reltix

Reltix property managers no longer spend their mornings manually sifting through a queue of tenant requests in German. The AI Tenant Request Auto-Responder has reduced their daily administrative load by 14 hours per week, shifting their role from reactive communicator to proactive portfolio overseer. Tenant satisfaction scores have risen by 22 points, as response times for urgent maintenance issues now average 47 minutes instead of 28 hours.

Today, property managers and WEG administrators manually read, categorize, prioritize, and draft responses to every tenant request—a process that consumes 2.1 hours daily per manager (source: internal time-tracking survey of 42 customers, Q4 2024) and creates a 28-hour median response delay for urgent issues (source: Reltix ticket data, Jan 2025). This delay risks lease non-compliance, tenant churn, and preventable asset damage. The business case: 1,250 active property managers on platform × 2.1 hrs/day × €45 blended hourly cost (source: regional PM salary benchmarks) × 220 working days = €25,987,500/year in recoverable administrative cost. If adoption reaches only 40% of target users: €10,395,000/year. This excludes the recoverable value from prevented tenant churn (estimated 3.7% annually, €1.2M) and avoided emergency repair markups.

This feature is an automated classification, drafting, and routing system that integrates with the existing ticket inbox to handle the initial triage and response for common, non-urgent tenant requests. It is not a fully autonomous tenant support agent, a replacement for human judgment on complex legal/issues, or a system that executes work orders or payments.

WHO / JTBD: When a property manager or WEG administrator starts their day, they want to efficiently triage and address all tenant requests to maintain compliance, tenant satisfaction, and property value—without being buried under hours of repetitive communication and manual prioritization.

WHERE IT BREAKS: Today, the manager opens the Reltix communications dashboard to a flat, chronological list of requests ranging from a leaking faucet to a noise complaint. Each request must be manually read, its urgency assessed (often without clear tenant context), its category determined, and a personalized response drafted in German. This universal manual process treats a billing inquiry with the same initial workflow as a broken heating system in winter. Critical issues get lost in the noise, response templates are copied manually from old tickets, and routing to the correct handyman or accountant requires manual @mentions.

WHAT IT COSTS:

JTBD statement: "When I open my request inbox, I want the urgent, actionable items surfaced immediately with a drafted response and clear next steps, so I can protect tenant safety and asset value without wasting time on categorization and first drafts for routine queries."

Integration Map: Reads From:

• Tenant Request Ingest API: Source of raw request text, sender (tenant), and property ID.
• Property Database: For tenant history, unit details, and assigned manager/WEG board.
• Vendor/Team Directory: For routing logic based on issue type and property.

Writes To:

• Ticket Database: Adds auto_category, auto_urgency, auto_response_draft, suggested_assignee, suggested_timeline.
• Communications Outbox: Sends auto-approved responses for non-urgent, templated cases.
• Manager Alerting System: Triggers push/email notification for P0/P1 tickets requiring human review.

Core Mechanic: Upon ticket creation, the feature’s AI pipeline 1) classifies request intent and urgency, 2) generates a personalized German response draft using ticket and property context, 3) applies routing rules, and 4) either auto-responds (for low-risk, templated queries) or surfaces the drafted response + routing suggestion to the manager for one-click approval.

Primary User Flow: The manager's primary interface remains their familiar ticket dashboard, now enhanced with auto-generated metadata and actions.

┌─────────────────────────────────────────────────────────────────────────┐
│ Reltix — Request Inbox                           [Filters: All ▼]       │
├─────────────────────────────────────────────────────────────────────────┤
│ 🔴 URGENT - Heizung defekt (Apartment 3B)                               │
│    From: Müller, 3B                  Property: Hauptstr. 10            │
│    ↳ Auto-Category: MAINTENANCE (Heating) | Urgency: P0                │
│    ↳ Suggested Assignee: HVAC GmbH (Klaus) | ETA: 4h                   │
│    Drafted Response: «Sehr geehrte Frau Müller, wir haben Ihren...»    │
│    [Review & Send]      [Edit Draft]      [Re-assign]                  │
├─────────────────────────────────────────────────────────────────────────┤
│ 🟡 ROUTINE - Mietzahlung Februar                                        │
│    From: Schmidt, 1A                 Property: Gartenweg 5             │
│    ↳ Auto-Category: BILLING | Urgency: P2                              │
│    ↳ Auto-Response Sent: ✅ «Ihre Zahlungsanfrage wurde erfasst...»    │
│    [View Thread]       [Re-open]                                       │
└─────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────┐
│ AI Response Draft & Routing Preview                                     │
├─────────────────────────────────────────────────────────────────────────┤
│ Request: «Sehr geehrte Hausverwaltung, die Heizung in 3B funktioniert…»│
│                                                                         │
│ Auto-Detected:                                                          │
│   ● Category: MAINTENANCE — Heating System                             │
│   ● Urgency: P0 (Keywords: "defekt", "keine Wärme", "Winter")          │
│   ● Suggested Resolution: Emergency HVAC call                          │
│                                                                         │
│ Drafted Response (German):                                              │
│ «Sehr geehrte Frau Müller, wir haben Ihren Hinweis zum Heizungsdefekt…│
│ Unser Dienstleister (HVAC GmbH) wurde informiert und meldet sich…»     │
│                                                                         │
│ Suggested Routing: HVAC GmbH — Klaus (Priority: Hoch)                  │
│ Suggested Timeline: Initial contact within 4h, repair within 24h.      │
│                                                                         │
│ [Approve & Send] [Edit Draft] [Change Category/Routing]                │
└─────────────────────────────────────────────────────────────────────────┘

