Senior Design LeadJun 2021 – Dec 2022Remote
Morgan Stanley
- The problem
- Clients waited on analysts for short research answers, and analysts waited on email to clear every public appearance.
- My role
- Senior Design Lead, 2021 to 2022. I owned UX strategy for the Research Portal and AskResearch, the assistant that cites its source and hands off to the analyst, and brought analysts' appearance approvals into it.
Headline results
- 150K+
- <1 min
- 93%
Context
Research became a product the year it got a price tag.
Before unbundling, the top fifteen banks sent over 40,000 research notes a week, and fewer than 5 percent were opened. In 2018 MiFID II split research from trading commissions, budgets fell 20 to 30 percent, and clients paid only for insight they could find, trust and act on.
Clients reached analysts mostly by email, then a call, and every interaction was billed. I joined Morgan Stanley Research in June 2021 to own UX strategy for the Research Portal and AskResearch.
- 20–30%cut to research budgets after MiFID II
- 40K+research notes a week from the top 15 banks
- <5%of those notes ever opened
- 70K+Morgan Stanley reports a year
The user funnel, as a diagram: Email blast, Analyst email and Organic each lead to Articles Reports, which leads to Email, Chat and Call. On the channel side, two labels: Analyst & Team, and Sales.
Research
Everything existed except the conversation.
We interviewed the analysts and brought clients into the workshops. Most client contact ran over email, and most of it was small: “60–70% of client inquiries are short and require 2–3 sentence responses from the team.”
Analysts named the gap themselves: “we got the articles, emails, calls, podcast, but no chat.”
The analyst interviews. The interview plan, under the heading Analyst interview: things to keep in mind, the goals of analyst chat and the research outcome. Then the notes, sorted into four topics: Analyst/Client interactions (Email/Call), eight; Analyst/Client interactions (Chat), nine; Content/ Publications, four; Qualifying Clients, five. Each list reads as written below.
Analyst interview
Things to keep in mind
- Objective of research
- Unusual grouping
- Triggers to action
- Current customer journey map (To compare with the future one based on ideation)
- Does the journey map has any gaps, questions? Do we need answers for those?
Goals of Analyst Chat
- Platform and availability
- Ease of reach and convenience
- Efficiency (Content integration, booking, etc..)
Research outcome
- Summary
- Objective
- List of questions
- Note of the interviewee and their role
Three personas in two groups, each with motivations, needs, main tasks and pain points. Clients: the client. Research: the lead analyst and the junior analyst.
The questions
Each cluster became two questions.
We clustered what we heard into research integration, engagement and access, and chat interaction, then turned each into two problem statements.
The research board, three clusters of stickies: Research integration, two; Engagement and access, fourteen; Chat interaction, four.
Research integration
Being able to connect with analyst while reading particular article
Integrating research with the chat with my clients
Engagement and access
Reach clients on mobile when I am on the go
I want my group chats to be persistent across mobile & web applications so I can pick up from where I left off
As a client, I want to access different sector analysts in one conversation - “role engagement”
Bloomberg is part of the clients daily behaviour
Bloomberg has “offline” chat mode aka send email
I like Bloomberg because my clients are using it
Engagement and availability for Clients
I don't have a Bloomberg and want to access multiple clients / analysts in real time conversations
I don't want to be bombarded w/ qs as the Lead Analyst; allow my team to help mitigate qs
Using chat is to compliment other ways to communicate with client
Having more than one access point to interact with the client
Ability to invite other analyst(s) to a chat with a client
Ability to interact with multiple clients at once (Group Chat)
Bloomberg can only invite 2 users to chat
Chat interaction
No real value in archiving chat history
Bloomberg will open a new chat for group chat
Bloomberg is based on command line UI
Able to copy/ paste (Templated scripts) quickly to client
Three research themes, each with its two problem statements. Theme 1, research integration. Theme 2, engagement and access. Theme 3, chat interaction. The statements read as written below.
