Overview
In 2023, our team came up with a solution, designed with a lean UX approach, to support profit growth across all grades within a consultancy. This product got white labelled to be an go-to-market offer to prove our competency in the radical GenAI era to the market.
CLIENT
Internal (Consultancy)
Project sponsored by EVP
DURATION
6 months and counting
ROLE
Lead product designer
DELIVERABLES
User flow, Wireframe, UI delivery, Design system
Background
The Ask
It is all about the rise of conversational GenAI technology in the industry. The leadership aims to assess the organisationās readiness to adapt the emerging technology trend by sponsoring an internal project
The Objectives
To design a conversational genAI PoC that removes the barriers to access corporate knowledge, encourage consultants to participate in business development and accelerate the bid process
The users
All consultants in the corporate.
Common characteristics are: tech-savvy, fluent in searching information, and confident in using self-serving products
The scope and focuses
PoC phase
Feasibility Viability
The PoC was completely experimental. The engineers put extraordinary effort in data training, not to mention to thank all sectors and units to onboarding their business data.
Pilot
Feasibility Desirability Usability Performance data-driven
After the beta launch, our team actively looked for primary feedback to refine the data and improve the conversational experience for 2.0
TLāDR
ARIAās beta phase launched for 1 year and counting. We engaged over 700 users (64% active) and received 3900+ prompts in the first 3 months! As of February 2025, the 2.0 is on preproduction environment and ready to test.
Kudos to our proactive team for pushing our momentum for 6 months
- Product owner
- Lead BA
- Lead engineer
- Front end engineers
- Backend engineers
- Solution architect
- User researchers
- Content designer
- Lead product designer << me šš»
- Interaction designers
- Performance analysts
- Marketing reps
- A new born! š£
01.Pilot
The Start.
āI wish I was better at UX and design but I lack the creative spark to get nice colour palettes and usable designsā - George H., the lead engineer (June 2023)
It was started from an ask from the lead engineer to prepare ARIA (formerly called CapGPT) concept presentation. This is the abridged timeline in 2023.
Early 2023
Successful feasibility and risk discovery to continue a GenAI solution, incorporating a low-code PowerApps for data collection.
Mid 2023
More than 1200 data points were collected and under review as well as testing other technologies such as prompt engineering.
Started bringing front-end effort to the conversation agent and the data collection PowerApp for demoing to the executive leadership team
End of 2023
The roadmap had set the goal at year end to scale ARIA up as a usable product.
Discovery
How can I help you, George?
To fill in Georgeās skill gaps, I needed to understand which group of people within the corporate will use ARIA and the leadershipās expectations.
01 Who is your user(s)?
- All consultants in the corporate
- Including director and above level presumably
02 What is the objectives?
- Save time and resources in bid process
- Potential in improving bid winning rate
03 Where is it going to be?
- MVP will be on web (ideally)
- PowerApp for data collection
Hypothesis (concieved by the business)
We believe that integrating a GenAI solution into our internal platform will benefits consultants (users) by supporting their participation in business development activities and improving bid successes.
Problem statement(s)
Unknown usersā perspective toward GenAI
GenAI is a buzz word and a controversial topic. Users could be skeptical towards the product and its data usage.
Unknown technology adoption
Where do our consultants (users) fall on the Technology Adoption Lifecycle as genAI made radical entry to the market in 2023, even if the users are IT consultants?
GenAI was new to the corporate internally
No other genAI products could reference within the corporate. The design direction could rely on other products in the market.
Short timeline and time cost
Given the tight timeline, traditional design process with in-depth interviews, observations would be too long for an internal product development.
Assumption to move forward
We believe lean UX would be the best approach to design ARIA as it enables rapid development and foster continuous improvement (after launch). To address the uncertain users perspective to GenAI and their actual needs, the objective of ARIA pilot is to provide an experience that is as familiar and intuitive as the daily work apps that users already know.
Design process
User flow
At least I was told this will only be a conversational AI platform, a user flow can estimate the interface effort from the given concept reference.
Design direction
Thank goodness we have access to the corporate branding playbook.
Given that ARIA MVP will be internal facing, incorporating the global branding is inevitable.
Despite the materials are unavailable on Figma for the later UI development, we can reference the official branding playbook to establish a familiar look and feel for our internal product, ensuring a seamless experience across internal tools while enhancing credibility for our data-heavy product.
I also took branding guideline as a starting point to upskill my peer designers who are relatively lack of experience in developing UI component from scratch.
I conducted rapid interviews on understanding active bid participantsā perspectives on daily work apps.
