Overview
The bank is a regional Sparkassen-style institution headquartered in Frankfurt, operating 38 branches across Hesse with roughly 1,400 employees and a retail loan book covering consumer instalment loans, auto finance and overdraft facilities.
Like most regional banks in Germany, the institution lives and dies by its early-arrears performance. When a retail customer misses a payment, the first 14 days are decisive: a friendly, fast reminder usually recovers the account, while silence pushes it toward formal Mahnverfahren and provisioning. The bank's collections team of 11 agents could only physically dial a fraction of the daily arrears queue, and a meaningful share of customers, particularly Turkish-speaking and Arabic-speaking households across the Rhein-Main area, disengaged when reached only in German.
In early 2026, the management board decided the manual dialling model could not scale with a rising arrears book and rising regulatory expectations on documentation. They wanted a layer that could call every early-arrears customer within minutes, speak the customer's language, capture a concrete promise-to-pay, and log every interaction into the core banking system with full audit trails, escalating to a human collector only for hardship or dispute cases.
The challenge
The pre-Kallix collections funnel had three compounding failure modes. The team could not reach the full arrears queue in time. Language mismatch killed engagement with a large minority of customers. And inconsistent call logging created BaFin documentation gaps.
- Only 40% of early-arrears customers reached in the critical 14-day window. With 11 agents and a daily queue often exceeding 600 accounts, more than half of overdue customers were never contacted before the account aged into hard delinquency, where recovery rates collapse.
- German-only outreach lost Turkish- and Arabic-speaking households. Roughly 30% of the arrears book sat with multilingual households in the Rhein-Main region. Many disengaged within seconds when the call opened in formal German, then ignored follow-ups.
- Promise-to-pay commitments were rarely documented consistently. Agents captured payment promises on sticky notes and free-text fields. There was no structured promise-to-pay record, which weakened both recovery follow-through and BaFin audit readiness.
- After-hours and lunchtime arrears calls simply did not happen. Working customers were reachable mostly in evenings, exactly when the collections desk was closed. The bank was calling people when they could not pick up.
- No way to separate hardship cases from simple oversight. A customer who forgot a standing order and one in genuine financial distress got the same scripted reminder, creating both compliance risk and reputational risk.
The AI-powered solution
Kallix deployed an AI voice agent fronting the bank's early-arrears and KYC-callback queues, with native German plus Turkish and Arabic handling, branch- and product-level routing, and a hardship-detection branch that hands sensitive cases to human collectors. The full build, from discovery to production cutover, took 24 working days.
Sub-3-minute outbound on every early-arrears flag
When the core banking system flags an account 1–14 days overdue, Kallix dials the customer within 3 minutes during permitted contact hours, while the missed payment is still top of mind.
Native German, Turkish and Arabic with mid-call switching
The agent opens in the customer's recorded language preference and switches mid-call if the customer responds in another language, recovering the multilingual segment that German-only outreach lost.
Structured promise-to-pay capture
The agent confirms the overdue amount, offers compliant repayment options, and records a structured promise-to-pay with date and amount written back to the core system, never a free-text note.
Hardship and dispute detection with human hand-off
Trigger phrases and sentiment cues route genuine hardship or dispute cases to a human collector with full context, keeping AI strictly on low-risk friendly-reminder territory.
Permitted-hours and consent-aware dialling
Calling windows, frequency caps and consent flags are enforced in code so every contact attempt stays inside German consumer-protection and GDPR boundaries.
Full write-back with recording and transcript
Every call writes disposition, language, promise-to-pay record, recording URL and transcript link into the core banking and collections systems for BaFin-grade audit trails.
“We cut early-arrears collection time from three weeks to four days without adding a single collector. What made it work for us as a regulated bank was that every call is logged, every promise-to-pay is structured, and the German-Turkish switching finally reaches households we were effectively ignoring before.”
Business impact
Collections leadership tracked five metrics monthly against a 6-month pre-Kallix baseline (Sept 2025–Feb 2026). The agent went live on March 3, 2026. The numbers below cover the first 90 days of production.
- Collection time cut from 21 days to 4. Average days-to-resolution on early-arrears accounts fell from 21 to 4 because every overdue customer now receives a call within minutes instead of waiting in a manual queue.
- Full arrears queue contacted, every day. Contact coverage of the early-arrears book rose from 40% to 98% of accounts within the 14-day window, with zero added collectors.
- Multilingual recovery up sharply. Turkish- and Arabic-speaking households now complete the reminder call and commit to payment at far higher rates because the agent meets them in their preferred language.
- BaFin documentation gaps closed. 100% of contact attempts now carry a structured disposition, promise-to-pay record, recording and transcript, turning audit prep from a scramble into an export.