Theme 1
Research integration
Bringing the research, and the analyst behind it, into the conversation, so a short question is answered where it is asked.
Theme 1 Problem statement(S)
How might we answer clients' short research questions inside the chat, with the source report, without waiting for an analyst?
How might we connect a client reading a report with the analyst and the team who wrote it?
Theme 2
Engagement and access
Letting clients reach the right analyst, and the whole sector team, from wherever they already work.
Theme 2 Problem statement(S)
How might we let clients reach analysts on the portal, on mobile and in Symphony, in one conversation that follows them?
How might we let a sector team share the client questions, so the lead analyst is not the only way in?
Theme 3
Chat interaction
Chat as quick as the tools clients know, without archiving every word.
Theme 3 Problem statement(S)
How might we let analysts answer in a few taps, with templated replies they can send as they are?
How might we keep a group conversation in one place, where Bloomberg opens a new chat for every group?
Ideas
We prototyped it as a text conversation.
We worked the questions through with the analysts, the tech and marketing teams, and clients we invited. I prototyped the assistant as a text thread on a phone: a price, who covers Apple, then a hand-off to an analyst.
The AskResearch prototype: four phone screens of a text conversation under a Morgan Stanley AskResearch header. The greeting, Good morning Harvey!, offers three replies: Top picks, Podcast collection and Chat with an analyst. Asked for the Apple stock price, AskResearch answers $150, and it names Jessica Pearson as the analyst covering Apple, Inc. (AAPL). Asked who to chat with, the client answers Chat with Louis Litt, and his card follows: Louis Litt, Senior Analyst, with a Chat button. In the chat, the client asks Louis to email the latest research on IBM, and he answers Sure thing! Check your inbox.
Wireframes
We drew the assistant where the conversations were.
Clients and analysts already chatted in Symphony and Skype, so the first wireframes put AskResearch in both. Symphony, the main channel, went forward, beside the portal and the app.
Wireframes of AskResearch, desktop then mobile. Desktop, three wireframes. AskResearch on Symphony: a chat window with the left rail of chats, their status dots and a new chat button; a search strip with six round icons; the chat header, Ask Research, Company name; an empty state with two chat bubbles between two people and the line Start chatting with Ask Research!; under Yesterday, Username at 1:50 asking Apple; under Today, Ask Research at 1:50 answering with a card, Heading, two lines of text and a Button; and the composer, Type your message here. AskResearch on Skype, with no text: a title bar with window controls, a green presence dot, a small avatar dot and a large avatar circle over a hairline, a date rule, an outgoing bubble, the assistant avatar beside an incoming card bubble with a header, a body and a footer, the empty composer between two rules, and a toolbar of five round controls. The portal home: top navigation, a banner strip, a centred search pill, the Morning Summary with seven tabs and five article rows, and the docked AskResearch chat rail with three contacts, message bubbles, a video tile and a composer. Mobile, five phone wireframes in a row: disclosures, a headline bar, two body bars, a dashed disclosure box and an accept pill; My Feed, filter tabs, time bands and cards with a floating AskResearch pill; mention, a chat thread whose composer holds an at-sign stub with a floating @AskResearch suggestion card; answer card, the same thread plus a card with a title bar, two link stubs, four key and value rows and one green delta stub; and group chat, an outgoing question and an incoming answer under a sender label.
Desktop
Username1:50
Apple
Ask Research1:50
Heading
Mobile
What made the cut
Three to four epics make the MVP.
The squad voted the ideas into epics, and four made the MVP. Analysts' approvals waited for the next release. I drew the road to MVP with the product owner.
The votes on the ideas, rolled into epics. Above the line, 3-4 Epics = MVP: Ask in the chat, 22 votes; Reach the analyst and team, 21 votes; Sector team chats, 18 votes; Know it's there, 8 votes. Below it, later: six more ideas, the last of them, analysts' public appearance approvals, for the next release.
