ESG Covenant Compliance Platform

We started with a paid four-week discovery. From it we designed the complete platform and a phased delivery plan with costs attached. It all runs in the Microsoft 365 and Azure tenant the client already has: no extra vendor, no extra licence in phase one, and no data crossing the EU boundary.

Layer 1, intelligence: read each document once

Azure Document Intelligence runs OCR over every scanned document and evidence file. Microsoft Fabric processes the whole repository, Azure AI Search indexes it, and Azure OpenAI answers questions in three languages, using nothing but the client's documents as its source.

Five AI assistants, inside Microsoft Teams:

Layer 2, operations: every job title becomes a name

• AI pulls out the tasks. Each task in the 70 documents is extracted together with its owner, asset, domain, frequency, source clause, deadline logic, evidence needed and status. A confidence score puts uncertain items at the top of the review queue. The ESG team checks them in workshops by domain, and each job title is mapped to a named person.
• Reminders nobody can miss. Teams messages and emails go out 14, 7 and 1 days ahead. On the deadline the line manager is escalated to automatically, and after it escalation repeats every day.
• Evidence with a trail. Each uploaded file links to the task and the IFC commitment it supports. Each action carries a timestamp and a named person.
• Five automated workflows: task reminders, deadline escalation, alerts on evidence upload, task assignment and training acknowledgement.
• Data collected in a structure. Email gives way to one monthly form per asset, and a single submission serves both ESG and accounting.
• Training hub. A library and register of training, where a calendar event plus a form acknowledgement produces a training record that meets IFC standards.

Layer 3, insight: dashboards and the report to the lender

• Five Power BI dashboards: data collection, task completion, training compliance, KPI performance across energy, water, waste and GHG, and action-plan status tied to the IFC covenants.
• Screening the IFC report before it goes out. We check the draft monitoring report for phrasing an ESG language model would treat as negative or unclear, and suggest wording that is clearer, dated and backed by evidence. The goal is a report that shows real performance accurately. It never hides an issue.
• Governance. Entra ID gives role-based access. An AI governance policy and a data processing agreement are both conditions for go-live. Every register has a documented schema and is ready for when the group rolls out its ERP.

Proof first, then the big build

The team gets a working AI assistant in week four, ahead of the larger build. Delivery runs in two costed phases with clear acceptance criteria. One of them is a 30-question benchmark in three languages, which the assistant must pass at 80% or higher before go-live.

What we designed with