case study
Automated timesheets with
activity tracking and AI analysis
At Timely, we transformed the tedious task of time tracking into an effortless, AI-driven process, enhancing productivity and billing accuracy for thousands of businesses worldwide.
Company
Timely
Role
Lead Product Designer
Established
Norway, 2013
Funding
$20 M
Customers
5,000+
An interface that is sleek and simple, yet surprisingly powerful and customizable to support the diverse needs and requirements of solo freelancers, SMBs, and enterprise workspaces.
I redesigned the Timely activity tracker for Mac, Windows and Linux, built with Tauri, utilizing the best of native recognizable patterns with a custom cross-platform shell. Extra focus on easy installation and guiding users through the required system permissions.
Designing in the age of LLMs
From skepticism to expectations
Timely automates the tedious parts of time tracking. Before ChatGPT’s 2023 breakout, most AI lacked inherent trust and needed careful onboarding and transparency. In consumer products like Netflix, YouTube, or Google, recommendations already felt natural. But in a B2B productivity context where employees and managers rely on accurate records, the bar for trust was far higher. From ongoing research, I learned users had to understand how a recommendation was produced; accuracy alone was not enough.
Our ML grouped computer activity into time entry drafts that users approved or rejected. The system improved with use, but it had an onboarding gap. New customers often saw few drafts during their trials, meaning they rarely experienced the automation benefits in time. That gap hurt conversion.
ChatGPT reset expectations. People arrived with strong mental models and treated anything less as outdated. Manual approvals felt slow. Paradoxically, we no longer had to show exactly how conclusions were reached since default trust had increased. If results looked plausible, few users stopped to double-check. That shift created both opportunity and responsibility.
In this new landscape, we can ship automation that once required heavy trust building. But we also have to add guardrails so users are not harmed by confident errors.
At Timely we leaned in. Instead of recommending individual entries, we proposed whole days or weeks at once. We reused our proven ML pipelines and layered in LLMs for semantic matching and summarization. The impact was meaningful automation for new customers within minutes and new workspaces seeing value by day one. Links obvious to a human, such as a document and its project, became legible to the system.
When I left the company in 2025, a whole range of even more powerful AI features were already under development. I am genuinely excited to one day use them myself once they are released to the public.
The color palette was created using a custom tool built with Leonardo, for maintained contrast values a gorgeous shades and tones built around the established 2013 visual brand.
Example of how a single translucent color token ensures consistent contrasts across UI layers, simplifying implementation for engineers without needing multiple surface tokens.
Personal highlights
Over my four years at Timely, I've had the chance to work on almost every aspect of the product. Here are a few aspects that I'm extra proud of being part of.
🥳
Recruiting and leading the remote product design team
I recruited two product designers—writing the ad, running the process, and creating a fair design challenge. As their mentor, I aligned them with our culture, supported their growth, and balanced creative freedom with consistency. Watching their confidence grow was a peak moment for me at Timely.
🔥
Supercharging onboarding with a redesigned desktop experience
The old tracker flow was full of dead ends, blocking users before reaching Timely’s value. I redesigned it for with guided system permissions, automated OAuth, and a modern cross-platform shell. The result: 2× installs, +50% retention, and +113% trial conversions.
💙
Translucent color tokens is the GOAT
Gray-scale interfaces often create muddy contrasts as components shift between hover and active states. I solved this by introducing translucent color tokens, letting surfaces stack while preserving contrast and structure. It took careful tuning, but now most of Timely’s components use them. It feels great and is so much easier to work with.
Overview of Timely components