Ask in the chat
22 votes
@AskResearch in any chat
9 votes
Answer cards: price, rating, price target
7 votes
The source report with every answer
6 votes
Reach the analyst and team
21 votes
Chat from the portal and the app
8 votes
Hand off to the analyst when the assistant can't answer
7 votes
“Ask Jessica & Team” on every report
6 votes
Sector team chats
18 votes
Group chat with the lead analyst and team
8 votes
The same chats on web and mobile
6 votes
Invite another analyst into a chat
4 votes
Know it's there
8 votes
“Did you know” tips for @AskResearch
5 votes
Welcome screen with the disclosures
3 votes
3-4 Epics = MVP
Later
Notification of new reports
4 votes
Analyst feed: updates, top picks, events
4 votes
Chat history as a client timeline
4 votes
Blast a report to many clients
3 votes
Book a call from the chat
3 votes
Analysts' public appearance approvals
Next release
The road to MVP. User Research: user pain points, user needs, user goals. Research Themes: user research summary and synthesis. Problem definition: list of HMW problems to solve based on user pain points, needs and goals. Ideation: team collaboration to provide ideas to solve user problems. Visualization: design team provide a visualization for the ideas in a wireframe flow to help write feature list. Feature Prioritization: squad workshop to prioritize the list of features (Epics) and decide on MVP. Sprint Zero definition: identifying squad acclivities to prepare for sprint zero including writing user stories for the first Epic and stories in sprint zero; 3-4 Epics = MVP. Then the sprints: Sprint Zero, Epic 1, 3 user stories; Sprint 1, Epic 1, 5 user stories; Sprint 2, Epic 1, 5 user stories; Sprint 3, Epic 1, 5 user stories; Sprint 4, Epic 2, 5 user stories; etc.
User flows
Ask with @AskResearch, in the same thread.
In the analyst's Symphony chat with a client, @AskResearch summons the assistant, and it answers in the same thread: Apple research for the analyst to share, then the Apple price for the client.
The AskResearch flow in four frames, inside the analyst's Symphony chat with client Harvey Specter. Frame 1: Harvey asks Louis what is the latest on apple, and Louis replies that he will send the latest research on apple. Frame 2: the analyst's composer holds @AskResearch Show Apple Research. Frame 3: the command @AskResearch Show apple research is sent, and AskResearch answers with an Apple Research card: three reports, each with a checkbox, its title, By Jessica Pearson & 3 others, 22 Jan 2020 12:01 AM GMT, and Rating 100% or Rating 75%. Frame 4: under the card, Harvey asks @AskResearch Apple price, and AskResearch answers Apple price $150.
What shipped
One assistant, three surfaces.
The assistant shipped where conversations already lived: docked on the portal, in the mobile app, and inside Symphony. An answer card names its analyst, and a client can ask the whole sector team in one group chat.
The Morgan Stanley Research Portal home: top navigation with the Morgan Stanley Research wordmark, My Feed, My Collections, Themes, Asset Classes and Tools; a gold banner reading Morgan Stanley's Core Views: CIO Brief; the search bar; the Morning Summary feed with geography tabs from Macro to EEMEA and five tagged research rows with authors, dates and page counts; and the expanded AskResearch chat rail: the chats list with the digital assistant and two senior analysts, the assistant's greeting, Good morning Harvey!, a video at 0:00 of 1:39 and the composer, Ask your questions here. A dark feedback button sits at the bottom left.
Five AskResearch mobile screens: the welcome screen with its disclosures and an Accept button; My Feed with filter tabs, Last Week and Two Weeks Ago, three research cards and the AskResearch button; a chat thread with Louis where typing an at-sign brings up @AskResearch; the answer card for Apple, Inc. (AAPL.O) with Analyst Jessica Pearson, Stock Rating Overweight and Price Target 328.00 USD; and the Food & Beverage group chat, where the client asks the team and Donna answers.