Phase 1 — MVP: 10 weeks US#1 — Automated Classification & Urgency Scoring

• Given a new tenant request in German in the Reltix system
• When the AI classification pipeline processes it
• Then it assigns a category (MAINTENANCE/BILLING/LEGAL/NOISE/OTHER) with ≥95% accuracy and an urgency score (P0/P1/P2) with ≥99.5% accuracy on P0 detection (P0 dimensions: launch-blocking).
• If story fails, urgent issues are missed, leading to compliance violations and asset damage.
• Validated by QA Lead against a labeled test set of 500 historical tickets.

US#2 — German Response Drafting for Human Review

• Given a classified ticket (any category/urgency)
• When the drafting engine is triggered
• Then it generates a contextually relevant, grammatically correct German response draft, incorporating tenant name, property address, and issue specifics, with a professional tone.
• If story fails, managers reject drafts, wasting more time than saving.
• Validated by German-language Product Manager against 100 sample drafts.

US#3 — One-Click Approval & Send for Manager

• Given a P0 or P1 ticket with a generated draft and routing suggestion appears in the manager's inbox
• When the manager clicks "Approve & Send"
• Then the drafted response is sent to the tenant, the ticket is assigned to the suggested vendor/team member, and the status changes to "In Progress".
• If story fails, managers abandon the feature and revert to full manual process.
• Validated by Lead PM via user testing with 5 customer managers.

US#4 — Auto-Resolution for High-Confidence P2 Tickets

• Given a new ticket classified as P2 with a BILLING or GENERAL INQUIRY intent and confidence score ≥99.5%
• When the ticket is created
• Then a predefined template response is sent automatically, and the ticket is closed with status "Auto-Resolved".
• If story fails, tenants receive incorrect information, generating escalations.
• Validated by Customer Support Ops against a 1-week monitored beta rollout.

Out of Scope (Phase 1):

Phase 1.1 — 4 weeks post-MVP:

• Integration with major German vendor platforms (e.g., MyHammer, Jimdo) for automated assignment.
• SLA tracking dashboard showing response time deltas (Before/After AI). 1.2 — 6 weeks post-MVP:
• "Pattern alerts" for managers (e.g., "3 similar heating issues in Building B this month").
• Tenant-facing status page updates triggered by auto-routing.

Primary Metrics:

Guardrail Metrics (must NOT degrade):

What We Are NOT Measuring:

1. Number of drafts generated (vanity metric; doesn't measure usefulness or time saved).
2. Raw AI classification accuracy on all categories (we focus on P0 accuracy; overall can be low if GENERAL category is a catch-all).
3. Cost saved per ticket (derivative metric; we measure the input driver—manager time—directly).

Risk: AI misclassifies a life-safety urgent issue (e.g., "no heat in winter", "gas smell") as non-urgent. Probability: Low Impact: High Mitigation: ML Lead (Dr. Schmidt) is responsible for implementing a separate, rules-based P0 keyword filter ("gas", "flood", "no heat", "lock broken") that overrides the AI classification before any auto-resolution, by Week 4 of development. Trigger: Any ticket containing a keyword from the P0 list. If mitigation fails, we disable auto-resolution for all maintenance tickets before launch.

Risk: Generated German responses contain factual errors (wrong vendor name, incorrect legal advice) or inappropriate tone. Probability: Medium Impact: High Mitigation: Product Manager (Hofmann) is responsible for implementing a strict content safety layer that redacts/withholds drafts containing entities not in the property database and blocks generation for LEGAL category tickets, by Week 6. All drafts for P0/P1 will require human approval. Trigger: Any customer complaint about AI response content. If mitigation fails, we limit drafting to a strict set of 10 approved template responses.

Risk: Low manager adoption because the feature feels like an unreliable "black box". Probability: Medium Impact: Medium Mitigation: Head of Customer Success (Weber) is responsible for designing and executing a phased onboarding campaign for the first 100 customers, including a clear rollback path and weekly check-ins, to be completed 2 weeks before launch. Trigger: D14 adoption rate <20%. If mitigation fails, we add mandatory "confidence score" indicators and explanation ("Why this is P0") to every drafted ticket.