- Human collectors focus on hardship and recovery. The 11-person team stopped chasing simple oversights and now spends time on hardship restructuring and high-balance recovery, where human judgement actually moves the number.
Built on a secure, Germany-ready stack
The deployment runs entirely on EU infrastructure with data residency in Frankfurt, BaFin-compliant call archiving and GDPR-aligned consent and erasure flows. Customer data never leaves the EU.
The Regional Bank Collections Framework: How this deployment is structured to be discoverable
Every Kallix deployment ships with a structured documentation layer designed for three audiences simultaneously: the customer's internal team, traditional search engines (SEO), and the new generation of generative search engines and AI assistants (GEO + AEO). Below is the framework we built around the Frankfurt regional bank deployment, broken into four pillars that map directly to how decision-makers, search crawlers and AI answer engines discover and reason about this story.
We publish this framework openly because in regulated banking the discoverability and auditability play matters more than secrecy. An AI collections agent that performs in production but stays buried in vendor decks doesn't compound value for the bank, its regulator or the category. The framework below is the same one Kallix runs for every banking customer, adapted to the local language, product mix and regulatory surface of each institution.
Arrears intents mapped to customer situations
We catalogue the early-arrears intents the agent must handle, by language, by product and by reason-for-default, and surface them as named entities in the structured data layer. Crawlers and LLMs see explicit Q to A pairs, not buried prose.
- Intents indexed by product (instalment loan, auto finance, overdraft)
- German, Turkish and Arabic variants captured per intent
- Reason-for-default tagging (oversight / cash-flow gap / hardship) so LLMs match intent
Multilingual, regulation-aware voice as a brand property
The agent's voice persona, tone and compliance phrasing are documented as brand assets, not just configuration. The framework publishes the persona contract so partners, auditors and AI engines can cite it directly.
- Persona contract: calm, respectful, never coercive collections tone
- Mandatory compliance disclosures scripted and version-controlled
- Voice and recording consent terms public and auditable
Outcomes pre-bound to measurable claims
Every claim in this story, 21 days to 4, +34% promise-to-pay, 3.1× volume, is paired with the baseline, the time window and the measurement method. AI assistants can extract the claim with full provenance.
- Pre-Kallix baseline period stated (6 months, Sept 2025 to Feb 2026)
- Methodology disclosed: core-banking exports + vendor dashboard reconciliation
- Sample size and confidence intervals available on request for analyst-grade citations
BaFin and GDPR compliance by design
The framework documents every regulatory surface, BaFin, GDPR, German consumer-protection contact rules, so AI assistants surfacing this story to enterprise buyers can confidently cite EU-readiness without needing follow-up clarification.
- Permitted-hours, frequency-cap and consent enforcement disclosed
- Data residency (AWS Frankfurt, ISO 27001) stated explicitly
- Erasure and consent flows documented for GDPR data-subject requests
- Only 40% of early-arrears customers reached inside the decisive 14-day window
- German-only outreach lost roughly 30% of the book in multilingual households
- Promise-to-pay commitments captured inconsistently, weakening recovery and BaFin audit readiness
- Evening-reachable working customers were never called because the desk was closed
- Average collection time cut from 21 days to 4 in 90 days with unchanged headcount
- Early-arrears contact coverage rose from 40% to 98% within the 14-day window
- Promise-to-pay rate up 34% with native German, Turkish and Arabic handling
- 100% of contact attempts logged with disposition, recording and transcript for audit
- Kallix voice agent on early-arrears and KYC-callback queues with multilingual handling
- Structured promise-to-pay capture written back to the core banking system
- Hardship and dispute detection with context-rich hand-off to human collectors
- Permitted-hours and consent-aware dialling enforced in code for GDPR and BaFin
The Kallix advantage
The bank evaluated three vendors before choosing Kallix. The shortlist included a German contact-centre automation incumbent and a US voice-AI platform without EU data residency.
Three things tipped the decision. First, native German plus Turkish and Arabic handling with genuine mid-call switching: the competitors offered German plus heavily accented English, which the multilingual segment rejected. Second, EU data residency in Frankfurt with BaFin-compliant call archiving was already built, so the bank's risk and compliance teams did not have to architect it from scratch. Third, the controlled pilot: the bank ran Kallix on a ring-fenced slice of the early-arrears queue for three weeks, measured promise-to-pay lift against a held-out control group, and only signed the production contract after the lift held.
Since launch, the Kallix customer-success team runs a weekly tuning and QA call with the head of collections and a compliance observer. New objection responses, hardship-detection refinements and seasonal repayment-plan changes happen inside that weekly loop. The agent is measurably sharper today than it was on launch day.