Sources

    Nobody Could Use Our ML Engine

    ENTERPRISE AI · CASE STORY

    Nobody could use our ML engine

    Making machine learning usable for people with no ML background, without dumbing it down for the experts.

    4 min

    Imagine a professional espresso machine with no labels.

    It can make anything. Only the barista dares to touch it.

    WebFOCUS had a powerful machine learning engine. Adoption was low.

    To train one model

    12–15 clicks

    All inside Reporting Server: the + menu, Data Flow, a new browser tab, drag a dataset onto the canvas, Train Models, drag a model type onto the canvas, then find a hidden play button.

    Business users couldn't touch it.

    Two people had to be served by the same screens:

    BUSINESS ANALYST

    Zero ML background. Needs to get through it and trust the result.

    DATA SCIENTIST

    Wants depth and control, and not to feel spoon-fed.

    Rule 1

    Ask why behind the small request.

    It started as a request for one explainability popup. Asking why it was needed turned it into a revamp of the whole ML workflow.

    Rule 2

    Remove the dead end first.

    No results generated. Click play to train the model.

    A screen that tells you nothing happened, and hides how to make something happen, was the first thing to go.

    Rule 3

    One path, not two modes.

    The principal data scientist asked: "Are you trying to make this dumb?"

    No. Sophistication stayed. It became legible: one path for the novice and the expert, with explanations in context. Not a dumbed-down UI. An educated one.

    PRINCIPLE

    Reveal depth gradually

    Show the essentials first and let people open the advanced options when they need them.2

    Rule 4

    Four clear steps.

    Problem type

    →

    Target

    →

    Predictors

    →

    Hyperparameters

    Full-screen views, side-step layouts and two columns were all tried. A popup wizard won, because it solved the problem without asking the legacy product to become something it isn't.

    Rule 5

    Put the control where the eye already is.

    The threshold slider sits exactly under the point on the chart it moves.

    PRINCIPLE

    Make clear why the model did what it did

    People trust a result more when they can see how it was reached.1 Moving the slider and watching the chart respond teaches it.

    Rule 6

    Meet the behavior people already have.

    Right-click a dataset

    Predict Data

    Generate Insights

    Ask a Question

    In this product, right-clicking is religion. So Predict Data went there, and two more AI features later used the same entry.

    Rule 7

    Defend the design with evidence.

    When the data scientist pushed to bring back the old canvas, user research and session recordings showed two engineering directors and the principal data scientist why reverting would break the workflow.

    WHAT CHANGED

    Design done. Built and test-complete.

    ✓

    12–15 clicks → a 4-step wizard, as few as ~6 clicks

    ✓

    4 of 4 technical subject-matter experts finished with zero help, in usability testing

    ✓

    3 AI features adopted the right-click entry

    ✓

    The explainability popup shipped in WebFOCUS 9.2

    The testing validated the built redesign. It isn't production adoption.

    BEHIND THE STORY

    "I taught him machine learning through the UI."

    I owned the redesign end to end. Our design director, with no ML background, learned ML from the UI.

    "The clarity of her designs, in spite of the underlying data science and machine learning complexity, is impressive and has greatly contributed to the success of our products."

    Marcus Horbach, Principal Data Scientist

    What a mind-bending project.

    CHEAT SHEET

    Making powerful tools usable

    Ask why behind the small request.

    Remove the dead end first.

    One path, educated, not dumbed down.

    Few, clear steps.

    Controls where the eye is.

    Meet existing habits.

    Defend with evidence.

    READ MORE

    The full case study.

    The wizard, the confusion matrix, and the fight to keep it.

    READ THE CASE STUDY →

    Enterprise story 02: "Nobody could use our ML engine." Designed by Anuja Nimmagadda for WebFOCUS at Cloud Software Group.

    Sources

    1. Amershi et al. (2019). Guidelines for Human-AI Interaction. CHI 2019 (Microsoft HAX).
    2. Nielsen Norman Group. Progressive Disclosure.