In the use case of searching and reading case study, there are common perspectives from our in-house consultants:
- Feel comfortable reading text-heavy materials on Powerpoint with a dark-themed background
- Feel familiar with the corporate branded slides on Powerpoint
- Feel intuitive from searching on Microsoft Teams (links across Sharepoints)
- Fluent in using keywords to look up information via search engines
Based on these findings, I reference usersā daily working apps to drive the design:
ā
Ā Start the search on a landing page Resonating search engines
ā
Ā Daily message app conversation Resonating the heavy use of Teams
ā
Ā Open data sources in new tab Resonating day-to-day ways of working
ā
Ā Dark-themed background Resonating Powerpoint
- Inspired by the 2023 corporate Powerpoint master template
- Dark theme consumes less energy as a small contribution to sustainability from me
Additional refinement for testing
To accommodate the data quality and performance from the engineering team, the design iterations focused on the responses and new conversation.
The drama before launching
Testing and iteration
The user researcher team had conducted a thorough user research to justify the business case as well as taking the wireframe as the foundation of the output.
The research team completed the synthesis, followed by an interactive workshop with the whole delivery team to refine the solution for the Pilot launch untilā¦
The internal group IT department blocked us from hosting ARIA as a standalone internal website. Then legal team involved in terms of use and privacy policies. Oh! Branding team has entered the chat too!
Letās firefighting altogether!
While the amazing product owners and BAs were liaising with the legal team and the branding team, the engineering team was striving for an alternative way to host ARIA for everyone in the corporate.
Workarounds
1. The built is on Microsoft Teams as an built-in app
2. Users must download the app and manually install it.
To prevent an unhappy flow from the installing process, the research team ran extra rounds of rapid usability testing including the installing process and the design workaround.
Experience to tweak
- Adapt the conversation layout as similar as Microsoft Teams
- 1:1 conversations like text messages
- scroll up to view all chat history
- Descope in-page search due to Teams app constrains; Teams native search is not applicable
- Adapt Teams interactions as much as possible
- keyboard shortcut (i.e. Enter key)
Interface to tweak
- Align the typography and font-size for natural āChat-to-Ariaā eye flow and task transition
- Accentuate the colour on the āresponseā speech bubble as to catch usersā attention from the text-heavy page
- Tone the purple-themed colour to avoid clashing
- Responsiveness (deprioritised)
Outcome and next steps
ā Ā The installation process was smooth (Thank to the content designer the precise instructions!)
ā Ā Users were not too offended by using the pilot as a built-in app on Teams
āĀ Users found the text is too small (we set the font size identical to Teams 100%-zoom)
āĀ Despite the dark-theme is more favourable, users would like to see a light theme version
āĀ Not responsive. A few active workphone users prefer accessing ARIA on workphone as well
āĀ In-app search is not applicable. Active bid-writing users do not have a handy way to look for history or search keywords
These insights have been grouped as 2.0 opportunities for feasibility check and the roadmap planning.
Launch ššššš
The beta launched in February 2023 and we shamelessly ask users to provide feedback through the in-built feedback system.
Noticeable results in 3 months
Credit to the performance analyst who gathered all the results weekly.
700+
internal āsubscribedā users
64%
active users
3,900+
total prompts
8.35
prompts per active user
Positive impacts
Objective ā
Lightning process
āActivities which typically took days such as finding case studies or gaining insights for a certain client project now take a matter of seconds.ā
Objective ā
Win win win!
āCapGPT has been an extremely helpful resource during the completion of a recent {client} bid - which we were successful in winning!ā
Surprise š
Break the barrier
āCapGPT was intrumental in allowing me to engage and win my first bid as a mid-weight consultant in my first 6 monthsā
2.0 Phase
In development
- While the beta is live and we are actively gathering feedback, the product team and the research team kick-off research activities to steer the opportunity prioritisation
- The success brings noise within our corporate and multiple sectors requests be a part of the database
- as of January 2025 (1 year after launching), ARIA has 5,500+ data points sourced from all units in the corporate
- Considering Microsoft Teams will integrate Co-Pilot, the goal is to preemptively prevent potential confidential data breach from AI training. The main objective of 2.0 is to transition the product away from Microsoft Teams to a SSO internal website while introducing new features to retain recurring users
- As the foundation was a success, the designers uses agile methodology to refine new interface features:
- SSO onboarding
- Sidebar
- Conversation history
- Grouping
- Light and dark themes (new design!)
- UI refinement
Takeaway and self reflection
Design system does matter
Without a design library, the design iterations and having 2+ designers caused some visual inconsistency and dated branding time to time. In addition to basic human-computer standard, we rely on designs and developersā experiences to contribute to the process.
No shortcut in launching a product
Look at a holistic corporate view in launching new product! Took the lesson from IT, legal and branding departments, launching an internal product need to seek external guidance no matter it is a CEO or VPās call.
