A reviewed version of the final steps before creating a model. Decided to give more visibility over what’s going on under the hood with meaningful steps instead of a list of AI jargon. On the right, the final step of the new tool walkthrough is introduced.

Main Challenges

Turning complexity into a learnable experience

• Ensure users don’t need data science experience to use the tool.
• Hide the complexity under the hood, without making it look less powerful.

Homework, before changing a single word

• Go through dozens of already existing screens and user journeys within the product.
• Understand a highly complex field of which I knew nothing beforehand.

Writing for a better experience

• Keep the experience consistent in itself and across the different company products.
• When I joined, there was a lot of already identified friction on the tool.

Different versions of a mandatory initial configuration. On the left is how it was before I joined the team, on the right the reworked version.

Revamped configurations

I knew from the beginning that in our tests, users were not figuring out how to complete a set of mandatory configurations that popped up before getting started. It was the first step on the tool and the title was impersonal, the process looked and was cumbersome, we only had an action-based task list and we were using a very heavy tone the whole way through.

To change it, I designed a more welcoming first step in ML Builder:

• Dropped the descriptive title and went for an invitation, welcoming the user to the tool and encouraging them to get started.
• Made this the first step in the journey, instead of labelling it as a mandatory burden users had to deal with before getting to the real thing.
• Give a bigger picture first and let users digest each step in smaller bits later.
• Most of the copy focuses on the benefits instead of on the most troublesome side of configurations.
• The sub-title, descriptions, and main CTA all use the action “set up” instead of using different actions to represent the same concept such as “configuring” or “connecting to”.
• Kept an estimate of how much users would take on each step to better manage expectations and avoid frustration that would inevitably come up later otherwise.
  

In the end, we didn’t get rid of complexity, we just made it easier to understand.

One modal to rule them all.

Creating patterns

To guarantee consistency on all confirmation and cancelation modals, I created a system that could apply to all of them.

• Titles ask the user to confirm the action they selected.
• Buttons are always action-based and answer the question clearly.
• Descriptions provide more information about the action.
• On destructive actions, canceling becomes the secondary action to avoid mistakes.

All opportunities were good ones to guide the user to better outcomes.

In-product guidance

We knew that machine learning isn’t easy for most, so we tried to guide users through each step of the product and didn’t waste any opportunities without adding too much weight to the tool.

• Empty states give a glimpse of what can be found there while also encouraging action.
• Contextual tips raised the success rate through a well-informed happy journey.
• All guidance was focused on first-time usage and would disappear once the action was taken.

Other projects

Main Challenges

Turning complexity into a learnable experience

• Ensure users don’t need data science experience to use the tool.
• Hide the complexity under the hood, without making it look less powerful.

Homework, before changing a single word

• Go through dozens of already existing screens and user journeys within the product.
• Understand a highly complex field of which I knew nothing beforehand.

Writing for a better experience

• Keep the experience consistent in itself and across the different company products.
• When I joined, there was a lot of already identified friction on the tool.

Other projects