Continuity across channels is how an agent starts a conversation already holding what your systems know about the customer. Verified context and localization run in webchat today. The cold start they remove is costing you the customers you most want to keep.
A customer who has ordered from you nine times opens a chat. She is signed in. Her cart is full. The agent's first move is to ask for her email and a one-time code to prove she is the person already logged into the account she is looking at. By the time she reaches an actual question, she has answered three she didn't need to.
An AI agent recognizes a returning customer the way a good human one does: by already knowing who walked in. Continuity across channels means the agent starts each conversation with verified context (who the customer is, their status, their history) carried in from your own systems, in the customer's own language, from the first turn. It has two moving parts: verified context passed securely from your business to the agent, and localization that renders the conversation in the visitor's language from the first click.
The cold start, which treats a known customer as anonymous, is the clearest sign an agent isn't wired into the business behind it.
Nothing in your business told the agent who she is
Three systems in your business already hold the answer:
- Your site knows she is signed in.
- Your loyalty program has her at Gold-tier.
- Your order history has last month's purchase.
None of it reaches the agent, so it does the only thing it can and asks. It asks for the identity she established by logging in, and it re-pitches the enrollment she completed two years ago.
Each redundant question reads, to the customer, as proof the agent is stapled to the business rather than part of it. The customer who feels it most is the one who has been with you longest, the regular who expects to be recognized and meets friction at the door instead.
She can give up and call your phone line, which is the live-agent cost you were trying to contain. She can also give up on you. Push a loyal customer through the cold start once too often and she does not come back, and her lifetime value walks out with her.
The fix is to give the agent what your website already knows.
What continuity across channels actually looks like
The same customer opens the same chat. The agent greets her by name, already aware she is signed in and a Gold-tier member, and moves straight to her order. Nothing is re-verified, because nothing needs to be.
The same customer, the same question, three turns apart:
| Turn | Without continuity | With continuity |
|---|---|---|
| 1 | "Can I get the email address on the account?" | "Hi Maria, I can see your recent order. What can I help with?" |
| 2 | "I've sent a one-time code. Can you read it back to me?" | Maria asks her question. |
| 3 | "Before we continue, would you like to join our loyalty program?" | The agent answers it against her order history. |
| 4 | Maria asks her question. | Conversation over. |
Maria is signed in, and she has been a Gold-tier member for two years.
Answering a question and resolving a need are different outcomes. The left column does the first, the right column does the second, and the difference between them is context your business already had. Continuity is that context, present from the first turn.
How does the agent know the customer without exposing their data?
This is the question your security team asks first, and the answer is what makes continuity safe to ship. The customer's own system passes the agent a cryptographically verified statement of who the person is: identity, loyalty status, whatever you choose to share. PolyAI verifies that signal from your systems to the agent and holds it in a dedicated internal service, keeping tokens and personal data out of the language model's context by default.
That matters for regulated buyers. The identity the agent acts on is verified by cryptography, not inferred from a browser attribute anyone could spoof, so it holds up in the procurement review a bank runs before it lets an AI agent near a customer record.
Customers see none of this. They notice only that they weren't interrogated.
What about customers who don't speak English?
A customer in Montreal opens the chat in French and is greeted in English. She switches to English herself, because it's faster than fighting the widget, and gets her answer. Nothing broke. You just found out which language your brand actually operates in, and so did she.
Continuity means the widget opens in the visitor's language from the first click and follows her if she switches mid-conversation. No half-English seam, no reset. PolyAI runs French-Canadian in production today.
In Quebec this is not a preference. The Charter of the French Language, strengthened by Bill 96 in 2022, requires businesses to serve consumers in French, and the chat widget is the most public place to fail that test.
Verified context and localization are live today
Your agent can open a webchat already holding who the customer is, what tier they're on, and what they last ordered, in the language their browser is set to. Both halves are running in production now: the verified context service that carries the signal from your systems, and localization that opens the widget in the visitor's language. Carrying that same context between voice and chat is in development.
Count the questions your systems could already answer
Look at your agent's first three turns with a known customer. Count the questions it asks that your own systems could already answer.
That number is your cold start, measured. It's also the fastest thing you can fix, because every question on that list is one your business already knew the answer to.
Book a demo today to see the only purpose-built dialog model for customer service in action.
