Transforming HR Operations with AI-Powered Solutions at Scale

nubank · Finance · 2025 · source Agentic: no · promotional bias: high · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: AI-powered HR operations

Nubank needed to scale HR operations and employee experience for more than 8,000 employees across countries using AI, LLMs, forecasting, and knowledge-base quality systems.

Constraints:

• The company had more than 8,000 employees across multiple countries.
• Employees struggled to find internal HR information.
• The People & Culture function needed to analyze thousands of open-ended feedback comments.
• Turnover and promotion planning required predictive models.
• The article provides limited operational detail on guardrails, model choices, and failures.

Architecture

Nubank deployed a portfolio of HR AI systems: AskNu as a Slack-based knowledge assistant, generative AI for engagement-survey feedback analysis, time-series turnover forecasting, machine-learning promotion budget planning, and AI quality scoring for the WikiPeople knowledge base. HR experts work with engineers, data scientists, PMs, and designers in embedded cross-functional teams.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  asknu_slack_assistant["AskNu Slack assistant"] --> wikipeople_knowledge_base["WikiPeople knowledge base"]
  wikipeople_knowledge_base --> ai_quality_scoring["AI quality scoring"]
  ai_quality_scoring --> documentation_gap_analysis["Documentation gap analysis"]
  documentation_gap_analysis --> asknu_slack_assistant
  feedback_analysis_llm["Feedback analysis LLM"] --> people_analytics_workflows["People analytics workflows"]
  turnover_forecasting_model["Turnover forecasting model"] --> people_analytics_workflows
  promotion_budget_model["Promotion budget model"] --> people_analytics_workflows

Components

Data flow

Employees ask HR questions in Slack through AskNu, which draws on internal documentation. HR survey comments are analyzed by generative AI for insights, turnover data feeds forecasting models, promotion data feeds budget planning models, and knowledge-base content is scored to identify gaps and improve WikiPeople.

Why this architecture

Nubank chose a portfolio architecture because HR scale problems were heterogeneous: knowledge access, open-text insight extraction, turnover prediction, budgeting, and documentation quality each needed different model workflows.

Models used: Large Language Models · Generative AI · time-series forecasting models · machine learning models

Scale

• Nubank had over 8,000 employees across multiple countries.
  “with over 8,000 employees across multiple countries”summary, stated
• The feedback workflow analyzed thousands of open-ended comments.
  “analyzing thousands of open-ended employee feedback comments from engagement surveys”summary, stated
• Turnover prediction accuracy improved by 14 percentage points.
  “14 percentage point increase in turnover prediction accuracy”summary, stated

Alternatives considered (roads not taken)

Centralize AI as a purely IT function

Rejected because: Nubank embedded technical capabilities directly in P&C teams.

Rely on colleague-to-colleague HR knowledge sharing

Rejected because: That created interruptions and inaccurate information flow.

Patterns

rag-retrieval (knowledge assistant)

AskNu appears to retrieve internal HR information from documentation to answer Slack questions.

data-quality-loop (knowledge base improvement)

AI identifies documentation gaps and quality issues that P&C teams then improve.

batch-offline-processing (HR analytics)

Generative AI analyzes batches of open-ended employee survey comments.

Tradeoffs

Embed AI capability inside People & Culture.

Chose Cross-functional P&C product teams over Purely centralized IT delivery.

Rationale: Embedding improves domain fit and speed but requires governance to avoid fragmentation.

Use several specialized HR AI systems.

Chose Assistant, feedback analysis, forecasting, budget planning, and knowledge scoring over One generic HR chatbot.

Rationale: Different HR operations require different model types and outputs.

Gotchas (what bit them)

Company blog posts may omit LLMOps controls.

Root cause: The case emphasizes successes but does not disclose model choices, guardrails, hallucination metrics, or access controls.

Resolution: Treat implementation maturity claims as partially uncertain and note missing operational details.

Embedded AI teams can increase speed but risk fragmentation.

Root cause: Technical capabilities sit directly inside P&C rather than a centralized AI function.

Resolution: Use governance if embedded teams proliferate.

Evaluation

Nubank reports business and HR analytics improvements, including turnover prediction accuracy, feedback insight speed, promotion forecasting, and knowledge accessibility.

Metrics: Over 8,000 employees. · Thousands of feedback comments analyzed. · 14 percentage point turnover prediction accuracy increase.

Outcomes

• Nubank deployed AskNu for internal HR knowledge access.
• Generative AI accelerated feedback analysis.
• Forecasting and budget models supported turnover and promotion planning.
• WikiPeople quality scoring improved knowledge accessibility.

The case covers over 8,000 employees and thousands of open-ended survey comments.