Case study · Insurance · Cyprus
Two FTEs of manual email-and-phone outreach replaced by a multi-channel AI cadence that respects ICCS rules, GDPR, and the customer's quiet hours. Built in 2 weeks on n8n, then re-built into a custom application integrated with their insurance management system.
Service used: Workflow Automation
Renewal cadence
Customer: Andreas P. · Language preference: Greek
Action: Renewal email sent in EL. No premium change — cadence continues.
Day 0 — Renewal email (EN / EL / RU)
Personalized · Language matched
Day 3 — SMS with one-click renewal link
If no reply · Consent checked
Day 7 — Voice agent call (quiet hours enforced)
No calls before 09:00 or after 20:00
This week: 1,000 policies queued
Run time: <1 hour
~1,000
policies per renewal cycle
52,000
renewals processed in a year
~2 FTE
of manual outreach freed up
100%
logged interactions for ICCS / GDPR audit
n8n → Node.js
Phase 1 to Phase 2 build path
Client
DigiCare Insurance is a Cyprus-based brokerage, the trading name of Costas Matheou Insurance Agent. They write personal lines (motor, home, health, travel, life, pet, motorbike, expat) and commercial lines (employers' liability, public liability, professional indemnity, block-of-flats). Their customer base is multilingual — English, Greek, and Russian — which is the norm for Cyprus brokers serving locals, expats, and a long-standing Russian-speaking community on the island.
| Field | Detail |
|---|---|
| Country | Cyprus (EU) |
| Regulator | Insurance Companies Control Service (ICCS), Ministry of Finance |
| Customer languages | English, Greek, Russian |
| Renewal volume | ~1,000 policies per week, ~52,000 per year |
| Lines written | Personal + commercial |
| Site | digicare-insurance.com |
| Engagement | Phase 1 PoC in n8n (~2 weeks) then Phase 2 custom integrated app (~8 to 12 weeks) |
Before
Every Monday morning, DigiCare's team opened the renewal list for the week. Around 1,000 policies. Their job was to reach each customer, confirm whether they wanted to renew, and update the policy status before the expiry date.
The mechanics were simple and brutal:
Two full-time staff did almost nothing else. Roughly 80 person-hours a week, on outreach. And it still wasn't enough. By Wednesday they were behind. By Thursday the cycle was leaking renewals — customers who never picked up, never replied to email, and silently let coverage lapse.
The pattern is common in brokerages of this size. Renewal outreach is the highest-leverage activity in the agency, and it almost always sits in the lowest-leverage hands.
The other layer of pain was the audit trail. Cyprus brokers are regulated by the Insurance Companies Control Service (ICCS) under the Insurance and Reinsurance Services Law of 2016, which is aligned with the EU Insurance Distribution Directive. Every customer interaction needs to be logged with timestamp, channel, and outcome. Manual outreach produces spotty notes. Spotty notes are a compliance risk. Compliance risk eats agency-owner sleep.
DigiCare didn't need a CRM. They had one. They needed the renewal cadence to run itself, in three languages, without breaking any of the rules a Cyprus brokerage has to live by.
Before
After
Solution architecture
We built the renewal automation in two phases. Phase 1 was a fast n8n proof-of-concept to prove the flow worked end-to-end and to start producing ROI inside a month. Phase 2 was a custom application, built directly into DigiCare's insurance management system via API and webhooks, which is what runs in production today.
The flow itself stayed the same across both phases. The architecture is the case study — not the platform.
The flow, end to end
Ingest renewal list
Pull weekly batch from agency system
Extract policy data
AI reads price, type, policy #, expiry, channels
Premium guardrail check
>10% YoY change → branch to human
Multi-channel cadence
D0 email · D3 SMS · D7 voice · D10 email
Voice agent confirmations
Coverage Q / neg. sentiment → branch
Consent + opt-out checks
STOP · unsubscribe · press 9 to opt out
Audit log
Every interaction: timestamp + transcript
Status writeback
Confirmed / declined / needs human / no contact
Weekly digest
Monday email: channel performance to agency owner
Human licensed agent
Human licensed agent handles escalations
Premium-change advice (AI cannot advise on coverage changes — Cyprus/EU regulatory requirement)
Coverage questions · Negative-sentiment escalations from voice AI
Ingest renewal list
Pull weekly batch from agency system
Extract policy data
AI reads price, type, policy #, expiry, channels
Premium guardrail check
>10% YoY change → branch to human
Multi-channel cadence
D0 email · D3 SMS · D7 voice · D10 email
Voice agent confirmations
Coverage Q / neg. sentiment → branch
Consent + opt-out checks
STOP · unsubscribe · press 9 to opt out
Audit log
Every interaction: timestamp + transcript
Status writeback
Confirmed / declined / needs human / no contact
Weekly digest
Monday email: channel performance to agency owner
Human licensed agent lane
Escalation from step 3 (>10% premium change) and step 5 (coverage questions / negative sentiment). Required by Cyprus/EU regulation: AI cannot advise on coverage changes.
The flow runs every Monday at 09:00 Cyprus time, finishes the queue inside an hour, and the human team starts their week looking at the exceptions — not the entire renewal list.
How we built it
We don't deliver an 8-week build before anyone sees a working version. The point of Phase 1 is to validate that the flow makes sense for the actual policies, the actual customers, and the actual edge cases — before anyone signs off on the full integration.
Phase 1 — Fast PoC
A working multi-channel cadence wired to a sample renewal list, sending real emails and SMS to a small cohort, with a feature-flagged voice agent. Goal: prove the flow, surface integration edge cases, generate immediate ROI.
Phase 2 — Production app
Direct API and webhook integration with DigiCare's insurance management system. Hardened error handling, retries, observability, role-based access, audit-grade logging, regional EU data residency. The system DigiCare runs today.
