Cross-channel customer context: recognise and hand over

Short answer

Context survives a channel switch on two conditions. First, the system has to recognise that it is the same person by a reliable key — usually the phone number. Second, the next agent or bot needs a short, checked summary rather than the whole history: who this is, what they want, what was promised and what the next step is. Without a key, merging profiles on a guess is more dangerous than having no context at all.

Where the problem starts

A customer asks for a price on Instagram on Monday, writes on WhatsApp on Wednesday about "what we discussed yesterday", and calls on Friday. To them, this is one conversation. To the company it is often three separate threads in front of three different people.

The result is familiar: the customer introduces themselves every time, a promised discount is forgotten, and the second agent repeats questions the first one already asked. The broader channel strategy is covered in omnichannel customer service; this piece is specifically about recognition and passing context on.

Every channel knows the customer differently

The core difficulty of cross-channel context is that every channel identifies the customer by its own ID:

• WhatsApp: the phone number — the strongest key, because it belongs to the customer.
• Instagram, Facebook, TikTok: the platform's internal account ID and username. The number is known only if the customer writes it.
• Website widget: often an anonymous session. If the visitor gave no contact details, there is no way to link them to another channel.
• Phone call: the calling number — which may belong to a family member or an office line.

So to recognise "leyla_design" on Instagram and a +994 number on WhatsApp as the same person, you need a bridge between them: a number the customer wrote on Instagram, or their own confirmation.

Matching rules

Large platforms call this identity resolution: systems such as AWS Customer Profiles combine external CRM data and contact history in one profile, and CRMs recognise duplicates by a unique key — HubSpot, for example, deduplicates contacts by email address. For a small or mid-sized business the practical rule is:

1. 1Strong keyPhone number or email, stored in one normalised format (for example with +994); the same key merges automatically.
2. 2Weak signalSame name, a similar username, writing on the same day — these only warn the agent that it is "probably the same person".
3. 3ConfirmationOn a weak signal the agent asks the customer: "Was it you who wrote on Instagram yesterday?"
4. 4NeverAutomatic merging by name alone. A city has thousands of people with the same first and last name.

Why a wrong merge is worse

When context is lost, the customer is irritated. When context goes to the wrong person, one customer's order, address or health question becomes visible to someone else. That is a privacy problem and it is hard to reverse: many CRMs cannot split merged records. HubSpot's own documentation states plainly that a merge cannot be undone.

What to pass to the next person

A complete history is good, but an agent has no time to read 40 messages. They need a five-line summary:

1. 1WhoName, main number, which channels they have used.
2. 2What they wantOne sentence: "credit terms for a three-bedroom flat".
3. 3What was saidThe price quoted, the discount, the promise — with dates.
4. 4Open questionThe customer's unanswered question, if there is one.
5. 5Next stepWho will do what, and when.

AI can draft the summary, but a person should check the "what was promised" line: if a promise is copied wrongly, the customer will hold you to it.

Chat to call is a special case

The moment context is lost most often is the switch from messaging to a call: if the caller cannot see the chat, the customer tells the whole story again. The rules for that handover are described in chatbot to voice agent handover.

An illustrative example

This is an illustrative example. A customer writes to an aesthetic clinic on Instagram, asks the price of a procedure and mentions their number. Two days later a WhatsApp message arrives from the same number: "Do you have a slot on Friday?"

Because the number matches, the WhatsApp conversation lands in the existing profile. The agent sees which procedure the customer asked about and what price they were quoted. The reply is "Friday 15:00 is free — we'll book 40 minutes for the procedure you asked about on Instagram", and the customer repeats nothing.

When context goes stale

Context is not permanent. A price quoted six months ago may no longer be valid, and a complaint from a year ago should not be the first thing raised in a new conversation. So every item in the summary should show its date, and the team should agree that prices and promises older than 30 days are checked before an agent repeats them. The history is not deleted; the older part is simply read as background, not as an agreement still in force.

Common mistakes

• Storing numbers in different formats (055…, +99455…, 99455…) so the system sees one person three times.
• Merging automatically by name or username.
• Passing on the entire history as "context" without a summary.
• Treating an office or family number as one person's key.
• "Reminding" a customer of what they said in another channel without asking — it feels like being watched.

Limitations

No system knows exactly who a customer is across every channel: there is no bridge between an Instagram user who never gave a number and an anonymous website visitor. Combining data across channels is also processing of personal data — check with a lawyer what you are allowed to combine and keep under your local law.

How Vexvon keeps context

In Vexvon, a customer's history is collected in one timeline from ten sources: Instagram, WhatsApp, Telegram, Facebook, TikTok, the website, phone calls, the AI call agent, manager notes and the system's own events. Phone numbers are detected in chats in five formats, and when the same number comes in again no new record is created — the conversation is added to the existing customer's timeline. A one-sentence summary of the chat can be attached to the lead notification. More on the CRM page.

Next step

Take 30 customers from last month and check: how many wrote from two or more channels, and how many of those appear in your system as a single profile? The gap is your context loss. The day-to-day side is in the omnichannel inbox, and we can show an example with your own channels during a demo.