When a customer starts a case on chat, picks it up by phone an hour later, and emails about it the next day, an AI keeps the case continuous: same identity, same context, same history. The new ingredient on this rung is persistent state across channels with strong identity matching, so the customer does not start over because the case did not.
A customer chats with the AI about a refund at 11am, gets put on hold for a back-end check, and goes for lunch. They call back at 2pm to follow up. Before AI, the phone agent had no record of the chat. The customer explained again. The agent looked up the order again. The case effectively restarted. The AI version picks up where the chat left off because the case ID is shared.
How this used to be a decision tree
Each channel ran its own tree. Chat had a queue, a script, a ticket system. Voice had a separate IVR, a separate queue, a separate ticket system. Email lived in a third place. The customer was the only thread that connected them. Switching channels meant the customer paid the start-over cost: explain again, prove identity again, wait again.
Why AI doesn't make this a decision tree anymore
A unified case record holds the customer's interactions across channels. When the customer reaches any channel about a known case, the AI loads the case state and resumes. The chat-to-phone-to-email path becomes one continuous case rather than three new ones. The channel is a surface; the case is the thread.
What people in the field are saying
DCX Newsletter's "The customer is still doing the orchestration" names the failure mode this rung addresses: when the system forces the customer to carry the context between channels, the channels are doing less work than they appear to be doing.
How does cross-channel continuity look today?
One case record per customer issue, identified by a stable case ID. Each channel writes its interactions back to the case. When the customer reaches any channel, the AI matches them by identity (session, phone number, email) and loads the case. The AI's reply starts from where the case is, not from a greeting. The customer does not have to repeat themselves because the case did not lose its memory.
What does it take to make this work?
A shared case data layer that every channel reads from and writes to. Strong identity matching across channels (phone number to account, email to account, session to account). A policy for what state the next channel is allowed to load (some history is appropriate; some may need re-verification before reading). Audit logs that span channels so any single customer experience can be reconstructed end to end.
Where does this go wrong?
Identity mismatch: the AI cannot tie the phone call to the chat session because the phone number on file is different from the one the customer called from. State staleness: the chat AI moved the case forward, the voice AI loaded an older snapshot. Privacy leakage: the channel that loaded the case had less verification than the channel that wrote it. Over-loading: the AI reads back the customer's entire history on every interaction, which feels invasive.
Which tools handle cross-channel cases?
- Gladly: one conversation thread per customer is the product.
- Sierra: cross-channel autonomous handling.
- Kustomer: customer-history-first helpdesk.
- Cognigy: enterprise multi-channel orchestration.
- Decagon: action-taking with shared case state.
How would I start doing this?
Pick the two channels with the highest cross-over traffic (often chat and voice). Wire one shared case data layer behind both. Pilot on a low-risk case type (status updates are a common start). Read the audit trails after a month to see where context was dropped, where identity was mismatched, where the customer still had to repeat. Fix those before adding the third channel.
Next: a case that does not resolve in one contact, but takes weeks. Managing a complaint that takes weeks to resolve.