A Morgan Stanley research report page: the dark bar with Editions: Global, Chats, My Feed, Shortlist, Collections and the H Specter user menu; the research navigation; a row of six promo cards, Chat with Analysts via AskResearch among them; a pick-up-where-you-left-off banner; the Global Insight article Global Technology Supply Chain and the Data Worth Watching by Jessica Pearson, Managing Director, Equity Analyst, with its share actions; a contents rail listing the report's sections; the alphawise player, Listen to the Introduction, at 01:23 of 5:26; the report body; the Response to significant event and In this report cards; and the docked chat with Jessica on the far right.
Public appearances
Analysts had a second inbox: approvals.
Before an analyst speaks in public, at a media appearance, an event or a conference, the appearance needs approval through the Public Appearance Tool, PAT. Analysts and their assistants called it dated and manual, with no view of where a request stood.
The Analyst Feedback board, in two parts. Must Dos, in four categories: automate and speed up approval process; modernize UI, simplify submission fields and process; access to PAT submissions and attachments; proactive alerts and reminders. Then New Enhancements: mobile friendly, bulk request, a new search field and Methode integration. Each note gives its category, then the analyst's words, as written below.
Analyst Feedback
Must Dos
New Enhancements
PAT Personas, with motivations, needs, main tasks and pain points for each column. Requestors: the lead analyst and the team assistant. Approvers, one column for five groups: Res Risk, Res Mgmt, SAs, Control Group*, Corp Comms. Under the approvers, the mandatory approval channels: Res Mgmt for sensitive topics, Supervisory Analysts for attachments and materials, the Control Group for all PATs 24 hours before the event.
PAT Personas
The votes
Transparency drew the most votes.
The research came down to three themes and six problem statements. In the workshop with analysts, assistants and every approver group, transparency won with 28 votes: a progress tracker like parcel tracking, with an approval status indicator.
Three research themes, each with its two problem statements. Theme 1, automate and speed up approval process. Theme 2, modernize UI: simplify submission fields and process. Theme 3, access to PAT submissions and communication exchanged with attachments. The statements read as written below.
Theme 1
Automate & speed up approval process
Enhancing the PAT approval process through automation, accountability, and improved communication to reduce delays, redundancies, and errors.
Theme 1 Problem statement(S)
How might we optimize all stakeholder group approval workflows for various types of PATs? (e.g. sensitive topics, materials, media)
How might we enable approvers to communicate and provide PAT updates to submitters, respective approvers, other FYI approvers?
Theme 2
Modernize UI: simplify submission fields/process
Improving the usability and design of the PAT tool to enhance efficiency and reduce stress for analysts.
Theme 2 Problem statement(S)
How might we provide a simple user UI and interaction experience that has minimal cognitive load, clear simple language, clear user flow, provide help, guardrails, education and prepopulated data?
How might we provide transparency around PAT requests with proactive status updates, reminders, regional consideration (time-zone & business units) and respective approver assignments?
Theme 3
Access to PAT submissions & communication exchanged with attachments
Improving the functionality and collaboration within the PAT tool, with focus on search (historical PATs) and attachment approvals (SA offline workflow)
Theme 3 Problem statement(S)
How might we provide the ability to edit and tweak submitted PAT requests while setting strict time parameters and expectations for PAT approvals?
How might we provide search functionality to surface past PATs and provide direct collaboration functionality within the PAT application to easily exchange and archive commentary with SAs around submitted attachments?
The votes, cluster by cluster: Physical Transparency in the approval process and Intuitive UI experience, 28 votes; Editing capabilities, 19 votes; Extended approver coverage and support, 12 votes; Physical notifications based on PAT actions/updates, 10 votes; Search and historical access, 5 votes. Under the top two, what each asked for; under the other three, the analysts' notes that asked for a conversation, for notifications and for search. Each reads as written below.
Physical Transparency in the approval process and Intuitive UI experience
28 votes
- Introduce an interactive 'Approver Directory' within the system for easy reference and communication.
