Phrase: Custom AI 
Cleaning Filters

We audited the flow and quickly found the biggest issue: the content was trying to explain everything. Users were met with long paragraphs that walked them through the underlying data science, but all they really wanted was to clean their dataset and move on.

There were multiple steps, multiple filters, and multiple opportunities for user confusion. We knew we needed to streamline the content, rethink the structure, and design with the job-to-be-done in mind.

The biggest friction point was clarity. Filters had vague names, overly descriptive copy, and inconsistent input behaviours.

We rewrote every single sentence with two goals in mind: clarity and brevity. Paragraphs were turned into single lines. Multi-step instructions became in-line guidance. Every word had to justify its place.

Content related to user interaction points on inputs was also reworked to provide more logical, straightforward instructions.

We mapped the ideal flow, realigned the IA, and defined a new structure for each filter:
- Clear title
- One-line description
- Simple editable input
- Immediate feedback

In the re-cleaning flow (where users review and edit their original cleaning settings), the layout was particularly difficult to parse.

The original designs had descriptions sat below the input fields. Existing filter values were hidden under new editable ones. Re-cleaning involved multiple repeat steps with no clear guidance.

We couldn’t achieve the ideal state (a single editable field showing pre-set values) due to dev constraints. Instead, we compromised:
‍
- Moved filter descriptions above inputs for faster scanning
- Displayed new values in a persistent banner at the top of the filter
- Included quick actions for editing and undoing

These adjustments gave users a way to track changes at a glance without scrolling or guessing what they’d already modified.

We delivered a clean, intuitive experience that helped users move through the flow confidently. No dense copy. No Help Centre dependency. Just a logical structure, clear content, and strong affordances.

We also provided in-product guidance that spoke directly to users’ goals, whether it was removing outliers or fine-tuning quality thresholds.

Usability testing confirmed the new flow was dramatically easier to use. Participants reported that the filters “just made sense,” even if they had no prior knowledge of data cleaning.

The final experience significantly reduced user reliance on support articles, lowered the barrier to entry, and allowed users to move through a once-daunting flow with clarity and ease.