Phase 1 doesn't replace Phase 2. It de-risks Phase 2. By the time we started the production build, every "wait, what about X" question had already been answered on a real cohort.
If you've ever been pitched a six-month build before you've seen anything working, you already know why we do it this way.
Built for ICCS and GDPR
A renewal-automation system that fires SMS and outbound calls into a regulated EU market is one bad design decision away from a regulator complaint. We designed for the boring cases first.
Anything that would constitute insurance distribution advice — premium changes over 10%, cover modifications, complex policy questions — routes to a licensed human agent. The voice agent is allowed to confirm, not to sell.
All workflow execution and data storage runs in EU regions (Frankfurt or Dublin), under the EU GDPR (Regulation 2016/679). No customer PII leaves the EU.
SMS marketing requires opt-in; renewal communications run under the legitimate-interest basis as a continuation of an active customer contract, with a clear opt-out on every message. Voice calls check the do-not-contact list before dialing.
No outbound calls before 09:00 or after 20:00 local Cyprus time. No outbound calls on Sundays. These are the rules; we wired them into the scheduler so they cannot be accidentally turned off.
Every email, SMS, and voice call is logged with timestamp, channel, intent, outcome, and (for voice) a transcript. If ICCS asks, or if a customer raises a GDPR access request, the full record is one query away.
STOP unsubscribes the customer from SMS. 'Agent' or sentiment-negative cues escalate the voice call to a human. There is no scenario where a customer is held in an automated loop against their will.
If your jurisdiction's regulator looks anything like the ICCS — most EU insurance regulators do — this same design pattern transfers directly. It's the layer most "AI insurance broker" pitches skip past, and it's the one that makes or breaks the build.
Outcomes
Framing note: every number below is approximate. Some are derived from defensible insurance-industry benchmarks. Others are based on early post-launch tracking against the manual baseline. Where uncertain, we hedge.
| Metric | Before (manual) | After (automated) |
|---|---|---|
| Policies processed | ~1,000 / week | ~1,000 / week |
| Renewals processed per year | ~52,000 | ~52,000 |
| Person-hours per week on outreach | ~80 (2 FTE) | ~6 to 10 (exception review) |
| Speed to first contact | 3 to 5 day backlog | Under 60 seconds |
| Renewal retention rate | Industry baseline ~84 to 85% | Estimated +5 to +9 pts post-launch |
| Cost per successful renewal contact | ~€2 to €3 | ~€0.20 to €0.30 |
| Audit completeness | Spotty notes | 100% logged |
| Languages handled in cadence | EN / EL / RU manual | EN / EL / RU automated |
The most underrated win is the audit log. Two FTE of time savings is the headline. But what makes the agency owner sleep at night is that every single one of 52,000 customer interactions is on file — searchable, exportable, and ready for an ICCS or GDPR access request without anyone having to find a sticky note.
“Two of my people used to spend their entire week chasing renewals — emails, SMS, calls, in three languages. Scale me AI shipped an n8n version in two weeks that paid for itself, then built it into our agency system properly. Now around 1,000 policies a week move through the cadence on their own. The team only sees the exceptions.”
Under the hood
| Layer | Tool / vendor | Why |
|---|---|---|
| Workflow engine, Phase 1 | n8n (self-hosted, EU) | Visual, fast to iterate, self-hostable for compliance |
| Workflow engine, Phase 2 | Custom Node.js / TypeScript app | Full control, hardened observability, deeper integration than a no-code engine allows |
| LLM (data extraction, personalization) | Modern frontier LLM — OpenAI and Anthropic both qualified | Both meet EU data-processing requirements for our use; final choice scoped at deployment |
| Voice agent | Vapi (with ElevenLabs voices) | Reliable Greek and English voice quality, low latency, transcript export, escalation hooks |
| SMS | Twilio | Coverage in Cyprus, STOP keyword handling, deliverability |
| Postmark | Transactional deliverability, EU data residency option | |
| Insurance system integration | DigiCare's existing insurance management system, via REST API and webhooks | Generic — the pattern transfers to any modern AMS that exposes an API |
| Hosting / data residency | EU region (Frankfurt or Dublin) | GDPR by design |
We don't fuse a stack onto a client because we have a partner deal. We pick the tools that fit the regulator, the language mix, the integration target, and the cost curve. If your stack is different, the design pattern still holds.
Applicability
If you recognize any of the following, this build pattern transfers directly:
The architecture in this case study is portable. The compliance pattern is portable. The phase-1-then-phase-2 sequence is portable.
What changes is the regulator surface, the language mix, the AMS, and the volume. We scope each engagement around those four variables on the first call.
Book a 30-minute discovery call. We'll map your renewal volume, your AMS, your regulator, and what a Phase 1 in n8n would actually look like for you — before you commit to anything.
Related
Workflow automation
The renewal cadence, the multi-channel scheduling, the audit logging — all built as a structured workflow that runs on a defined schedule and writes outcomes back to the system of record.
See workflow automation→AI voice agents
The day-7 outbound call layer with quiet-hours enforcement, language matching, and human escalation on sentiment or coverage questions. Confirmation channel, not a sales channel.
See AI voice agents→Customer support automation
Multilingual customer outreach, opt-out handling, and the human-escalation pattern. Same principles as the renewal cadence, applied to inbound support and post-sale communications.
See customer support automation→AI integration build
The Phase 2 custom application integrated directly with DigiCare's insurance management system. When a no-code workflow engine doesn't go deep enough, we build the connector.
See AI integration services→A 30-minute call. We'll tell you whether Phase 1 in n8n is the right move for you, and what it would cost — before you commit to anything.