AI-Powered Onboarding Email Personalisation
A project management platform's generic drip sequence wasn't converting trial users to paid accounts. We built a behaviour-aware system that generates contextualised emails — and lifted conversion by 42%.
The Situation
The platform attracted around 2,800 free trial signups per month, but only 19% were converting to paid. The product team suspected onboarding was the bottleneck — users weren't reaching their "aha moment" fast enough.
Their existing onboarding was a five-email drip sequence, identical for every user. A solo freelancer received the same emails as a 50-person team lead. Someone who signed up to manage client projects got the same tips as someone tracking internal sprints.
The emails were well-written, but generic — and open rates had been declining for three consecutive quarters. The team had tried segmenting manually, but with five segments and five emails each, they were looking at 25 separate flows to write, maintain, and update every time the product changed. Their two-person marketing team couldn't keep up.
"We don't need more emails. We need smarter ones. Something that actually knows what the user has been doing and meets them where they are."
What We Built
An automation pipeline that observes what each user does during their first 48 hours, classifies them into behavioural segments, generates contextualised onboarding emails using AI, and sends them through a human review gate before delivery.
How It Works
Signal Capture
A webhook listener captures key behavioural events in real time — account setup completion, first project created, teammates invited, features explored, and where users got stuck. By the 48-hour mark, we have a behavioural fingerprint for each user.
Behavioural Segmentation
Instead of asking users who they are (which adds friction and is often wrong), we let their actions tell us. A custom scoring script classifies each user into one of five segments based on their behaviour patterns.
Custom logicAI Email Generation
Each email is generated fresh based on the specific user's actions. The AI receives their segment, specific behaviours, and a strict template framework — subject line, personalised opening, one paragraph of guidance, one CTA. Guardrails keep output consistent and on-brand.
Custom prompts + validationHuman Review Gate
Generated emails appear in a Slack review queue. During the first two weeks, the team reviewed every email. After building confidence, we shifted to 85% auto-send with 15% random sampling plus keyword-triggered reviews.
Adaptive Sequencing
Unlike time-based drip campaigns, the next email depends on whether the user completed the suggested action. If they did, the journey progresses. If they didn't, the system tries a different angle on the same goal.
Custom logicPerformance Tracking
Weekly automated reports compare performance across segments, identifying which email variants work best. Data feeds back into prompt refinement and segmentation tuning on an ongoing basis.
The Five Segments
Rather than demographic personas, we defined segments by what users actually do in the product.
Results
Team Champions showed the highest conversion rate at 41% and responded best to fast follow-up within 4 hours of signup. Ghosts had a 12% re-engagement rate from AI-generated win-back emails, compared to 4% from the previous generic sequence. Integration Seekers had the shortest time-to-paid at 3.2 days once they confirmed their tools were supported.
The Tech Stack
No new platforms for the team to learn. The system plugs into what they already used.
What Made This Work
Behaviour over demographics. We never asked users who they were — we watched what they did. A CEO and a junior PM might land in the same segment if they exhibit the same patterns. More accurate, and no signup friction added.
Personalisation with guardrails. The AI doesn't freewheel. Every email follows a proven structure, references only real user actions, and passes automated validation before reaching the review queue. This is what makes AI-generated content reliable at scale.
The human review phase built trust. The marketing team was understandably nervous about AI writing customer emails. The two-week full-review period let them see the quality firsthand and build confidence before shifting to sampling.
Measurement from day one. Every email, action, and conversion was tracked from the first send. Weekly reports gave the team confidence and clear direction on where to improve.
Engagement Timeline
Have a workflow that needs to get smarter?
Let's talk about what AI automation could do for your onboarding, operations, or growth.
Get in touchThis case study represents a composite engagement based on real automation and product strategy work. Client details have been anonymised.