Web Application
Sep 2023 – Apr 2024
How I stopped worrying and embraced Speech Analytics
Spokn AI is an AI-powered speech analytics tool that simplifies call centre managers work.
Responsibilities
· Initial requirements gathering
· Competitors Research
· User Research
· UI Design and Prototyping
· Presenting to Stakeholders
· Close Collaboration with Dev Team
Team
Results
· designed and shipped MVP of AI-powered solution from scratch in 7 months
· attracted a new customer with 200+ seats
· simplified customer's Quality Assessment specialists working process
MaxContact is a Contact Centre Software Company serving businesses with dozens and hundreds of call centre workers. MaxContact has been ranked one of the top 50 fastest growing technology companies in the North consecutively from 2021-2023, and in 2022 received the IT Vendor of the Year Award.
I was involved in the project as a product designer having expertise with data structures and visualisation and having shipped several RNDs, what means I'm productive working in changing environments.
Dashboard
Playback
Interaction details
MaxContact came to us with a roadmap and had already developed the first features. The goals of the project were:
Discover
To understand our main user and ways to improve their efficiency, I started with:
· defining key system roles
· analysing working processes of the users as-is
The client has conducted research before we joined the project. Also, I was lucky to have clients who are real subject-matter experts and have worked in user roles for years before becoming managers.
I worked with available sources, also conducting desk research and competitor research.
We started with user persona for a system role crucial for the process – quality assessment specialist (QA). Persona’s goals, tasks and challenges were used in scope design and prioritization in accordance to the time and team capacity limitations.
Persona – Quality Assessment Manager
Define
in other words,
assessing call recordings
is exhausting and time-consuming.
How can I design a system that helps the QA stop spending effort on insignificant conversations and start dedicating more time to the real people?
To solve the primary user problem, I focussed on finding solutions allowing to reduce time QAs spend on selecting and listening to the calls. Help the QAs start spending their effort only on the meaningful calls and start dedicating more time to the real people.
We defined the main product principle to follow –
human-in-the-loop:
· User can edit the transcript and, in perspective, the analytics results.
· System doesn’t "spit out" the interactions without the context.
Based on it, we collected and prioritised the functionality requirements.
In the beginning, I had only a huge document with all the high-level user stories documented and used them to reconstruct the user flow. Based on them, the group of VP of Engineering, PO, developers and me created a story map that helped us focus on functionality that could form an MVP.
Story Map
MVP scope:
dashboard providing high-level insights with visualisation and drill-down to the selection of most meaningful conversations.
Develop
How to align the user problems, large amounts of data and emerging technological solutions, having to:
· implement navigation to the interactions using dashboard visualisation of meaningful parameters.
· provide the call info without making the user open call details.
Exploration and prototyping
Based on our story map and prioritisation processes, I created sketches resulting in a lo-fi screen flow. How can we combine high-level KPI values and detailed visualisations to access the calls selections?
Sketches
UI Searches
Screenflow Searches
Constraints
During prior project phases, our team met our main constraints:
· limited time,
· limited team capacity,
· limited technical opportunities,
· no direct access to the end users.
Outcomes
After reviewing first UI design explorations with stakeholders, we decided to make edits:
· Keywords and phrases changed to Topics because of more broad usage and opportunities of selected speech analytics method.
· Interactions list was moved to the Playback page similar to one existing in existing manager's tool to keep behaviour of the service familiar for existing customers.
· Coaching functionality, agent’s details got out of the scope because of tight time constraints.
Deliver
After finishing the MVP, MaxContact gave access to the platform to the beta clients right away.
By doing that, we
· tested the systems on the real amounts of production data – no tester can generate so much test data
· got direct feedback from live users
· brought value to their businesses immediately
Validation
We had two ways of validating the solutions.
· Calls with customers with prototype demo walkthrough in parallel with development process. After those sessions, we aligned on problems and made some edits in designs.
· Learning the hard way – getting feedback from active users after MVP launch.
This is the feedback we got from the beta users:
“Spokn's speech transcription and call summary features are helpful, especially for reducing the need for manual sales documentation”.
“These insights would allow us to address issues without extensive manual review, thereby saving time and improving efficiency.”
Requests and expectations:
· Proactive Insights in Dashboard
· Less effort to access individual interactions
· Advanced work of topic and sentiment detection
Results
· Spokn AI attracted in a new customer with 200+ seats
· I designed and shipped MVP of AI-powered solution from scratch in 7 months
· Speech Analytics simplified Quality Assessment managers’ work and boosted performance
Next steps:
· Improvements and refactoring in user experience based on user feedback
· Exploring generative AI in speech analytics
Reflect
My challenges:
· Constant change of requirements on first stages, with development already in progress.
Solution: accept the uncertainty; defend the scope with your life.
· Need for a second opinion.
Sometimes there is enough time, but not enough brains to focus on each field thoroughly enough – you have to be a jack of all trades.
Solution: asking coworkers, hallway testing, asking client’s UX lead for participation.
· Visually simple solutions can be a problem for the development – and vice versa.
Building real big data visualisation from scratch requires days and weeks of testing and causes unexpected issues. Line chart took us a month to test!
Solution: involving developers in the process as early as possible; but accept that there can be unexpected things anyway.
Main takeaways:
Understand and manage expectations
Even after demoing prototypes to the customers, we found out they anticipated proactive AI-generated insights, not just a descriptive dashboard.
Collaborate with development as early as possible.
I always do it, but there is yet another proof. Also, developers feel happier developing nice UI 🙂
Data analytics doesn’t have to be complicated
Never assume that a request for data analytics means a complex analytics tool. In 99% of the cases, the goal is simpler than exploratory data analysis.