05 — 2023
Onboarding & ActivationAI campaign generator
Replacing the empty state with a personalised, ready-made campaign: a full marketing campaign drafted from a short intake.
- Role
- Product manager
- TEAM
- 3 IT teams, 7-8 developers (front-end and PHP) with QA, UX and design experts, and data analysts; legal and compliance as SMEs on generated-content risk
Goal
Give new users a hand-guided start instead of a blank canvas: reduce cognitive load and choice paralysis at the moment of highest drop-off by offering personalised, ready-made draft content rather than empty state screens.
Context
New accounts landed in a powerful but empty product. Building a first campaign meant making a dozen decisions and a ton of editorial work: audience, structure, copy, design, palette, before seeing any value, and many accounts simply stopped there.
What I did
- Designed the intake as the product: a short set of plain-language questions (field of work, audience, product, tone) that carries enough signal to generate something a user recognises as theirs.
- Owned prompt design and supervision for the generation flow — the wording, the guardrails, and how brand and tone inputs shape output. OpenAI was selected by the architects as the only mature, reasonably safe and affordable option at the time.
- Scoped the generated output as a complete campaign — landing page, welcome email and first newsletter — so the user gets a working funnel, not a snippet.
- Added brand and palette controls (including logo upload) so generated assets look like the customer's brand from the first draft.
- Kept every generated asset editable and publishable in place, so the AI drafts into the existing editors rather than into a dead end.
Outcome
ShippedShipped and announced publicly. Time to first email dropped from over three hours to under three hours, and the share of users sending within their first seven days rose from below 15% to around 20%. Accounts that used the AI campaign generator on their first trial day reached 25%. Directionally, this translated into roughly a 10% lift in users reaching value.
Stack & methods
Visuals
