Role: Design, research, strategy
Project duration: 6+ months (Jul 2023 - Present)
Primary stakeholders: Founders (CEO + CTO), software developers
Results: Interactive prototypes, working Chrome extension and web portal
This case study focuses on the AI Agent workflow automation platform I have designed during Sphere's pivot, from user research to ideation, wireframing, and prototyping. AI Agents are an application of generative AI that essentially operate as follows:
1. User provides AI Agent with an objective (in this case, using a chatbot interface)
2. AI Agent breaks objective into tasks, using generative AI
3. AI Agent autonomously executes the tasks across your apps and/or browser until the objective is completed
How did we get here? User Research!
Since March 2023, we have conducted over 100 interviews with potential users via Zoom in an effort to discover the most pertinent problems they were facing at their companies that we could solve. The founders knew they wanted to remain in the B2B SaaS space, so we primarily interviewed people with roles in product, sales, customer success, finance, and HR.
Early on, the calls were exploratory research. But over time as we gained direction and built prototypes and products, the calls were more focused on collecting feedback and usability testing.
For the first few months I was in charge of taking notes manually, and eventually we got a Gong subscription to help with meeting notes.
After each call, I was in charge of extracting the key insights, adding them to a Miro board, and organizing them into affinity maps as we discovered themes and trends in the problems that users were facing.
What We Learned
Macro insight: Businesses are laser-focused on minimizing costs and maximizing employee efficiency, especially given the economic landscape of 2022-2023.
This may seem obvious, but this insight was critical to keep at top of mind when building software in 2023. It was also the very reason our previous corporate education product stopped growing; it was seen by companies as a nice-to-have rather than an essential service that would directly translate to increased revenue.
Why AI Agents?
Given the macro insight above paired with the rapid developments in generative AI technology that we followed throughout 2023, we decided to build AI Agents that would automate repetitive and time-consuming tasks in order to help businesses maximize employee efficiency and increase revenue.
(Note that this was not the first idea we landed on; we tested multiple other ideas from March - June before pursuing AI Agents in July. To learn more, read my Learnings From An Early-Stage YC Startup in 2023.)
AI Agents for Sales Teams
Of all the verticals we conducted research on, we were most excited about the opportunities in the sales space, so that is what we focused on first.
Here are the pain points from our user research that informed the initial workflows we focused on automating for sales teams:
Pain point #1: Sales reps are not consistent at updating the CRM after sales calls according to their company's best practices. This results in the loss of valuable customer information that could be used to close deals and identify expansion opportunities.
Pain point #2: Account research in preparation for sales calls is the most time-consuming task for sales reps. This takes away time that reps could be using for customer calls instead.
I conducted thorough competitor research on other companies building AI Agents (Adept, AgentHub, Automat, Beyond Work, Lindy.ai, Multion, Orby) and companies specifically focused on automations for sales teams (Managr, Rattle).
My research process was as follows:
1. Conduct background research on competitors via Google, LinkedIn, Twitter, Crunchbase, and YC Directory
2. Create an account with each competitor
3. Spend time as a user of each competitor's product
4. Map out each competitor's product in Figma
5. Create a list of features that competitors did well, features that could be improved, and how we might implement these learnings into our product
Based on our user research, competitor research, and technical capabilities, we first decided to build a product for the following user profiles:
1. Admin user = Sales Leader (e.g. Director of Sales Operations)
2. End users = Sales Reps (e.g. Account Executives)
We went through many early iterations which involved a lot of wireframing and prototyping. In order to avoid this case study becoming too long, I will not go into detail on all these iterations.
If you are interested in learning more, feel free to contact me at email@example.com.
At this stage, we decided to use a Next.js template for the Sphere web app in order to conserve our design and engineering resources. We picked the Vuexy template because the design is clean and matches Sphere's branding, it includes many of the features we would need, and it had many positive reviews.
The prototypes below were constructed using components from the Vuexy Figma library. They include flows for our 2 user profiles, Sales Leaders and Sales Reps.
1. Create Sphere account
First, the Sales Leader would create a Sphere account for their organization.
We decided to use PropelAuth for this in order to conserve our design and engineering resources.
Sales Leader connects Sphere to Salesforce and Gong via API integrations in Sphere web app.
3. Create first Agent
Sales Leader uses natural language to describe the workflow they wish to automate. Sphere uses these instructions to train our AI model to complete the described tasks in line with the customer's best practices.
I came up with 4 different options to accomplish this:
4. Invite Sales team
Sales Leader invites Sales Reps to Sphere web portal + Chrome extension.
We also used PropelAuth for this in order to conserve design and engineering resources.
5. Install Chrome Extension
Sales Reps install Sphere Chrome extension (which is connected to the Sphere web portal).
6. Sales Reps use Chrome extension
The prototype below shows the user flow for CRM data entry automation.
7. Manage agents
Sales Leaders can use the Sphere web app to view and manage:
1. All their team's Agents
2. Charts showing Agent usage by Sales Rep, number of automations complete, and amount of time saved
3. Output log showing suggestions Sphere's Agents have generated, the source the suggestion came from, and the status of the suggestion
4. Feedback that Sales Reps have submitted for the Agent
We built and tested the iteration above through design partnerships with real sales teams, primarily for the CRM data entry use case. Over the course of approximately 2 months, we discovered the following issues:
1. Sales reps were not consistent with accepting the suggestions generated by Sphere's Chrome Extension. The longer the reps put off accepting Sphere's suggestions, the more suggestions built up.
2. Following demos of the CRM data entry automation use case, companies were slower to convert into customers than we had anticipated.
We hypothesized that these issues were occurring for the following reasons:
1. Accepting suggestions via a Chrome extension was not the ideal UX; it still required too much labor from users.
2. The use case of CRM data entry was not urgent enough; companies had more important pain points to address.
In order to test these hypotheses, we developed more iterations that would:
1. Require little to no labor from users
2. Address different pain points besides CRM data entry
These prototypes use a chatbot interface and as mentioned above, they require little to no labor from users. They also address pain points we discovered from our user research across multiple verticals.
We are currently in the process of conducting usability testing on these prototypes. Please expand to view by clicking the arrow in the upper righthand corner and be patient while the prototypes fully load.
Use case: Competitor research
Use case: Enablement
Use case: Invoice processing
Use case: Candidate search
If you would like to learn more about this project, feel free to contact me at firstname.lastname@example.org.
Thanks for reading!