Vlad Lobchanka

Smartcat

August 2024 – January 2026

Turning more trial users into paying customers

In the growth team of an AI translation platform, I made the product’s AI credits clear to new users. Conversion to self-service purchase rose 17%, and more people signed up with a work email.

Visit smartcat.com

Role
Product Designer, Growth
Team
Growth team of five: a product manager, an analyst, two engineers and me
Stack
Figma · Storybook · Mixpanel · Grafana · Claude (code prototypes in my last six months)
Smartcat sign-up page: tabs for Business, Translation agency, University and Freelancer, a business email field, Microsoft and Google sign-in buttons, and on the right “Transform any content to any language with AI”.
The new sign-up asks for a work email and has its own route for agencies, universities and freelancers.

+17%

conversion to self-service purchase

5% → 31%

sign-ups with a work email

+5.2%

conversion to subscription

Context

Smartcat is an AI translation platform that learns from every translation it does for a customer. I joined its growth team in August 2024 as the product designer. Our part of the funnel ran from sign-up to the first translated file.

The problem

New users did not understand Smartwords, the credits that every AI action in the product spends. Onboarding never mentioned them, so when the 15,000 trial credits ran out, the product simply stopped. Support kept getting the same message, “not enough Smartwords, what does it mean?” And in sign-up, only 5% of new users gave a work email.

What I did

Explaining the AI credits up front

I added a welcome screen that explains the rules of the trial. Every AI action uses Smartwords, and the trial includes 15,000 of them. After a file upload, the product now shows the cost of the translation before anything is spent. Conversion to self-service purchase rose 17%, support got fewer questions, and the sales team said it saved them time.

Welcome modal “Hello, Frank! Welcome to your Personal Free Trial!” with three rules: all AI actions use Smartwords; you have 15,000 free Smartwords; free trial Smartwords expire in 15 days.
The welcome screen explains the trial rules before the user spends anything.
Analysis complete: the project contains 7,982 words, and the user will spend 7,982 Smartwords to translate it, with a button to start the translation.
After a file upload, the product shows how many Smartwords the translation will cost.
Project needs more Smartwords: a balance of 15,000 against about 17,892 words, and a note that the product will cover the difference this time.
When the trial credits are not enough, the product covers the difference once and says so.
Not enough Smartwords to translate remaining files: a balance of 15,000 Smartwords against about 17,892 words, with buttons to start over or upgrade now.
After that, the screen shows the balance next to the word count and offers an upgrade.

Building sign-up around work emails

I rebuilt sign-up around work emails and blocked personal addresses. Translation agencies, universities and freelance translators each got a route of their own. Sign-ups with a work email went from 5% to 31%. Conversion to subscription rose 5.2%.

The old sign-up: one email field for everyone, or a Google, LinkedIn or ProZ account.
Before the change, one form accepted any email address or a social account.
Sign-up, Translation agency tab: “Start translating your projects now”, an agency email field, Microsoft, Google and LinkedIn sign-in buttons, and a preview of the project workspace.
After the change, translation agencies sign up on a route of their own.
Sign-up, Freelancer tab: “Global network of expert translators”, an email field and a preview of a translator’s task list.
Freelance translators have their own route too.

Finding out where new users struggled

The team wrote its own hypotheses, and the growth director chose the order of tests. To learn where new users struggled, I designed a satisfaction survey inside the product, with the answers sent to our Slack. The answers showed hidden problems in onboarding and communication. For example, users wrote that they did not know how many free words they had left. This supported the Smartwords change. We also used transcripts of interviews run by our lead designer, and an AI agent that marked important points in client calls.

A survey over the translation editor: “Hi Karl, how do you like using Smartcat for now?”, a 0 to 10 scale and a field for a comment.
The survey appears after a user’s first translated file, and the answers go to Slack.

The design system, reviews and code

Smartcat has a design system in Figma, built from small parts up to full page templates. Engineers use the same components in Storybook, and I designed within it. I also reviewed colleagues’ designs, and a design went to engineering only with three approvals out of five designers. In my last six months, I built prototypes in code with Claude. Engineers reviewed about ten of my merge requests and merged them into production. Most of them were interface fixes in sign-up and onboarding.

Three paywall popups: broad assignment configuration, upgrade to Starter to unlock assignment templates, and upgrade to Unite to unlock automatic invitations.
These paywall popups are built from the parts of Smartcat’s design system.
A “Template description” field in three states: empty with the hint “Maximum 500 characters”, over the limit with the error “Please reduce text by 15 characters”, and corrected.
The system documents every state of a field, from the hint to the error and back.

First steps towards an AI agent

In my last months, Smartcat began to turn the product into an AI agent that does the translation work for the user. I worked on the growth side of this change. I designed how a new user meets the agent in onboarding and gives it the first task. Before the agent starts, it shows what the task will cost in Smartwords, as the rest of the product already did.

Results

  • Conversion to self-service purchase rose 17%.
  • Support got fewer questions about Smartwords, and sales spent less time explaining them.
  • Sign-ups with a work email went from 5% to 31%.
  • Conversion to subscription rose 5.2%.
  • Survey answers showed hidden problems in onboarding and communication.

Who did what

  • Me: the Smartwords onboarding, the new sign-up, the survey, the agent’s first steps for new users, design reviews, and code prototypes in my last six months.
  • The team: the hypotheses, which we wrote together.
  • Front-end and back-end engineers: built the new sign-up, the welcome screen and the cost estimate.
  • Analyst: set up the tracking and read the test results in Mixpanel and Grafana.
  • Growth director: chose which hypotheses to test, and in what order.
  • Lead designer: ran the customer interviews whose transcripts we used.

What I learned

  • Smartcat started to build an AI agent for customers before the team used agents in its own work. Today I would start with agents in the team’s own work, and let that experience shape the product.
  • People need to see what an AI action costs before they use it. Once the trial rules were clear, more users bought on their own.