Sources

    Three AI Engines, Three Separate Homes

    ENTERPRISE AI · CASE STORY

    Three AI engines, three separate homes

    The AI worked. People just couldn't find it. Designing one home for three AI features.

    4 min

    Imagine a store whose three best departments have no signs.

    One is behind a door, one is up a back staircase, one only opens if you know the knock.

    WebFOCUS had three AI engines: natural-language questions, automated insights, and machine learning.

    !

    Ask a question and Insights: buried several clicks deep in menus, then inside a popup.

    !

    Machine learning: reached by right-clicking a dataset, an entry added in the earlier ML revamp. Before that, it was hidden inside Reporting Server.

    Two paths, two mental models.

    Powerful. Paid for. Impossible to find.

    Rule 1

    Diagnose before you solve.

    THE FIRST READ

    The models need work.

    THE REAL PROBLEM

    People couldn't find them.

    The engines weren't the problem. Their addresses were.

    Rule 2

    Architect before the tickets exist.

    The workflows across all three engines were designed first. The tickets were written after the mockups.

    Rule 3

    Integrate. Don't rebuild.

    3 engines

    →

    1 home

    The engines stayed exactly as they were. Only their addresses changed.

    Rule 4

    Pick the data once.

    BEFORE

    Pick the data three times.

    Every tool remembers its own dataset.

    THE REDESIGN

    One action, all three tools.

    One action applies the same dataset everywhere. After testing with experts, each tool can still take its own.

    Rule 5

    Outcome first. Tool names never.

    Describe your goal. The hub routes you to the right engine.

    What do you want to find out?

    And a "What are…?" explainer that takes the fear out of AI.

    PRINCIPLE

    Recognition over recall

    Don't make people remember which tool does what. Show them the outcome they can choose.1

    Rule 6

    Win the room with evidence, not taste.

    List views were the exact reason these features were invisible. Large tiles that describe outcomes, backed by industry precedent and interaction logic, won the argument.

    Newest in a room of architects who'd been there 20 to 30 years. Won it with evidence.

    Rule 7

    Iterate until it lands.

    1

    Too dense

    2

    Too passive

    3

    Competed with the main navigation

    4

    Landed

    WHAT CHANGED

    Three menus, one hub.

    ✓

    Three menus into one hub

    ✓

    Zero new capabilities, just findable ones

    ✓

    Four iterations to land the page

    These are design outcomes.

    BEHIND THE STORY

    I'd already redesigned each engine on its own.

    I had taken over and redesigned natural-language questions, Insights and the ML workflow one by one. Each sat in its own menu, each barely found.

    So my PM and I pitched one home for all three, and I owned the design end to end.

    Not more features. One architecture.

    CHEAT SHEET

    Making AI features findable

    Diagnose first. Low use isn't always a model problem.

    Architect before tickets.

    Integrate, don't rebuild.

    Pick the data once. One action for every tool.

    Outcomes, not tool names.

    Win with evidence.

    Iterate until it lands.

    READ MORE

    The full case study.

    The routing fight, the four iterations, and the architecture.

    READ THE CASE STUDY →

    Enterprise story 03: "Three AI engines, three separate homes." Designed by Anuja Nimmagadda for WebFOCUS at Cloud Software Group.

    Sources

    1. Nielsen Norman Group. Recognition vs. Recall in User Interfaces.