From Experimentation to Scale: PhonePe’s AI Journey in HR 

At PhonePe, our strategy for AI in HR was never about deploying a single, generic solution. Instead, we set out to build a fit for purpose ecosystem prioritizing speed, flexibility, and accuracy embedded directly into daily workflows.

AI Readiness began before AI. We have been deliberate about maintaining a small number of core systems as the source of truth and building specialised experiences on top where necessary. The foundational work – data integrity, system interoperability, permissions and reliable process design is what makes more ambitious AI automation possible.

Our First Application in the Performance Management Process

In 2023, as generative AI emerged, PhonePe’s HR team began exploring how language models could reduce administrative burden across core HR processes. We started with one of our time intensive operational challenges: 360-degree feedback during performance review cycles. 

With managers overseeing multiple direct reports-each receiving upwards of ten detailed evaluations required managers to manually synthesize fragmented qualitative feedback.  Our first attempt was to summarise this feedback and highlight recurring patterns. 

Initial summarization models were a helpful starting point, but managers’ feedback highlighted opportunities to capture greater nuance and context. Approaching this as an iterative problem rather than a technology limitation , the team continuously improved data inputs, contextual framing, and theme extraction algorithms to deliver better insights.

Today, this Feedback Report has become a go-to tool for managers to understand performance patterns – not just within a single cycle, but across years. More importantly, it taught us an early lesson on creating an AI experience that managers can trust.

This was later complemented by a dedicated self-serve agent: the Feedback Generator for stakeholders providing 360 feedback-converting raw notes into objective, actionable feedback using structured prompts. 

With performance feedback structured, we tackled self-reviews – a process often hindered by goal details scattered across JIRA boards, google sheets and docs. To save employees from manually reconstructing a year of fragmented work, we launched the Goal Generator to automatically extract raw goal documents into structured PMS entries. 

Managers at PhonePe have regular 1:1s, but follow-through and tracking becomes an admin heavy task. To solve this, Note Assist structures routine check-ins end-to-end by preparing pre meeting agendas, transcribing calls, and logging clear action items. Currently being piloted, it will soon roll out across the organization. 

Scaling Learning Infrastructure

Large organisations typically possess a wealth of learning material, but employees may still struggle to locate the right information at the moment they need it. The Learning Hub acts as a 24/7 conversational interface over our internal knowledge base and learning platforms, providing relevant answers without requiring employees to search through multiple repositories.

Our philosophy centers on tapping in-house subject-matter experts to craft deeply contextual learning content. However, these experts often need significant support to create structured digital modules. To help them, our Learning Content Creator automatically parses raw domain inputs into organized curricula slashing prep time from weeks to hours. 

Finally, automated completion reports process learning-platform logs in bulk and generate executive summaries. This helps teams focus on understanding learning outcomes rather than assembling the underlying data.

Rethinking Frontline Hiring at Scale

While internal performance and learning tools optimized processes for our existing teams, streamlining the expansion of the frontline workforce became one of our larger business priorities. Meeting ongoing hiring demands required reviewing tens of thousands of top of funnel applicants each month. 

Field research including shadowing  candidate calls, discussions with Area Sales Managers and Regional Sales Managers,  revealed that candidate drop-off wasn’t caused by recruiter capacity – it stemmed from candidates entering the funnel without fully understanding the role particularly its field-travel requirements, daily responsibilities and total earning structure.

To conduct initial screening, evaluate intent, and explain role realities transparently, we implemented an AI Voice Bot. The primary hurdle wasn’t placing calls-it was building candidate trust and engagement. The team analyzed calls to map drop-off risks like unaddressed queries, background noise and pauses. Unlike humans who handle silence instinctively, the Bot had to be engineered to distinguish whether a candidate was thinking, distracted, or facing audio issues. Through continuous call reviews and field feedback, the team fine-tuned conversational pace, language handling, and talk-share balance.

Within three weeks, we piloted a prototype to make thousands of calls, balancing control, scalability, and speed. Going forward, the agent will cover over 60-70% of first-level screening, helping the organisation identify better-informed, higher-intent candidates at scale. 

Hiring Beyond Frontline

We have also been enhancing our core hiring platform to generate comprehensive Job Descriptions and dynamically provide interviewers with role tailored questions based on specific job attributes. Moving forward, the team is expanding the platform to run chatbot based screening-aiming to surface top talent faster and further reduce time to hire. 

Automating Employee Pulse Survey Insights

As PhonePe grows, extracting clear signals from employee pulse surveys becomes increasingly complex. While our dashboard tracked overall sentiment, analyzing quantitative and open-text feedback across different cohorts still required heavy manual effort. To solve this, we introduced an automated Insight Generator that analyzes survey data to surface key wins, focus areas, and cohort-specific action items. 

Employee Assistant 24/7

Acting as the central touchpoint, this conversational interface provides PhonePe employees with immediate, knowledge-driven resolutions for routine workflows and queries. By maintaining accessibility across multiple channels, the agent currently resolves nearly 70% of initial queries, significantly reducing administrative load and the volume of manual tickets.

Building an Agentic Layer for HR Operations

Beyond an employee help desk, our long-term ambition for HR Operations is to build an intelligent orchestration layer over existing systems of record progressively shifting the team from routine task execution to exception handling. Post-exit employee support is an early example. We receive a steady volume of queries from former employees and external agencies relating to full-and-final settlements, background verification, employment letters and other exit documents. Previously, an HR Operations team member needed to read each email, identify the request, retrieve the relevant information from the HRMS and prepare a response. Today, AI agents manage much of this workflow end to end: interpreting emails, classifying intent, retrieving the appropriate information and responding automatically. This allows the HR team to focus on sensitive, ambiguous or unusual cases that genuinely require human judgement. 

None of these solutions is a static implementation. Each continue to improve as we capture more context, observe user behaviour and learn from edge cases. 

Across performance, learning, hiring, employee listening and HR operations, the objective has never been automation for its own sake. It has been to reduce administrative drag, improve the clarity and consistency of decisions, and allow HR teams, managers and employees to spend more time on work requiring empathy, context and judgement.