top of page

AI Assistant

In this case study: user interviews, expert interviews, use cases, user journey map, design principles for AI and adapting classical frameworks to fit new needs of designing for AI

My role

Senior Product Designer


The client

January 2024 – present

World’s largest multinational communications & advertising company

The Product

The platform consolidates data, services and products, in order to break down silos and foster seamless collaboration between teams, becoming a daily destination for the company’s 130k employees and their clients. 

In this case study, we’ll focus on the AI Assistant, a launchpad for all things AI on the platform: chat, tools, and services.

The Problem

Following the public release of ChatGPT, agencies developed their own AI solutions to address their specific needs and those of their clients. However, with multiple AI solutions in place, there was a lack of a consistent user experience, which made it difficult to leverage AI capabilities to their full potential.

The Goal

A central AI Assistant solution will enhance workflows across agencies by providing access to the AI chat, tools and services from any point on the platform at any time.

My Responsibilities

  • Conducted user and expert interviews

  • Validated use cases

  • Crafted and communicated user journey maps

  • Collaborated closely with product teams that developed similar AI solutions in parallel 

  • Ran usability testing sessions


The central AI Assistant should be the launchpad for AI tools, services and chat. 

  • It has to assist with daily tasks by acknowledging the context of what task or project the user is working on at any given moment.  

  • It should help the novice platform user navigate the platform and it should speed up tasks for pro-users.

  • Thanks to the client, brand, market intelligence it’s powered by, it should provide the user with the most relevant results for their work.

  • It has to understand user intent, act as an orchestrator and call the necessary data, services or LLMs to provide the user with the synthesized answer or direct the user to a recommended tool. 

  • With so many useful features, the AI Assistant has to efficiently educate the user about its capabilities.

My Approach

  1. Discovery: expert interviews, user interviews 

  2. Definition: use cases, user journey map

  3. Develop: design principles definition, UX Design

  4. Deliver: usability testing, insight synthesis and analysis

AI Assistant - Process.png

Step 1: Discovery

To begin, I ran interviews with senior strategists, planners & technologists in agencies to understand how they were planning on adopting AI in their processes to help them run campaigns. 


At the same time, I interviewed users (creatives, media planners, strategists) to understand their day-to-day workflows and to get a feel for their goals, needs and pains.


These were the main pain points I uncovered in interviews:

  • User need help understanding how to navigate the platform and how to use AI Assistant effectively

  • When starting working on new projects, users get recommendations on relevant case studies

  • Strategists need help understand metrics (e.g. KPIs) and campaign results in natural language

  • Data scientists and strategists need to get quick answers to their questions about audiences and to delve deeper into the sources (internal data, 3rd party data resources) 

  • Strategists and creatives need to quickly generate images to put their ideas across in workshops and ideation sessions.

Step 2: User journey

I presented the interview findings in the form of a journey, where I mapped out all the steps of an advertising campaign as run on the platform.


Aside from the standard swimlanes on the user journey map (steps, actions, emotions, needs, pains, opportunities), as a slight improvement of this framework to accommodate the process of designing for AI, I tweaked the Touchpoints swimlane to include more information on data. This is crucial as the solution depends directly on the type, quality, availability and accessibility of the data. 

Aside from that, I usually add the following swimlanes to my user journey maps that give more context: 

  • Quotes. Here I paste direct quotes from user sessions in correspondence to each step. This allows anyone who’s reading the user journey map to step into the user’s shoes, promoting empathy. 

  • How Might We’s. In this swimlane, I try to distill the actual problem, by piecing together the information from other swimlanes. 

  • Ideation. This is where I put solution ideas from ideation sessions. 


It’s crucial to socialize the user journey map with the team and use it as a tool for building a common understanding of the needs we are solving.


Step 3: UX Design

The design challenge was to incorporate a previously disparate toolbar and chat into one single AI launchpad.


Before designing, my team and I defined some design principles:

  1. Discoverability 

  2. Utility and relevance

  3. Learnability

  4. Transparency


Here are the solutions we ideated around these principles.


Educate the user about the assistant’s capabilities (chat, tools, services, prompts).

  • Expose the AI tools in a way that allows browsing and discovery, both for first-time users testing them and also for pro-users

  • The toolbar should expand to show the tool names and collapse for a condensed view.

  • Prompts should help the user explore the chat’s capabilities. Regular prompts (e.g. improve text) return general LLM powered synthesized answers. However, services or mentions (e.g. @campaign, @data) are custom-made LLMs powered by company data that return relevant customized answers. 


Utility and relevance

Customize the view based on the user's needs, current task or project.

  • Prompts and services are suggested based on their popularity on the current tenant or relevance to the user’s current work context.

  • Depending on context change, some prompts will change. For instance, while working on a client’s project, users might switch to another client’s project and execute a similar task – we should let the user choose if they want to stay in the current context or change. 

  • Allow the user to utilize the generated content by copying, downloading into their current workflow.

Utility and relevance.png


Equip the user with knowledge on how to use the AI Assistant.

  • Prompts should help guide the user by pre-filling in the input field with the pre-prompt text. The user is free to change the prompt as they wish.

  • Services should be shown in the form of @mentions and described with a short description in the menu and tooltip.



Create a sense of trust by making the user understand in general terms how the AI Assistant works

  • The user wants to know that the system has conducted a complex calculation. If the calculation happens too quickly, users might not believe that the calculation has been done correctly.

  • Show the sources used while generating answers and upon completion. This way the user learns to trust specific sources and can delve deeper into each answer.


Final solution

The experience starts when the user interacts with the blue on the right edge of the screen. The tag sits above all other content of the page and is always shown, which allows easy access at all times.

Once the AI Assistant opens up, the user is presented with generic prompts, custom services in the form of @mentions and a toolbar with recommended and pinned tools. These tools open up in a window next to the chat to allow simultaneous use, for instance, a quick research about what audiences in a particular market prefer in the AI chat and then a quick image or idea generation in the AI tool.​​

Next steps

While constantly collecting quantitative metrics and user feedback, we are also planning to run usability testing sessions to help refine the AI Assistant.


Successfully unified disparate AI tools into a single, coherent AI Assistant, improving accessibility and user experience across the platform.



Traditional user-centered design frameworks are being slightly reshaped by the process of building AI products. New additions to classical approaches illustrate this shift:

  • Research, user interviews, user journey maps will delve deeper into data quality, availability and accessibility

  • Ideation sessions will be framed through the lens of key AI capabilities key capabilities (detection, prediction, generation) and functions (clustering, regression, classification) 

  • Business model canvases will include new commercial models

That being said, it’s more important than ever to define the real user need before rushing to solutions.

bottom of page