Microsoft IDA · Client work · 2020

A chatbot that teaches a research tool while you use it

How I designed Aida, an assistant inside Microsoft Insights and Discovery Accelerator (IDA), so new users learn the tool on the real screen instead of in a manual.

Role: Product designer
Year: 2020
Tools: Figma, Azure, Fluent, PowerPoint
Status: Client work

6 min read

Welcome to Microsoft Insights and Discovery Accelerator (IDA). Aida says hi, I can help you get started with IDA, and asks what brings you here: find documents fast, connect ideas across sources, or upload my own files, with Skip for now and Start tour — Aida’s welcome: pick a goal, then take the tour or skip it.

The 30-second version

Put the guide inside the tool

Problem

IDA had no onboarding. New users found the search process hard to understand and couldn’t tell where to find things.

My role

Product designer, end to end: interviews, synthesis, a competitive review, a MoSCoW, a journey map, sketches, mid-fi, the identity and the final UI.

Key decision

An assistant on every screen, instead of a one-time tour or a help center in another tab.

Outcome

A full design for Aida, from research to final UI: onboarding by goal, plain-English answers, and visual aids on the real interface.

Context

A powerful tool with no front door

IDA, Microsoft Insights and Discovery Accelerator, is a research platform built on Azure AI for searching large sets of documents.

It had no onboarding. New users met a search bar with its own syntax, such as exact phrases and words near each other, and nothing to explain it. The brief was onboarding plus a chatbot: clear to new users, and out of the way for people who already knew IDA.

The IDA dashboard: a sidebar with Dashboard, Datasets, Saved Searches, Portfolios, Reports, Shares and Timeline, a single search box, and empty Recent searches and Portfolios panels — Fig. 1What a new user saw: one search box, and nothing to explain it.

My role

Product designer, from interviews to final UI

Role: Product designer
What I did: Research. User interviews, then a synthesis of pain points and motivations, and a persona.

Competitive review. Westlaw, JSTOR and LexisNexis, on five features.

Prioritization. A MoSCoW for the assistant.

Design. A journey map, sketches, annotated mid-fi, Aida’s identity and the final UI.
Tools: Figma, Azure, Fluent, PowerPoint

Research

Show me, on the screen, when I ask

From my interviews and synthesis. The persona was a journalist who needed quick answers and connections between sources, and found new technology hard.

Pain · the search processHard to understand how searching worked.
Pain · finding thingsUnclear how to locate information.
Wanted · visual aidsA chatbot that onboards you with a visual aid.
Wanted · connectionsCompare, contrast and cross-reference articles, with recommendations like Amazon’s.

Motivation

Step-by-step onboarding means less time spent learning a new research platform, and a better-informed user.

Competitive analysis

Help existed, but never on the screen you were stuck on

Three research platforms on five features, from my 2020 review. Limited is shown as Partly.

Reviewed in 2020 · 3 platforms

Westlaw (Thomson Reuters)
JSTOR
LexisNexis

How three research platforms supported new users in 2020, and what IDA aimed for
Product	Chatbot	User onboarding	Chatbot on all screens	Visual aids	Voice control
WestlawLegal research	Partly	Yes	No	No	Yes
JSTORAcademic research	Partly	Partly	Partly	No	No
LexisNexisLegal and news research	No	Yes	No	No	No
IDA with AidaThe target	Yes	Yes	Yes	Yes	Yes

WestlawLive chat and phone support, with a long wait. The chat opens on a new page. A tour for onboarding, but features are hard to find.
JSTORThe chat is a person, 8:30 AM to 4:15 PM EST. Onboarding is articles under Support. Visual aids only in support articles, not on the site.
LexisNexisNo chatbot: it points you to video trainings or a phone number. A training center for onboarding. No visual aids.

The gap

Nobody had visual aids, or a chatbot on every screen.

Help meant a separate page, office hours or a phone line. Our answer: an assistant on every screen that points at the real interface.

Prioritization

What the assistant had to do

My MoSCoW, from the research and the review.

MustUse natural language. Be clear to new users and unobtrusive to existing ones. Use clear visual aids. Be accessible from any page.
ShouldSupport all features. Give error messages and explanations. Match IDA’s visual language.
CouldVoice control and voice prompts. Help from the sign-in page.
Won’tA live support representative. Aida had to stand on its own.

The key decision

A tour, a help center, or a guide inside the tool?

Three ways to onboard new users, checked against the musts.

