeldung for all 45 product lines — adapting its data collection to the specific line. For Kfz-Schäden: natural conversation collecting all required FNOL fields ("Können Sie mir bitte schildern, was passiert ist?" — then systematically collecting missing information based on the accident description). For Hausrat-Schäden: incident type determination (Einbruch, Leitungswasser, Sturm, Feuer), damage assessment, emergency needs (Notmaßnahmen — emergency repairs, temporary accommodation), and documentation guidance. For Haftpflicht-Schäden: third-party claim details, liability assessment, and witness information. Data completeness: 96% of required fields collected (from 82%) — because the agent systematically covered all fields for each specific product line without being affected by time pressure or product unfamiliarity. (2) Policy service: the voice agent handled routine Vertragsservice — Beitragsanfragen (premium explanations, payment status), Adressänderungen (address changes), Bankverbindung changes, Versicherungsschein requests, and coverage enquiries ("Bin ich gegen Glasbruch versichert?"). For complex changes (coverage modifications, Bündelverträge restructuring), the agent provided preliminary information and scheduled Berater appointments. (3) Retention during Kündigung: when customers called to cancel (Kündigung), the agent: analysed the customer's portfolio (all policies, premium total, claims history, Kundenwert — customer lifetime value), identified the reason for cancellation (Beitragshöhe — price, Leistungsunzufriedenheit — service dissatisfaction, Wettbewerbsangebot — competitive offer, Bedarfsänderung — changing needs), presented personalised retention offers (Beitragsanpassung — premium adjustment, Leistungserweiterung — coverage enhancement, Bündelrabatt — bundle discount), and processed the retention or Kündigung as appropriate. Retention rate: 45% (from 35%) — the agent's systematic analysis and consistent offer presentation outperformed agents who had less time to review customer data. (4) Senior-optimised service: for customers identified as senior (age data from policy system), the agent: spoke more slowly and clearly, used simpler vocabulary (avoiding Versicherungsfachdeutsch — insurance jargon — where possible), offered to repeat information, allowed longer conversation duration without rushing, and proactively checked whether the customer had related questions. (5) 24/7 Schadensmeldung: claims could be reported at any time — particularly important for Kfz-Unfälle (which occur at all hours) and Einbruch/Notfälle (break-ins and emergencies). The voice agent dispatched emergency services (Abschleppdienst, Notglasung, Schlüsseldienst) from the insurer's approved Dienstleister network around the clock.

Call handling: voice agent handled 67% of calls without human intervention — 18.8 million of 28 million annual calls. Hold times: eliminated for voice-agent-handled calls. Peak renewal season: wait from 30 minutes to under 4 minutes. Schadensmeldung completeness: from 82% to 96% — reducing Nachfassgespräche by 78%. Across 3.2 million annual claims: 448,000 follow-up calls eliminated, saving 2.7 million minutes of agent time. Retention: 45% rate (from 35%) during Kündigung season — 18,000 additional retained policies. Average annual premium per retained policy: EUR 1,850. Revenue protected: EUR 33.3M annually. Senior satisfaction: 4.5 out of 5 (from 3.4) — senior customers specifically valued the unhurried, patient service. 24/7 availability: 22% of Schadensmeldungen now filed outside business hours — capturing claims immediately rather than the next morning. Emergency Dienstleister dispatch: average 72 minutes 24/7 (from next-morning for after-hours claims). Agent redeployment: 1,100 of 2,400 agents redeployed to Beratung (advisory), complex claims handling, Gewerbekunden service, and retention specialisation. Annual value: EUR 33.3M (retention improvement), EUR 18.5M (agent redeployment — 1,100 agents to higher-value roles), EUR 8.2M (Schadensmeldung quality — reduced follow-up calls and faster claims processing), EUR 4.5M (24/7 emergency response — reduced claim costs from faster intervention), EUR 2.8M (senior satisfaction — reduced churn in growing demographic) = EUR 67.3M. Development cost: EUR 360,000. Payback: under 2 days.