- Design a 'Progress Tracker' (similar to parcel tracking) for real-time tracking of PAT journey and provide Approval Status Indicator.
- Showcase 'Active PAT Assignment' on the landing page to prevent duplication of work (Approver point of view).
- Refine the email communication module to deliver precise and easy-to-understand approval updates.
- Overhaul the system's UI, focusing on typography, color scheme, and usability of elements like calendars.
- Highlight 'Media' versus 'Non-media' PATs and improve the process of flagging non-covered companies.
Editing capabilities
19 votes
- Introduce 'Dynamic Submission Editing' allowing submitters to retrieve, correct, and resubmit PATs when needed (where material changes will be clearly highlighted for approvers’ convenience), or make amendments at any stage at least 24hrs prior to event.
- Implement 'Real-Time Submission Editing' that locks once a reviewer starts reviewing it to maintain the integrity of the review process.
- Introduce PAT in “draft mode” if a user is not confident with submission (for both submitters & approvers)
- Introduce a 'Commentary Inclusion Feature' to ensure all approver comments are included and centralized in submissions.
- Enable 'PAT Withdrawal' for cancelled events to reduce unnecessary follow-ups.
- Allow 'Selective Editability' to avoid unnecessary approval cycles; only certain fields may be open for edits.
- Reuse previous PAT submissions and respective data to speed up creation process
Extended approver coverage and support
12 votes
Chat for help
One star.
Small digital assistance (chatbot) with in PAt UI for common faq or operations.
Chat tool w/in program to ask questions
PAT request must contains some conversation message exchange section UI within submitted PAT ticket, so that submitter/approvers should be able to exchange messages with in that section itself for any clarifications for that ticket.
Physical notifications based on PAT actions/updates
10 votes
Notify respective audience whenever PAT status changes through desktop alert, Microsoft Teams alert, email alert so on so forth based on approvals timezone
One star.
Email reminders for missing documents/info
Search and historical access
5 votes
Ability to view list of past PAT requests
It would be good if we could search for PATs submitted by others so we can check the status of them and avoid having to follow-up with the person who submitted it
Into AskResearch
The approvals belonged in the conversation.
The ideas kept asking for a conversation: help inside PAT, a thread on every request, reminders before a deadline. Analysts already had that conversation in AskResearch.
So I proposed that PAT stay the approvers' tool and the record, and that AskResearch become the analyst's way in. I took it to both product owners, the Control Group and the supervisory analysts, and we agreed one rule: the assistant prepares and relays, people approve.
What moved into AskResearch and what stayed in PAT. Under the rule, a table of the clusters the workshop voted for, from 28 votes to 5: each with what it became in AskResearch, for analysts, and what stayed in PAT, for approvers.
AskResearch prepares and relays. PAT keeps the record. People approve.
Physical Transparency in the approval process and Intuitive UI experience
28 votes
- A tracker in the chat: each approver, each status
- Message an approver from the request
- Approver Directory
- Active PAT Assignment, so no one reviews twice
- New type, colour and calendar
Editing capabilities
19 votes
- A draft to check, started from the last request
- Edit and resubmit from the thread
- Withdraw a cancelled event
- Only some fields open for edits
- Edits lock once review starts
- Every comment kept with the request
Extended approver coverage and support
12 votes
- Ask for help in the chat
- A conversation on every request
- More approvers per group
- Cover across time zones
Physical notifications based on PAT actions/updates
10 votes
- A reminder before each deadline, so a Monday request is not left for Friday
- Status changes in the thread
- Teams and email alerts by time zone
Search and historical access
5 votes
- Find my past requests
- A searchable archive
- Management reports
An appearance request's path. It starts one of three ways: AskResearch reminds the analyst, the analyst asks @AskResearch, or the team assistant starts it. A draft is filled from the last request and the analyst's coverage, and the analyst checks it and submits in the chat. Materials attached? If yes, the Supervisory Analyst reviews the materials. Sensitive topic? If yes, Research Management reviews the topic. The Control Group reviews every request, 24 hours before the event. Changes requested? If yes, the analyst edits in the thread and resubmits, the changes marked for the approver, and the request goes back to the reviewer who asked. If no, it is cleared: the thread says so, and the team sees it. From the draft or any step, a cancelled event ends with the request withdrawn.
