AI Automation platform

AI Automation platform

AI Automation platform

AI Automation platform

AI Automation platform

AI Automation platform

AI Automation platform

INDUSTRY

AI / Marketing Automation

OWNED

Product, Brand, Design System

LIVE LINK

Overview

Design Lead for an AI platform for business automation. Owned product, brand, and AI platform strategy in partnership with the founders.

Context

Subscription business. One number ran the company: trial-to-paid conversion. And it was stalling.

The easy assumption was the paywall. It wasn't. People were reaching the end of their trial without ever seeing the product do the thing it was built for.

The pre-paywall story was broken.

My role

I led design across product, brand, and platform — and shaped direction with the founders rather than executing a finished plan. Framing the problem was part of the job, not something handed to me. I set the visual system, ran the component library, and personally owned the flows the business lived on: onboarding, payments, and every surface where the AI acted on a user's behalf.

Key decisions

Rebuilt onboarding around first-content activation. Stopped teaching people the product. Started getting them to one finished piece of content in their own brand voice, as fast as possible. That's the moment value stops being a promise and becomes a thing on their screen. Everything before it is friction with better copy.

Set the rule for how the AI behaves. The product wrote, scheduled, and published. The design question for two years was where the line sits between what it does alone and what it asks about. The rule I landed on and defended in every review after: propose quietly, act visibly, and never cross the line where something goes public or costs money without a human. That rule came from watching real behaviour — people forgive a bad draft the second they see it. They never forgive a post that went out at 9am Monday they didn't approve.

Didn't let it over-explain. Narrating every decision is as much a failure as a black box. An AI that shows its working like a schoolchild proving long division isn't confident, it's insecure. Confidence should be readable from the interface, not recited.

Built the design system on intent, not appearance. Functional roles — stroke/primary, stroke/hover, backgrounds, text, icons — instead of a raw palette. A change lands once and propagates everywhere. Structured core → atoms → composed components, moving toward tokens as the system matured.

Repositioned to premium. Rebuilt the visual language and pricing surfaces for higher tiers and enterprise buyers. Defaulted subscriptions to annual.

Outcome

  • Trial-to-paid conversion: ~21% → ~39%

  • Weekly active paying users: ~200 → ~2M+

  • ARR: ~$100K → ~$1.3M

Roughly 10x on revenue and paying users over two years.

The number I watched most nervously was the one that never made it into a slide: people who converted in month one and churned in month two. That's the failure mode where trial-to-paid looks great and the business is quietly getting worse.

What I’d do differently

Lock the AI interaction patterns earlier. We shipped a few one-off agent surfaces before the component system caught up, and retrofitting them cost more than designing them right the first time.

And test the core claim sooner. We treated "personalised to your brand" as obviously valuable. It took months to find out which brand details people actually cared about — turns out "sounds like us" and "sounds like our marketing team wants us to sound" are two completely different products.

akréitor

2026 © Andrew Kréitor

akréitor

2026 © Andrew Kréitor

akréitor

2026 © Andrew Kréitor

akréitor

2026 © Andrew Kréitor

akréitor

2026 © Andrew Kréitor

Andrew Kréitor

2026 © Andrew Kréitor