As the founding designer at an AI B2B SaaS startup that predicted location-based insights, I improved 30-day active user retention from 43% to 75% by redesigning workflows to make the insights actionable rather than exploratory.
The context
Datasutram is a B2B AI-driven platform helping mid-sized businesses make strategic decisions like where to open new stores, who to target, how to expand. These companies didn’t have data science teams. So decisions were often made based on legacy knowledge and intuition, which made the process fragmented, slow, and error-prone.
Fig. Process typically followed for shortlisting sites for retail store opening at mid-sized companies
The product promised to bring structure to this using GIS, internal CRM data, and ML models to identify high-potential locations.
Fig. Datasutram used proprietary datasets and ML models to predict successful locations for store-opening
But adoption wasn’t happening.
What wasn’t working
The CXOs loved the product during demos and approved the product for internal deployment.
The intended users, the sales managers, weren’t using it after onboarding and user retention was very low.
Overview
Role
Founding Product Designer
Impact
Improved 30-day active user retention from 43% to 75%
Reduced support tickets by 40%
Duration
6 months
Team
Product, Engineering, Data Science, Sales
Scope
0 → 1
UX research
End-to-end design
Usability testing
Working with me
Rajit Bhattacharya, CEO, Datasutram
“Debeshi played a key role in shaping our B2B product’s UX, bridging user needs and technical feasibility through self-driven collaboration and rapid iteration. Her contributions directly supported onboarding 50+ enterprise clients and gaining Forbes 30 Under 30 recognition.”
The previous platform
To understand customer pain points with the product, I attended customer and sales calls, and identified usability issues which prevented the users form effectively synthesizing information to identify locations that would be ideal for opening new stores.
Fig. Summary of feedback given to the customer teams by the clients
Brainstorming the screen and filter
I streamlined the interface into a single-screen flow, fixed the filters to reflect the number of results in real time to prevent user frustrations.
Fig. Low fidelity prototypes integrating the screen components
Initially shipped design
I streamlined the interface into a single-screen flow, fixed the filters to reflect the number of results in real time to prevent user frustrations.
Fig. The initial improvised design showing areas with high business potential where business can search for properties to open stores in
But adoption didn’t move.
Talking to stakeholders to dive deeper
Customer calls and stakeholder interviews
6 customer feedback calls (client)
4 sales manager interviews (client)
4 customer & sales rep interviews (team)
Insight
The sales managers weren’t trying to explore locations. They were trying to find properties.
The recommended locations didn’t always have available properties.
That gap made the product feel useless.
Pivot from recommending locations to properties
We shifted the product from showing zones to showing available properties.
Score calibration
Adjusting location scores to match availability of properties
Actionable next steps
Pulling in real estate listings and their contacts to suit existing workflow.
Instead of just seeing “Area X can earn $100,000 annual revenue”, users could now click on a property in a high revenue potential area, and contact them for more details.
Iterating through stakeholder feedback
The redesign happened in 3 major cycles:
1. Rapid prototyping to test layout and workflow changes
2. User testing across 3 rounds — evaluating time to task, confidence, and perceived value
3. Final implementation in collaboration with engineering and data teams
“...but there are too many steps”
The first level of validation from the sales and customer teams revealed that there were too many steps involved in obtaining the list of properties, and the setting the filters was too overwhelming.
Solution
Separating the flow into:
1. An onboarding flow where the customer teams would help the clients set up the filters for stores meeting required parameters
2. A regular flow where the team can log into the portal and directly view the list of properties, reducing it to 2 steps only.
Fig. Reduced the regular flow to 2 steps from 7 steps
“...but how do we circulate the options within the team?”
The next level if iterations followed from customer validation calls, where, while context and images served well for shortlisting, exporting the property details to CSV allowed teams to circulate the data internally, so that each sales representative could make calls and speed up the shortlisting process.
Solution
I introduced:
A card view to quickly compare locations in context
A table view with exportable CSV format, so teams could shortlist and circulate internally
This addition significantly sped up their decision cycles.
Fig. Added a card view with parameter details chosen by the client for download and internal circulation
“...we have to manually check that the property meets all parameters”
When using the product, the sales managers struggled to trust the recommendations, and cross validated it with data points like footfall, income levels, and more in areas where they have successfully operating flows.
Solution
I decided to keep the data points such as footfall, income, and other parameters that are important to the business as optional toggles that they can turn on to match against the property recommendations.
Fig. The datapoints support the property recommendations and build trust
Final Design
The Outcomes
Beyond metrics, sales teams reported feeling more confident demoing the platform, and clients found it easier to get buy-in from internal stakeholders.
