Overview
Our client is a leading Indian consumer fintech platform. Its merchant lending vertical, which offers working capital and top-up loans to a nationwide merchant network, was scaling rapidly and needed an outreach layer that could scale with it.
The engagement focused on the drop-chase journey, targeting two critical milestones in the funnel: Offer Selection at the top and Agreement Signing at the bottom.
The Challenge
Merchant lending was one of the client's fastest-growing verticals, and human-led outreach was becoming a growth bottleneck.
Low conversion on the existing human-agent workflow, especially on drop-chase
Limited bandwidth meant a significant share of eligible leads were never called at all
Scale mismatch, as the vertical was scaling faster than a human-only model could support
Channel fragmentation, with IVR and email outreach becoming redundant and the client wanting to consolidate them into a single AI-led omnichannel stack
The client needed an omnichannel outreach layer that could convert better, catch every eligible lead, and scale quickly.
The Solution
SquadStack deployed a Voice AI Agent as the first interaction layer on every dropped lead, with conversations optimized in natural Hinglish for two milestones: Offer Selection at the top of the funnel and Agreement Signing at the bottom.
Three things made the system work in production:
Persistent Memory: The AI Agent carries memory of the last few interactions with the same merchant, so follow-up calls do not start from zero and merchants are not asked the same questions twice
Context Management: The agent holds full conversational context within a call, so multi-minute merchant conversations stay coherent end to end without losing the thread on eligibility, offer terms, or next steps
Optimize, SquadStack's multivariate testing engine: Every prompt variant, voice, and call strategy was run as a controlled experiment in production, measured against Offer Selection and Agreement Signing’
How the system improved every month:
What separated this engagement from a typical pilot was the cadence at which the Voice AI Agent improved once it was live.
Mid-November. Same voice, different scripts. Isolated the impact of script variants on Offer Selection and Agreement conversion.
Mid-November. Same script, different voices. Isolated the impact of voice choice on connectivity and completion.
Mid-March. Persistent memory. The AI agent now carries memory of the last few calls with the same merchant, eliminating the "start from zero" problem on follow-ups.
The Impact
Across three months of production traffic, the AI-native workflow consistently outperformed the incumbent process across both critical milestones of the merchant lending funnel.
Key Outcomes
Offer Selection uplift of ~20–30%
Agreement uplift of ~60–70%
Incremental CAC for Offer Selection uplift dropped by ~48%, and for Agreement uplift by ~18%, as the system kept learning on more data.
The Not-Visited cohort insight
The cohort-level story is where the real value shows up. Across all three months, uplift for Not-Visited cohorts, i.e. leads that had not yet been reached by a human agent, was dramatically higher than for Visited cohorts.
On Agreement, Not-Visited uplift ran at 76-94%, confirming that AI is exceptionally strong at converting cold and never-reached leads: exactly the leads that previously leaked out of the funnel because of bandwidth constraints.
Looking Ahead
What started as a POC on merchant lending drop-chase has expanded into a multi-use-case engagement across consumer lending, collections, insurance, and merchant support. The engagement shows how Voice AI, deployed as the first interaction layer, compounds: every new use case builds on the infrastructure and learnings of the previous one.
