Overview
An IT support platform (ITSM) is architecturally distinct from a customer support system. It must manage not just conversations but assets, configurations, change requests, and service catalogues - all linked together in a queryable graph. This guide covers the key architectural decisions in IdeaSprout's IT Support module.
1. The CMDB: Configuration Management Database
Why a Graph, Not a Table
IT assets don't exist in isolation. A laptop depends on a network switch, which depends on a router, which depends on an ISP circuit. When the router fails, you need to instantly know which assets and users are affected. A flat table cannot answer this query efficiently.
IdeaSprout models the CMDB as a directed graph where nodes are Configuration Items (CIs) and edges are dependency relationships.
User → Laptop → Network Switch → Core Router → ISP Circuit
User → Software Licence → Licence Server
Laptop → OS → Patch LevelGraph Storage
We use PostgreSQL with recursive CTEs for the CMDB graph rather than a dedicated graph database. For SMB scale (< 10,000 CIs), recursive CTEs on a well-indexed adjacency list table outperform graph databases in operational simplicity and query latency.
CREATE TABLE ci_relationships (
parent_ci_id UUID REFERENCES configuration_items(id),
child_ci_id UUID REFERENCES configuration_items(id),
rel_type TEXT, -- 'depends_on', 'runs_on', 'connected_to'
PRIMARY KEY (parent_ci_id, child_ci_id, rel_type)
);
-- Find all CIs affected by a failure of ci_id X
WITH RECURSIVE affected AS (
SELECT child_ci_id FROM ci_relationships WHERE parent_ci_id = $1
UNION ALL
SELECT r.child_ci_id FROM ci_relationships r
JOIN affected a ON r.parent_ci_id = a.child_ci_id
)
SELECT * FROM configuration_items WHERE id IN (SELECT child_ci_id FROM affected);CI Lifecycle States
ORDERED → RECEIVED → IN_USE → UNDER_MAINTENANCE → RETIRED → DISPOSEDState transitions trigger automated actions: RETIRED assets are flagged for licence reclamation; DISPOSED assets generate disposal certificates for audit.
2. Ticket Ingestion and Classification
Sources
IT tickets arrive from multiple sources with different trust levels:
| Source | Trust Level | Auto-assign Priority |
|---|---|---|
| Self-service portal | Low | Based on category selection |
| Email to helpdesk@ | Low | NLP classification |
| Agent-created (phone call) | High | Agent-assigned |
| Monitoring alert (auto-ticket) | System | Severity from alert |
| Manager [escalation](/blog/customer-support-escalation-management-system) | High | P1 by default |
NLP Classification Pipeline
Email and portal tickets go through a lightweight classification pipeline:
- 1Category detection: keyword + TF-IDF model trained on historical tickets → maps to service catalogue item
- 2Priority inference: urgency keywords ("production down", "cannot login", "data loss") → P1/P2 override
- 3CI association: entity extraction for asset names/IDs → auto-link ticket to CMDB CI
The model is retrained monthly on resolved tickets with agent-corrected categories.
3. ITSM Ticket Lifecycle
Ticket Types
IT support distinguishes four ticket types with different workflows:
- •Incident: Unplanned disruption - restore service ASAP
- •Service Request: Planned fulfilment (new laptop, software install) - follows approval workflow
- •Problem: Root cause investigation of recurring incidents
- •Change Request: Planned modification to infrastructure - requires CAB approval
Incident vs Problem Linkage
When 3+ incidents share the same CI or category within 7 days, the system automatically creates a Problem record and links the incidents to it. Resolving the Problem auto-resolves all linked open incidents.
4. SLA Engine with Priority Matrix
Priority Matrix
| Impact Urgency | High | Medium | Low |
|---|---|---|---|
| High (many users affected) | P1 - 1h response / 4h resolve | P2 - 2h / 8h | P3 - 4h / 24h |
| Medium (team affected) | P2 - 2h / 8h | P3 - 4h / 24h | P4 - 8h / 48h |
| Low (single user) | P3 - 4h / 24h | P4 - 8h / 48h | P4 - 8h / 48h |
Business Hours Calculation
SLA timers use the same elapsed-seconds pattern as customer support, but IT support adds maintenance windows - scheduled downtime periods during which SLA timers are paused for all affected CIs.
CREATE TABLE maintenance_windows (
id UUID PRIMARY KEY,
ci_ids UUID[], -- affected CIs
starts_at TIMESTAMPTZ,
ends_at TIMESTAMPTZ,
approved_by UUID
);5. Self-Service Portal Architecture
The Deflection Goal
Every ticket deflected by self-service saves ₹150–₹300 in agent cost. The portal is designed to deflect 25–40% of L1 tickets.
Components
- 1Service Catalogue: Structured list of requestable services with forms, approval chains, and fulfilment SLAs
- 2Knowledge Base: Searchable articles with full-text search (PostgreSQL
tsvector) and view-count-based ranking - 3Chatbot: Rule-based decision tree for top-20 ticket categories; escalates to agent if unresolved
- 4Status Page: Real-time CI health status derived from monitoring integrations
Search Architecture
Knowledge base search uses PostgreSQL full-text search with a custom ranking function:
SELECT id, title,
ts_rank(search_vector, query) AS text_rank,
view_count * 0.01 AS popularity_boost
FROM kb_articles,
to_tsquery('english', $1) query
WHERE search_vector @@ query
ORDER BY (text_rank + popularity_boost) DESC
LIMIT 10;6. Asset Lifecycle Automation
Warranty Expiry Alerts
A daily job scans all CIs with warranty_expires_at within the next 90 days and creates a Service Request for renewal review. The job uses a sliding window to avoid alert fatigue: one alert at 90 days, one at 30 days, one at 7 days.
Software Licence Reconciliation
Monthly reconciliation compares:
- •Purchased licences (from procurement records)
- •Installed instances (from endpoint agent inventory)
- •Active users (from Active Directory / Google Workspace sync)
Discrepancies (over-deployed or under-utilised licences) generate a reconciliation report and optionally trigger reclamation workflows.
7. Common Architectural Pitfalls
| Pitfall | Consequence | Fix |
|---|---|---|
| Flat asset table instead of graph | Cannot answer "what's affected by this failure?" | Adjacency list with recursive CTE |
| Single ticket type for all requests | Incidents and service requests need different workflows | Separate ticket types with type-specific state machines |
| SLA timer as absolute deadline | Maintenance windows and timezone changes break SLAs | Elapsed-seconds with pause/resume events |
| No CI linkage on tickets | Cannot identify recurring problems | Auto-link via entity extraction + manual association |
| Knowledge base without search ranking | Users can't find articles, deflection rate stays low | TF-IDF + popularity boost ranking |
Summary
A production ITSM platform requires a graph-modelled CMDB, type-aware ticket workflows, a business-hours-aware SLA engine with maintenance window support, and a self-service portal designed for maximum deflection. The architectural investment pays back in reduced agent load, faster incident resolution, and audit-ready asset records.