Materials attached?
Sensitive topic?
Changes requested?
Appearances in chat
Prepare, submit, track and answer, in one thread.
A Monday panel shows the loop: a reminder so the request is not left for Friday, a draft started from the last one, each approver tracked, and their comments in the thread.
Five screens of Jessica Pearson's AskResearch thread on her phone, one Thursday. At 8:00am AskResearch reminds her of Monday's panel at the Global Technology Conference: the Control Group needs every request 24 hours before, so it should not be left for Friday. At 8:05am the request is drafted from her 12 Sep panel: non-media, her coverage of Apple, Seagate and Sonos, the panel slides, and the Supervisory Analyst and the Control Group as approvers. After she submits at 9:42am, a tracker shows Harold Gunderson, the Supervisory Analyst, reviewing with edits locked, and Faye Richardson of the Control Group waiting. At 11:20am Harold asks in the thread for Microsoft on slide 7 to be marked non-covered; she edits and resubmits, and he sees only that change. At 3:06pm the request is cleared for Monday, Gretchen Bodinski can see it, and her Reg AC survey follows the panel. Numbered marks on the screens point to the eight notes below.
- 1
A reminder before Friday. Analysts asked for alerts so a request for a Monday is not left for Friday, time zones counted.
- 2
Started from the last request. The editing votes asked to reuse previous requests and their data.
- 3
Media or non-media. The top vote asked to set media apart from non-media appearances.
- 4
Coverage filled in. Analysts could not pull their coverage into a request, and added each stock by hand.
- 5
The tracker. The top vote, 28: a tracker like parcel tracking, with each approver named. The editing votes added that edits lock once review starts.
- 6
Comments in the thread. Analysts and approvers asked for one place to exchange comments on a request.
- 7
What changed. Approvers asked for every change to be marked, so they review only that.
- 8
The team sees it. Assistants could not see whether a request was cleared.
Design system
A design system adopted across teams.
I built a scalable design system, adopted across teams: Figma component libraries with tokens, variants and documentation, built and governed with design QA. It cut the time from design to build by 35 percent.
One AskResearch message doing five jobs, each drawn as its screen draws it. First the answer card from the app, for Apple, Inc. (AAPL.O), with its five parts numbered: the sender, AskResearch; the title; the detail, IT Hardware, North America and the analyst, Jessica Pearson; the rows, from Stock Rating Overweight to Diff. from Close; and the time, 8:01am. Then four more jobs: the Apple research list in Symphony, three reports by Jessica Pearson with their ratings; the welcome in the portal, Good morning Harvey, with the intro video; the request tracker for the approvals, each approver named; and changes requested, its title in red with one button, Edit the request.
AskResearch
Apple, Inc. (AAPL.O)
IT Hardware, North America
Analyst Jessica Pearson
8:01am
- 1
Sender. AskResearch names itself, so a client always knows who answered.
- 2
Title. One line says what the message is about.
- 3
Detail. Where the answer comes from: the coverage and its analyst.
- 4
Rows. A label and a value on each line, so a number reads at a glance.
- 5
Time. Bottom right, as on every message in the thread.
Outcome
Three products shipped.
The Research Portal, AskResearch and the Public Appearance Tool shipped. The portal and the assistant served more than 150,000 institutional clients. The assistant answered over 6,000 staff questions a month, finding a report took under a minute instead of ten, and 93 percent of its cited answers were rated helpful.
- 150K+institutional clients served
- 6K+staff questions a month
- <1 minto find a report, down from ten
- 93%of cited AI answers rated helpful
- 35%less time from design to build
- 3products shipped