Our client, Torch Partners, is seeking to leverage AI to enhance their deal closure rate by providing actionable insights from their meeting notes data.
Duration
8 weeks (Ongoing)
Team
Product Designer
Data Analyst
Analytics Manager
ML Engineer
Analytics Engineer
Background
Torch Partners, a leading M&A and private capital adviser in the software and technology sectors is currently facing inefficiencies in its existing meeting notes system. Their existing system stores thousands of meeting notes and valuable information (such as client locations, recurring themes, or cross references of businesses of interest), all of which remain untagged and unsearchable without extensive manual effort.
Problem statement
Torch Partners' inability to easily find the insights required from meeting notes has led to missed opportunities and inhibit efficient use of meeting note data.
Solution
Designed and developed an MVP solution to transform these inefficiencies by developing a custom AI-powered solution that can automatically analyse, structure, and extract information previously hidden within unstructured meeting notes. This solution will make details like topics of discussion, and businesses of interest searchable and actionable, reducing time spent on manual reviews and empowering Torch Partners to make faster, more informed decisions.
As this was an eight-week project, our team needed to maintain a lean and agile process. This required tasks across discovery, design, and implementation to follow a strict timeline.
One of the initial challenges was gathering enough pain points to gain valuable insights into users' workflows and key problem areas. To address this, both the process and the team had to remain agile.
Phase 1: Research & Discovery
SME Workshop
We conducted one workshop with the SME team to better understand their services and end to end workflow. With this workshop we can begin to determine the MVP requirements and the opportunities available for full build.
Objective
Gain a comprehensive understanding of the end-to-end process of the four services that Torch Partners provides and the role of meeting notes within each service.
Activities
•Service Definitions: Understanding how Torch Partners defines each of the four services they offer
•Process Mapping: Creating process maps for each service, including key steps, decisions, systems/tools, governance processes and meetings
•Meeting Notes: Identifying how meeting notes are used in the process flow
Business Process Map: Deal Process
Business Process Map: Service map
SME Workshop
Key Takeaways
The majority of their deals are carried out through two services, M&A and Private Capital Markets
The majority of the meeting notes are generated during the deal origination stage, therefore their current tool, is used primarily capture meeting notes pre a deal pitch
They have an accompanying CRM tool which is used during the post-deal phase to track the progress of a deal
Whilst their existing meeting notes tool and CRM tool have no connection, an optimimal solution, especially for C-suite Execs is to combine the two workflows.
Next steps:
Use the insights from these workshops to define Business requirements
Use the services highlighted and fleshed out as a foundation for the structure and the questioning in the User Workshops
User Discovery Workshops
While the Business Process Map provided an understanding of key services, the overall deal-making process, and how various tools are used throughout, it was essential to delve deeper into individual user workflows. This was crucial for identifying user needs and pain points.
Insights gathered from these user workshops directly informed the user stories, which were a core deliverable at the end of the discovery phase.
SME Workshop
Key pain points
Accessibility of information: inefficiency in navigating, searching, and retrieving relevant meeting notes due to incomplete tagging and reliance on manual processes.
Sharing of knowledge and actions: Difficulty in managing information across systems, teams, and processes, leading to duplicative tagging, manual action tracking, and reliance on team members for clarity.
Structure of information: Lack of consistency in note structure and tagging, making it difficult to trace information sources and extract key data points.
Next steps:
Use the key pain points to formulate user stories
Seek feedback from the client and collaborate on the refinement of user stories
Phase 2: Define
User stories
Our goal was to create between 10 to 15 user stories, which would later be refined into a select few. To develop these stories, I first organised the pain points from the user workshops into themes. These themes helped the team establish focus areas for the user stories.
I then facilitated a workshop with the project team to transform these themes into user stories. The session included a Designer (myself), the Project Lead, an Analyst, and the ML Engineer. Having diverse perspectives enabled us to construct clear user stories with measurable success metrics.
SME Workshop
Key Challenge
Once the user stories were developed, refined and prioritised we played back a shortlist of the recommended user stories to the client. A key challenge we found here was aligning with the client what a user story consisted of. A way to mitigate this was be fully transparent and create an open feedback channel where they can prioritise the user stories in the way they envisioned.
Final User Stories
Below are the final user stories agreed with the client
User Stories
Phase 2: Design and implementatation
Wireframes
Due to time constraints and limited availability on the client side to provide feedback and a clear understanding of the output I drove straight into mid-high fidelity wireframes which were designed in Figma.
Key challenge:
The MVP product was to be built in Streamlit, a tool used to quickly build and deploy custom web apps. Due to this, the design must consider the limitations of Streamlit.
The resulting solution was an AI powered search functionality that hosted 3 main features:
Free Text chat Functionalilty
A query builder
Action Items
This feature ensure that all documents extracted assessed for visual quality, this is to improve user processes and document quality.
Phase 3: Development and testing
During the testing phase, I selected User Acceptance Testing (UAT) as the primary method for gathering user feedback. However, I also incorporated opportunities for light usability testing, including live calls for real-time observations and pressing questions, as well as group chats to facilitate seamless communication.
UAT (structure)
Duration: 3 weeks
Round one (2 weeks): Focused on the main functionality which was the chat feature
Round two (1 week): Focused on the 'Action Items' feature/page
UAT ( Resources)
This phase was enabled by a UAT script created with excel given tot he users to complete:
Structure of the script:
Intro to the app
Functionality script
User stories
UAT ( Communication)
Users were able to provide feedback through two main channels:
Email
Team group chat
Teams calls
UAT ( Feedback)
A significant amount of feedback were centered around finetuning the quality and comprehensiveness of the model responses. There was also feedback around having easier access to the next steps from meetings and the ability to predict the companies and firms users were mentioning.
A few examples below:
(Regarding the lack of predictive text) "Is this something which can be changed? It will limit the usage quite a bit if people always have to spell out the full name when they refer to investors"
"It would also be great if we could export the table of data to excel"
"Whilst the page does list the action items coming from notes, the list itself does not add value as there is no context or ownership of the items, only links to meeting notes."
Next steps
Iteratively implement changes and overall improvements to the user experience
Diversify user feedback methods
Transition from Streamlit to React
Time constraints
Time constraints were one of the biggest challenges, with the entire discovery, design, build, and testing process compressed into just eight weeks. This led to misalignment with the client on user stories and requirements, as well as a shortened user discovery phase due to scheduling issues. While I managed both effectively, they did consume valuable time. To mitigate these challenges, I focused on upfront work while the client handled scheduling and reused pre-existing templates.
Client Alignment
The misalignment was most evident during user story creation and UAT. To address this, we adopted a 'guide' approach with the client, ensuring we could gather and implement feedback while also educating them on AI capabilities. This approach fostered transparency and trust, especially with a client who was not highly technical.
Next project