What it does

Routes all work items (voice, chat, SMS, social, cases, email) through a two-stage pipeline: Classification (enrich the work item) then Assignment (match to the right agent). All channels run through the same engine.

Key facts

• Two stages: Classification (rules/ML add context: skills, priority, capacity profile) → Assignment (match enriched item to agent by skills, capacity, presence, workload)
• Channels handled: voice, live chat, SMS, Facebook, WhatsApp, Teams, LINE, WeChat, custom messaging, email, cases
• Agents must use Copilot Service workspace: Customer Service Hub does not support unified routing assignment
• Not supported in Unified Service Desk: hard blocker for orgs on legacy USD
• Intelligent Skill Finder (ML routing) requires AI Builder and is not available in all geographic regions
• Configuration changes can take time to propagate: allow for propagation delay in test scripts

When to use / skip

Unified routing is mandatory for D365 Contact Center and Customer Service routing scenarios. There is no alternative; if the org is on USD, migration to Copilot Service workspace is required first.

Configuration decisions

• Classification strategy: rule-based (logical conditions) vs ML-based (Intelligent Skill Finder): use rule-based at go-live; ML needs training data
• Queue structure design: the most consequential architectural decision; drives workstream, capacity profile, and assignment method choices downstream
• Capacity model: profiles or units: use profiles for all new implementations

Gotchas

• Classification before assignment. Common mistake: jump straight to queues without designing classification rules. Poor classification means assignment can't match items properly: they arrive with no useful context.
• Copilot Service workspace is mandatory. Agents on Customer Service Hub won't receive work. Non-obvious blocker for hybrid rollouts where some agents haven't migrated yet.
• Start simple. Most clients try to replicate all existing routing complexity on day one. Start with 3–5 queues and basic rule-based classification. ML and complex skill taxonomies come later, after you have real data.

Consultant notes

• Queue structure design is the decision everything downstream depends on. Get it wrong and you're reworking workstreams, classification rules, and capacity profiles weeks later. Push for a complete queue architecture sign-off on paper before anyone touches the admin centre.
• The instinct on every project is to replicate the full complexity of the existing system on day one. That complexity is usually a workaround for limitations that don't apply in D365. Challenge it: start with 3–5 queues and basic classification, then add complexity based on real data.
• USD as a hard blocker needs to come up in pre-sales, not after kick-off. If a client is on Unified Service Desk and nobody flagged it before the project started, that's a scope conversation waiting to happen.

---

Source last updated: 2026-04-15 | Worth revisiting when: Routing architecture changes in release wave plans