Rakshith Mallikarjun
Mindsprint · Case study

AtSourceEnterprise sustainability platform, re-architected around AI

Took a manual, spreadsheet-driven sustainability and traceability process and rebuilt it around AI deforestation risk modelling and automated reporting — cutting administrative effort by ~70%.

Enterprise AIESGPlatform StrategyUX Revamp
6
Major initiatives, Q3–Q4 2025
~70%
Administrative effort reduced
~50%
Manual data entry reduced
~15
Engineers and designers led

The problem

AtSource is the sustainability and supply-chain traceability platform for ofi, covering sustainability data, supplier verification, carbon footprint analysis and reporting across global sourcing programmes. Deforestation risk for farmlands was assessed manually and inconsistently, 'AtSource Ready' status was tracked in spreadsheets, and customer-facing sustainability reporting took weeks to assemble.

Discovery

A Strategic Review of the platform surfaced the pattern: analyst time was going into reading, collating and chasing evidence rather than deciding, and master data drift between DMP and AtSource was driving ingestion failures.

Customer research

  • Interviews with sustainability analysts, supplier managers and account leads across sourcing regions
  • Workflow shadowing: mapped every artefact from supplier submission to signed-off report
  • Pain points: inconsistent risk scoring, no evidence trail, reporting cycles measured in weeks
  • Insight: analysts did not want AI to decide — they wanted AI to prepare the decision

Product strategy

Position AI as a preparation layer, not a decision layer. Model deforestation risk on farmland and supply-chain data, auto-generate the sustainability and traceability narrative with citations, and keep the analyst as the accountable approver — then use the speed advantage in enterprise conversations.

Prioritisation & tradeoffs

  • Sequenced the roadmap from the AtSource Strategic Review: 6 major initiatives across Q3–Q4 2025
  • Deferred a full data-model rewrite in favour of harmonisation plus an adapter layer
  • Chose modular architecture improvements and a UX revamp over net-new module breadth

Solution

  • AI deforestation risk assessment for farmlands with environmental impact tracking across the supply chain
  • AI-generated reporting engine producing summary reports on sustainability risk and product traceability
  • 'AtSource Ready' lifecycle automation replacing manual spreadsheet tracking with system-driven verification
  • Master data harmonisation across Traceability & Transparency (DMP → AtSource)
  • UX revamp and modular architecture across DFC, DRH, DMP and Onboarding

Architecture

  • Geospatial and supplier data ingestion → farmland risk features → deforestation risk model
  • LLM reporting layer with structured output schemas and source citations
  • Human-in-the-loop review queue with full edit history as the audit trail
  • Harmonised master data layer fronting DMP and AtSource to cut ingestion failures
  • AI-enabled SDLC prototype: automated BRD creation, wireframing and workflow automation

Execution

  • Leads a cross-functional team of ~15 engineers and designers across discovery, backlog and delivery
  • Two-week cadence with fortnightly SME review of AI output quality
  • Staged rollout: internal SMEs → design partners → general availability behind flags
  • AI-assisted discovery used to accelerate BRDs and support enterprise upsell conversations

Lessons learned

  • Ship the evaluation harness before the feature. Without a golden set you cannot safely change a prompt.
  • Citations converted more enterprise stakeholders than accuracy claims ever did.
  • Master data harmonisation is unglamorous and is the single highest-leverage thing I did on this platform.