Skip to content

Streaming Lifecycle Message Agent

About 21% higher watch engagement against a holdout group, and open rates up about 40%.

Request an intro

An agent writes every onboarding, churn-prevention, and win-back message for each subscriber of a German streaming service at send time, from about 20 viewing and preference signals.

What the agent does

Decides for each subscriber which viewing signals should govern the message, then writes and orders it at send time with no human review.

Brand
German streaming service (name withheld)
Built by
The streaming service's engagement team, with its implementation partner, on Braze
Contributed by
Braze (Agent runtime and message assembly)
Audience
B2C
Category
Marketing & PersonalizationContent & Experience
01
The problem

Why this exists

The service has millions of subscribers across big screen, web, and mobile, and had already built strong segmentation and trigger logic. Its engagement team put the gap plainly: the right users got the right trigger at the right time, but the message itself stayed generic.

Read more

In streaming that gap is commercial. Retention depends on reconnecting a subscriber to content they will actually watch before their interest lapses. Churn prevention runs inside a three-day cancellation window, so a generic message is a wasted send at the moment the subscription is decided.

Rules could not close it. About 20 signals with several states each cannot be enumerated as branches, and the content slate changes weekly. The signals also disagree, and new subscribers in onboarding have almost no viewing history to work from.

02
Outcomes

What changes when this is in place

  • About 21% higher watch engagement in the low-engagement journey, measured against a holdout group. Watch engagement is the leading indicator of a retained subscription.
  • About 40% higher open rates for agent-composed messages than for standard messaging.
  • One campaign definition replaces a hand-authored variant set and keeps producing individually written messages as the content slate turns over, with no per-variant maintenance.
Read more
  • The signal hierarchy keeps messages relevant for subscribers with thin viewing history, which is what let the agent go live on onboarding, the hardest journey to personalize.
03
How it works

Architecture

Marketers set the journey, the objective, and the brand guardrails. The agent runs inside the send path of three lifecycle journeys: onboarding, churn prevention, and win-back.

Read more

For each message it reads about 20 signals for that subscriber, including strongest and top-five genre affinities, 7-day and 30-day viewing intensity, format diversity, first-viewed genre, sports preferences, subscription history, and preferences declared at onboarding. It resolves conflicts between them with a defined hierarchy: strong recent behavior first, then softer signals and declared preferences, then brand-appropriate defaults.

It then writes the subject line, preheader, headline, recommendation framing, and editorial copy, and decides the order of the content blocks. No person reviews individual messages.

The agent returns structured, named components rather than a finished message. Liquid templating and custom HTML layouts in Braze assemble them into email and in-app messages at send time, which keeps reasoning separate from presentation.

The inputs come from the service's own systems: the subscriber data layer, which resolves viewing telemetry from TV, web, and mobile into derived profile attributes, and the content recommendation engine. Prompts were first developed in an enterprise ChatGPT instance, then moved into BrazeAI Operator for refinement. The next version reasons over specific titles and upcoming episodes rather than genres.

Viewing telemetry from TV, web, and mobile flows into the subscriber data layer, which holds a unified profile with about 20 derived signals plus preferences declared at onboarding. That profile, the content recommendation engine, and the journey, objective, and guardrails set by marketers all feed the reasoning layer, an agent on BrazeAI Agent Console. The agent arbitrates conflicting signals per subscriber, writes the subject, preheader, headline, framing, and copy, and orders the content blocks. It passes structured, named components to the assembly layer, Liquid templating, and custom HTML layouts, which renders an email or an in-app message at send time.

Viewing telemetry from TV, web, and mobile flows into the subscriber data layer, which holds a unified profile with about 20 derived signals plus preferences declared at onboarding. That profile, the content recommendation engine, and the journey, objective, and guardrails set by marketers all feed the reasoning layer, an agent on BrazeAI Agent Console. The agent arbitrates conflicting signals per subscriber, writes the subject, preheader, headline, framing, and copy, and orders the content blocks. It passes structured, named components to the assembly layer, Liquid templating, and custom HTML layouts, which renders an email or an in-app message at send time.
Tech stack
BrazeAI Agent ConsoleBrazeAI OperatorBraze Liquid templatingCustom HTML code layoutsStreaming service subscriber data platformStreaming service content recommendation engine
Next step

Request an intro

Braze is behind this solution. Ask about personalizing what the message says, not only who receives it, and the MACH Alliance will put you in touch with the person who can answer.

The MACH Alliance passes your details to a named contact at Braze who knows this work and can answer your questions. We aim to reply within 2 business days. Your details are handled under the MACH Alliance privacy policy.

Not ready to talk to anyone? MACH Insider is the Alliance mailing list. New solutions, program openings, and the research behind them, once a month.

Join MACH Insider

Everyone credited on this solution

  • BrazeAgent runtime and message assembly

Every organization named here has a page listing the rest of its work on the Hub.

Keep reading

Related solutions

Verification

Contributor submission

The detail on this page comes from the contributor's own account of their own work. The MACH Alliance reviews a submission before publishing it and does not independently verify the figures in it.

Where this came from

Published from the winner's case study and the 2026 MACH Impact Awards announcement. Solution detail is drawn from Braze's 2026 MACH Impact Awards entry. The brand is not named at Braze's request.

Status: Live in productionAdded Oct 26, 2026A listing is an editorial decision, not a MACH Alliance certification.
Put a solution forwardBack to the Hub

Built something worth showing?

Put your agentic solution in front of the enterprise leaders who use the hub to decide what to build next. Submissions are open to MACH Alliance Members, Collaborators, or Ambassadors.

Put a solution forward