A conversational marketing platform: from CPA offer to bot funnel. I designed the visual language and UX of the core modules of a platform that grew from a commercially successful bot into a funded SaaS startup. Sole product designer on the team.
AiSales grew out of a commercially successful bot that sold leads for credit products to the CPA network leads.su. The bot led the network in turnover — its stability earned the team its first investment and a place in the Innopolis accelerator, where the product was packaged.
Startup hypothesis: messengers will replace classic landing pages as a lead generation channel.
When I joined, the product was a set of fragmented screens with no unified visual language.
I was the only product designer on the team.
← swipe the chart →
Three layers of pain
The operator couldn't see the whole funnel at once — each node's settings opened separately. The “what connects to what” context got lost.
Each messenger had its own set of connection steps. There was no unified pattern.
Only conversions were visible, not the dialogue's emotional trajectory — which, in credit funnels, decides whether a user reaches the application form.
Flow Builder — the funnel on a single canvas
In competitors' products, the funnel lived in a tree of nested folders: the operator opened nodes one at a time and lost the “what connects to what” context.
I unfolded it into an open canvas: nodes of different types — traffic sources, bots, brokers, target offers — are linked by clear, visible curves.
A funnel is a graph, not a list: you need to see it whole, not assemble it in your head. What took 4–6 screen switches in competitors' tools reads here at a single glance.
Channels — 8 messengers, one pattern
Each of the eight messengers had its own set of connection steps — the operator had to relearn the process every time.
I reduced them to a single pattern: one connection modal, an identical intermediate screen and a header color-coded to each messenger's brand.
The operator can tell where they are by color alone, without reading labels, and connecting the eighth channel is no different from the first.
Dashboard — metrics and the “emotional background”
Standard dashboards on conversational platforms showed only conversions. But in credit funnels, emotion drops before conversion does: the user gets irritated and leaves, and the numbers only show it after the fact. Together with the CTO, we introduced a “campaign emotional background” layer: a gauge of negative / neutral / positive reactions, counters of atypical events and an engagement index.
This gave the operator an early “the bot is losing the user” signal long before conversion dropped. No competitor at the time had a dialogue-quality layer like this.
CPA Offer Wizard — choosing the offer and the model
Before I joined, launching a campaign was a string of forms: the operator filled in fields without knowing where they were in the process or how the monetization models differed. I built a step-by-step wizard: first choose a CPA offer, then the payout model.
Revenue Share vs. Cost per Action is presented as comparison cards with numbers, not fine print. A step indicator on the left shows where you are in the overall setup flow. Choosing a monetization model is the operator's business decision. The interface should help make it, not just collect data.
How I worked
I joined a project that consisted of a few disconnected screens with no unified visual language, and I was the only product designer on the team. My contribution: the platform's identity and logo; the visual language and design system; UX/UI for six functional modules — Flow Builder, Channels, Dashboard, Segmentator, Tasks and CPA Offer Wizard. I made every decision myself, from wireframes to detailed mockups.
An honest caveat: at that stage of the startup, the team had no research function. Hypotheses were tested by directly observing operators at work and through fast iterations. The Flow Builder concept and the “emotional background” system were co-created with the founder and the CTO. The foundation I designed went into production as Chatbot Funnel Designer in August 2021 — after I had left.
How we taught bots to understand people in 2019
Segmentator is the module where the semantic core was built by hand. The bot brought in thousands of unrecognized phrases, and the operator sat down and sorted through them: grouping questions into intents and answers into groups, tagging entities right in the text. Four thousand seven hundred and ninety-three unrecognized phrases. By hand.
Today an LLM does the same in a couple of hours — but in 2019 it was a product challenge in its own right: design an interface so that a person wouldn't lose their mind on phrase number three thousand. Hence the three-column layout, bulk operations and in-place entity tagging. A genuinely interesting problem, by the way: you were literally building the bot a creditor–debtor dictionary.
RESULT
The platform grew out of a commercially successful leads.su bot, attracted investment and went into production. The numbers above are the work of the whole team; my contribution to them is the design foundation the product was built on:
What I'd do now. With experience in agentic workflows: AI agents in Flow Builder — the operator describes the campaign goal in text and the agent builds the first version of the canvas; automatic training of the semantic core via LLM; a mobile dashboard for operators and a direct link between the emotional background and conversion.
- Designed Flow Builder — a visual canvas instead of the competitors' typical tree list: the entire campaign reads on one screen.
- Introduced the “emotional background” — a dialogue-quality metrics layer atypical for fintech, with an early signal that a user is being lost.
The visuals in this case study are real project mockups from 2019–2020. The final version of the platform launched in August 2021, after I left; I was not involved in shipping it.