Risk: Violation of GDPR "right to explanation" for automated decision-making, as we auto-close some tickets. Probability: Medium Impact: High Mitigation: Legal Counsel (Frau Voss) is responsible for ensuring our Terms of Service and Privacy Policy are updated to disclose automated processing for non-urgent inquiries and providing a clear manual override path, obtaining final sign-off by Week 8. This is a legal BLOCKING item under GDPR Article 22. Consequence: If not cleared by Week 8, we cannot launch the auto-resolution component in the EU.

Kill Criteria — we pause and conduct a full review if ANY of these are met within 90 days:

1. Median response time for P0 tickets fails to drop below 8 hours (D90).
2. Manager time spent on tickets does not decrease by at least 25% (D90).
3. Tenant CSAT drops by more than 10 percentage points (D90).
4. More than 2% of auto-resolved tickets are re-opened by tenants due to incorrect resolution (D90).

Decision: Scope of Auto-Resolution Choice Made: Auto-respond and auto-close only for P2 (Non-Urgent) tickets matching specific, low-risk intent categories (e.g., "Billing Inquiry", "Parking Permit Info") with 99.5%+ classification confidence. P1 and P0 tickets always require human review/approval. Rationale: Eliminates risk of auto-escalating a misclassified urgent issue. We rejected full auto-resolution for all non-urgent tickets to maintain manager oversight and trust in the system initially.

Decision: Primary User Interface Choice Made: Enhance the existing ticket inbox with auto-generated metadata and action buttons. Do not create a separate "AI Inbox" module. Rationale: Reduces cognitive load and adoption friction by not introducing a new surface. Managers continue working in their familiar environment. We rejected a separate module to avoid creating a shadow workflow.

Decision: Language Model Sourcing Choice Made: Use a fine-tuned, hosted LLM API (e.g., OpenAI GPT-4 Turbo, Anthropic Claude 3) for classification and drafting, not an open-source model trained from scratch. Rationale: Time-to-market and accuracy for German language generation. The cost per ticket is negligible (<€0.01) relative to time saved. We rejected training our own model due to the 9+ month data collection, training, and validation timeline.

Decision: Fallback Behavior Choice Made: If the AI pipeline fails (e.g., API timeout, low confidence), the ticket defaults to the current state: "Uncategorized", "P2 Urgency", and appears in the manager's inbox with no draft. Rationale: Fails silent and safe, preserving the existing manual workflow without introducing errors. We rejected a "default P0" or "notify admin" fallback to avoid creating panic or alert fatigue.

Decision: Tenant Transparency Choice Made: Auto-sent responses will include a standard footer: «Diese Nachricht wurde automatisch erstellt, um Ihnen schnellstmöglich zu antworten. Ihr Reltix Team.» Rationale: Maintains transparency and sets appropriate expectations. We rejected not disclosing automation to preserve trust and avoid potential legal misrepresentation claims.

Before / After Narrative Before: Petra Müller, a WEG administrator for a 12-unit building in Berlin, starts her Monday. She has 22 unread tenant messages from the weekend. She spends 90 minutes reading each one, copying boilerplate responses for billing questions, and finally spots a message from Unit 3B: "Heizung kaputt seit Samstag, sehr kalt." Realizing this is urgent, she scramble-calls three HVAC vendors, gets voicemail, and drafts a response promising to look into it—28 hours after the tenant's first message.

After: Petra opens her Reltix inbox on Monday. The AI has already categorized and prioritized the 22 messages. At the top, flagged in red: «URGENT - Heizung defekt (Apartment 3B) — Suggested ETA 4h». A professionally drafted response to the tenant, acknowledging the issue and noting the assigned HVAC vendor, is ready for her review. She clicks "Approve & Send". Eleven routine billing inquiries below are marked "Auto-Responded". Petra's active work is now focused on reviewing two other P1 items. She has addressed the critical issue in under 5 minutes.

Pre-Mortem It is 6 months from now and this feature has failed. The 3 most likely reasons are:

1. The P0 detection accuracy was lower in production than in testing, causing several critical maintenance issues to be auto-filed as "Non-Urgent." This led to a high-profile incident (e.g., flood damage), eroding trust with our largest property management customers, who disabled the feature entirely.
2. We underestimated the integration burden for routing. The "suggested assignee" was often wrong because the vendor directory data was stale, forcing managers to manually re-assign every single ticket, adding a step rather than removing one. The promised time savings never materialized.
3. A key competitor (say, Livendo) launched a similar feature 8 weeks before us, but with English/French support and direct integration with major European DIY store APIs for parts ordering. They captured the narrative of innovation, and our "German-only" MVP was perceived as playing catch-up.

What success actually looks like: Six months post-launch, property managers in our quarterly review cite the Auto-Responder as the primary reason they can manage 15% more units without hiring additional staff. The support team no longer hears complaints about slow response times. Tenant satisfaction survey comments shift from "takes forever to answer" to "issues are acknowledged immediately." In a board review, the Head of Product points to a dashboard showing a 65% reduction in "first response time" across the portfolio, calling it the most efficient capital deployment of the year.