Tour

A tooltip tour at sign-up

Step new users through the screen once, like Westlaw’s tour.

Sketch: a dashboard with a tooltip reading tip 1 of 9, this is the search bar, and a Next button

Pros: Familiar, and quick to build.
Cons: Shown once, then gone. It can’t answer the question you have on day ten.

Help center

A help center with articles and videos

Write it all down and link to it, as JSTOR and LexisNexis did.

Sketch: a help center page of article lines, with a note that it opens in a new tab

Pros: Complete, and easy to keep up to date.
Cons: It lives somewhere else, so users leave the work to learn the work.

AssistantChosen

An assistant on every screen

A chatbot inside IDA that onboards by goal and answers in place.

Show the original sketch or the polished redrawOriginalPolished

Hand sketch: a landing page with a chatbot area at the side, and a closed chatbot area below it — A landing page with the chatbot at the side, sketched in 2020.

The IDA screen with an Aida panel down the right-hand side — The IDA screen with the Aida panel, redrawn.

Pros: There on the screen you’re stuck on, and it can point at the real UI.
Cons: The most design work: tone, states, and what to do when it can’t help.

The decision

Build the guide into IDA, on every screen

Why

Only the assistant met the musts: natural language, visual aids, reachable from any page, and quiet for people who don’t need it.

What I gave up

The simplicity of a tour. I kept a short tour, as one thing Aida offers.

My hypothesis: guidance on the real screen reduces the time new users spend learning IDA.

The solution

Log in, get oriented, ask, see it

From my journey map: Aida starts onboarding at the first login, then answers with visual aids on the real screen.

User
Logs in
First time in IDA
Aida
Begins onboarding
Says hi and offers a short tour
User
Asks
“How do I save this search?”
Aida
Shows it
Highlights the real button

The answer panel

Answer, point, and say where it came from

IDA search results for battery recycling with Save search glowing in the toolbar, and the Aida panel answering how to save the search, with the top result card, suggested prompts, an input with a mic and an AI disclosure line — Fig. 2Aida answers and highlights Save search.

The Aida panel: a note that Save search is highlighted on the page, the question how do I save this search to a dataset, the answer, the top result card, suggested prompts, the input with a mic, and the AI disclosure — Fig. 2Aida answers and highlights Save search.

Highlights the real UI
Ask how to do something and Aida lights up the actual control, here Save search.
Suggested prompts
Chips such as What can I say? and Is this a real person?, both from my 2020 screens.
Answers with the source
The top result comes back as a card you can open.
Honest about being AI
A line under every conversation: answers can be wrong, check the source.
Ask in plain English
Type, or use the mic. No search syntax needed.

Beyond the tour

Summing up, and saying no

Aida panel: a search for battery recycling, the top result card, and a summary across 184 results: most mention Contoso and Rotterdam, second life comes up more after mid-2019, five documents compare two plants — Fig. 3Results: the top hit, and what the set has in common.

Aida panel: a request Aida can’t do, with an honest answer and an alternative — Fig. 4When Aida can’t help, it says so and offers another way.

The 2020 screens

What I designed then

Onboarding screen: Hi I’m Aida, I can help you get started with IDA, with the Aida bot and Back and Next buttons — Fig. 5Aida introduces itself.

Aida Chat Bot panel: search for COVID-19, here is the top result for COVID-19 with a document excerpt and Click to Open, click here to see all the results, and the prompts More results and Add keywords to search — Fig. 6The top result, in the chat.

Identity: Ai plus IDA equals Aida, with Aida Happy and Aida Sad faces — Fig. 7Ai + IDA = Aida.

Tour step on the IDA dashboard: the search box is lit while the rest dims, and the tutorial bar says you can start a new search from here, with Exit tutorial, Back and Next — Fig. 8A tour step.

Outcome

One assistant that teaches IDA on every screen

Designed

Research, a persona, the competitive review, a MoSCoW, a journey map, sketches, annotated mid-fi, the Aida identity and the final UI.

Direction

Onboarding by goal, plain-English answers, and highlights on the real interface, in one panel on every page.

Next round would test

Whether guidance on the real screen cuts the time new users spend learning IDA.

Reflection

What I took from it

Teach in context
The best help is on the screen where you’re stuck, pointing at the real button.
Say it’s AI
My 2020 screens offered “Is this a real person?” as a starter prompt. An honest answer, and a source, make an assistant easier to trust.
Design the unsure states
What Aida does when it can’t help matters as much as the happy path.

Appendix