Use case · AI content creation

AI personalised video that respects the data it uses.

Personalised video works as a fixed, approved template with variable slots filled from customer data: name, product, plan, usage, local offer. AI makes the voice, visuals and languages practical at scale. The limits are legal and human: use only data you have a lawful basis to process, the minimum needed, and nothing that surprises the viewer about what you know. Personalisation that feels like surveillance does more harm than generic video.

Templates, variables and fallbacks

A personalised video is a template with slots, not a thousand bespoke films. The master script, visuals and music are approved once; the variables, such as first name, product owned, renewal date or nearest store, are inserted per viewer. Most failures come from the data: names with apostrophes, accents or non-Latin scripts, very long names, empty fields, and joke entries typed into forms. Define a fallback for every variable, such as a generic greeting when a name is missing or doubtful, filter offensive entries, and check how the synthetic voice pronounces names, because a mispronounced name undoes the point of personalising. Test with the messiest real records you have, not a tidy sample.

Lawful basis and data minimisation

Under the GDPR and UK GDPR, every data point in a personalised video needs a lawful basis, and Article 5 limits processing to the data necessary for the purpose. The email or message that delivers the video also falls under ePrivacy rules, such as the UK's PECR. Use data the customer knowingly gave for a relationship they recognise: a name, a product they bought, a plan they hold. Keep special category data, such as health, religion or sexual orientation, out of marketing videos entirely. Never put personal data in video URLs; use opaque tokens and set an expiry on hosted videos. Keep personal data out of render logs, and delete per-viewer files when the campaign ends.

Rendering and delivery at scale

There are two ways to deliver. Pre-rendering creates a file per recipient before sending, which is simple to check but costly to store and hard to change once rendered. Rendering on demand builds the video when the viewer opens the link, which keeps data fresh and storage low but needs infrastructure that copes with the spike when a campaign lands. Either way, review a structured sample: every variant path, every language and the edge cases from the data. Keep it short, because a personalised opening does not hold attention through a long generic middle. The studio produces the approved master, its visual variants and voice-over in 30 languages with timing preserved, and approvals with an audit trail cover the template and every variant path.

Where personalisation turns creepy

The test is whether the viewer would expect you to know what the video shows. A thank-you that names the product they bought is welcome. A video showing their street, referencing their browsing history, their location last weekend, a health condition inferred from purchases or a life event they never told you about feels like surveillance, even where it is technically lawful. The GDPR itself asks, for legitimate interests, whether processing fits the person's reasonable expectations. Keep personalisation to data the customer gave you on purpose, say where the data came from when it is not obvious, and make opting out easy.

Updated 25 September 2026

Questions

AI personalised video, answered.

Do we need consent to send personalised videos?

It depends on the data and the channel. Personalising with data customers gave you, for a relationship they recognise, can often rest on legitimate interests, but the email or message delivering the video falls under ePrivacy rules such as the UK's PECR, which usually means consent or the existing-customer soft opt-in. Keep special category data out.

How do personalised videos handle unusual names?

Badly, unless you plan for them. Define rules for apostrophes, accents, non-Latin scripts, very long names and blanks, fall back to a generic greeting when data looks wrong, filter offensive entries, and check pronunciation in every language the voice-over uses.

Can AI personalise the presenter's words for each viewer?

With a real presenter, only within what they agreed to in writing, and synthetic lines should be disclosed where the presenter looks real. A safer pattern is to keep the presenter's recorded script fixed and personalise the on-screen text, graphics and product shots around them.

Get started

See your AI studio generating this week.

Thirty minutes. A live studio in your colours, your brand hub loaded, a real campaign brief.

See plans