When video belongs in a support conversation, and when it is theatre

Adding video to support is easy to sell internally and easy to get wrong, because the failure is quiet. It does not break anything; it just makes most interactions slightly worse — slower, more demanding, harder to search afterwards — while a small number get dramatically better. The work is in telling those apart, putting video inside the conversation rather than beside it, and settling recording, consent and data location before the first call.

When the customer cannot describe the problem. A physical object, an installation that is wrong in a way they lack words for, damage, a device's lights. Thirty seconds of camera can replace fifteen minutes of questions.

When identity or trust is the bottleneck. Identity checks, high-value authorisations, disputes — situations where the customer needs to see a person engaged, or you need to see the customer.

When the agent must demonstrate. Showing beats describing for anything spatial or sequential: a setting, a connection, a sequence of steps.

• Anything the customer could have typed. Video is synchronous, and synchronous is expensive for both sides.
• Anything requiring a lookup. Silence while an agent searches a system is fine in chat and painful on camera.
• Anything people would rather do unseen. Billing problems, complaints, anything embarrassing. Offering video here reads as pressure.
• Poor connections. For customers on weak mobile data, an audio call or photos in the thread may work better.

Three details make the difference:

• Camera optional, audio first. Much of the value is voice plus the rear camera or a shared screen. Requiring the front camera loses people for no benefit.
• Accessibility. Offer captions where possible, and always keep a text route for people who cannot use audio or video.
• Hours and capacity. Only offer video when an agent who can do it is actually available; a link that leads to a wait is worse than no offer.

Support video is usually two people, which browsers handle with WebRTC. Media is always encrypted in transit, and when a direct connection between the two devices is possible, it flows peer to peer. But direct paths often fail on corporate and mobile networks, and then media passes through a relay (TURN) server. Recording, transcription, or adding a third participant typically requires a media server that receives the streams.

So "peer to peer" is not a data-location promise. Before you tell customers where their video goes, find out where your relays and media servers run.

None is universally right. The deciding questions are volume, how deeply video must sit inside your support tool, and where media and recordings may be stored.

If you record, that is a consent and retention decision before it is a technical one — especially when a customer may be showing you their home, their documents or other people. Rules differ by jurisdiction: some require all parties to agree to recording, and data protection law treats video of identifiable people as personal data. Ask before recording, say why and for how long, store recordings where your policy says, and delete them on schedule.

A home-appliance brand adds video to WhatsApp and web chat support.

• Rule: agents may offer video only for "appliance fault" and "installation" issues, after the customer has described the problem in text.
• Flow: the agent sends a one-tap link in the thread; the call opens in the browser with audio on and the rear camera available; no recording by default.
• Outcome: the agent writes a summary and next steps back into the thread — "Drain filter blocked; steps sent; engineer visit not needed."
• Measurement: after three months, video calls resolved installation issues in a single contact far more often than text alone, but billing-related video calls (tried briefly) took longer and were dropped.

Do not fold video into overall support metrics. Measure resolution rate, repeat contact and handle time for video interactions against comparable text-only ones. If video is not resolving faster or more completely, narrow when it is offered rather than adding more of it.

• Where are your media and relay servers, and can I restrict calls to a region?
• Can customers join from a browser with no install or account, on both Android and iPhone?
• How is recording enabled, where is it stored, for how long, and who can access it?
• How is a call linked back to the support conversation and customer record?

For when a media server stops being optional, read video without a vendor. For the queue this escalation lives inside, see the support queue as a system; for recording consent as data, consent architecture.