On-Demand Operations/Customer Score: 3.55/5.0
On-Demand Knowledge Work | Internal/External audience
Banks receive 500-5000 customer complaints/month via multiple channels (email, phone, chat, complaint form, social media, regulatory ombudsman referral). Complaint types vary: product quality (service disruption, fee issue), conduct (sales practice, advisor misconduct), processing (slow service, error), and regulatory concerns. Current process: customer service team manually classifies complaint by product/issue type/severity, routes to specialist team, and generates acknowledgment. Median classification time: 15-30 minutes per complaint. Mis-routing occurs (30-40% of complaints routed to wrong team initially), causing rework and customer dissatisfaction. Escalation patterns are unclear; complaints that should trigger regulatory reporting are sometimes missed.
Data Sources:
Data Classification:
Data Quality Requirements:
Complaint ingestion completeness: 99.9%+ of complaints captured (no dropped complaints). Classification accuracy baseline: 85%+ (established from historical manual classifications). Escalation accuracy: 100% (no missed regulatory reporting thresholds). Routing accuracy: 90%+ (baseline from historical routing accuracy). Response time: <15 seconds from complaint ingestion to routing decision.
Integration Complexity: Medium , Requires integration with multiple intake channels (email, web form, phone/chat systems). NLP classification model requires training on bank's historical complaints (domain-specific terminology). Routing logic depends on complaint taxonomy and product matrix. Escalation rules require regulatory knowledge and mapping to thresholds. Complaint management system integration (Salesforce, Zendesk) requires API access. Auto-response generation requires email/communication system integration.
| Criterion | Weight | Score (1-5) | Weighted |
|---|---|---|---|
| Time Recaptured | 15% | 4 | 0.60 |
| Error Reduction | 10% | 4 | 0.40 |
| Cost Avoidance | 10% | 3 | 0.30 |
| Strategic Leverage | 5% | 3 | 0.15 |
| Data Availability | 15% | 4 | 0.60 |
| Process Clarity | 15% | 4 | 0.60 |
| Ease of Implementation | 10% | 3 | 0.30 |
| Fallback Available | 10% | 4 | 0.40 |
| Audience (Int/Ext) | 10% | 4 | 0.40 |
| Composite | 100% | 3.55 |
Complaint data is available (arrives via documented channels). Classification taxonomy is explicit (bank maintains complaint classification matrix per regulatory requirement). High volume (500-5000 complaints/month) justifies investment. Clear time savings: reduce classification time from 15-30 minutes to <2 minutes via agent. Fallback is simple: customer service team manually classifies if agent fails. Mix of internal (routing) and external (acknowledgment) communication creates some complexity, but both are templated. Clear measurement: track mis-routing rate, resolution time, escalation accuracy.
Sprint 0 (2 weeks) + 3 build sprints (6 weeks)
Sprint 0: Complaint channel integration, classification taxonomy design, routing matrix definition, escalation threshold definition
Build Sprints 1-3: NLP complaint classification model, routing logic, auto-response generation, escalation workflow, reporting/measurement
From zero to a governed, production agent in 6 weeks.
Sprint Factory Schedule a Briefing