AI in Finance: Lessons from the Frontline

In an environment marked by volatility, rapid technological change and rising expectations from business partners, Finance leaders are being asked to do ever more with less. AI has moved from theory to practice, and the conversation has shifted from whether to adopt AI to how to operationalise it responsibly and at scale. At a recent India CFO Forum session in Bangalore, Madhu Hosadurga, Global Vice President, Enterprise AI at Schneider Electric, outlined how to implement AI in Finance at scale, with concrete examples and lessons in change management.

From Pilots to Scale: A Maturity Playbook

A simple way to assess where an organisation is along its AI journey is to look across six dimensions:

• Strategy: The end-goal is a strategy tied to operating objectives like efficiency, growth and risk. The key is bringing AI into ongoing transformation projects, looking outward to ecosystems, not inward to functions.

• Competency: Move beyond isolated data science teams. Pair product managers with data scientists and MLOps (machine-learning operators). Invest in change managers. Aim for broad employee level AI literacy so that every employee becomes an AI ambassador.

• Technology: Standardise around secure AI platforms and right size models to the task. Optimise cost by using small models for routine tasks and large models where needed. Track and manage cloud and inference costs.

• Governance and Responsible AI: Set guardrails for data use, model behaviour and human approvals. Redesign processes so AI augments decisions rather than bypasses controls.

• Applications: Prioritise ‘useful’ over ‘cool’. Move beyond demos to stable production services embedded in business workflows. Target end-to-end journeys.

• Impact: Define common metrics up front and instrument use cases to report outcomes that matter to the P&L, compliance and customer experience.

Agents in the Enterprise: Escalating Levels of Ambition

• Agent-augmented workforces: Here, personal copilots and knowledge assistants help employees draft, search and create content safely and faster.

• Agent augmented workflows: At this second stage, agents are embedded in processes (sales, tendering, Finance). They read documents, extract data, take actions and prepare recommendations.

• Agentic systems: This entails full end-to-end orchestration across internal and external systems, including machine to machine interactions. This stage is still nascent but is arriving fast.

Currently, the early wins are mainly concentrated in the first two levels and human review remains part of the flow for sensitive and high-value steps.

Finance Use-Cases That Moved the Needle

• Credit risk and limit recommendations: Monthly recommendations that combine external insurer and credit data with internal payment behaviour and relationship history. Country specific models respect different risk perceptions and improve acceptance.

• Cash forecasting: Country- and entity-level models that raise forecasting accuracy and give treasury greater confidence in weekly and monthly visibility.

• Procure-to-pay and reconciliations: Automation of routine shared service activities like payment matching, with exception handling routed to people. The focus is shifting from isolated steps to endto-end process re-design.

• Tendering assistant: An AI solution that reads tender specifications and single-line diagrams to generate a bill of materials. This is integrated with the ERP to minimise change management.

Buy Versus Build: A Pragmatic Split

• Buy platforms where hyperscalers or enterprise vendors have invested heavily, for example sales copilots and horizontal model hosting.

• Build firm-specific workflows that reflect country, product and channel complexity. Cash forecasting and tendering are examples where bespoke models deliver advantage.

• As products mature, expect a 50:50 portfolio: e.g., buy a platform and build on top of it.

Operating Model and Funding

• Start with a central nucleus to establish standards and platforms. Evolve to a hybrid model where business led teams deliver smaller use-cases while large bets remain centrally funded.

• Report progress in a quarterly portfolio review with clear business cases and value tracking.

Measuring Impact

Adopt a simple scorecard that is consistent across use cases:

• Efficiency: cycle time, touch time, straight through rates

• Productivity: output per FTE, employee time saved and redeployed

• Compliance and risk: policy adherence, error rates, fraud detection

• Service quality: response time and resolution quality for internal or external customers

Risk, Control and Trust

• Use responsible AI guardrails to constrain data access, model actions and set thresholds that require human approval.

• Keep humans in the loop where the stakes are high. Over time, move toward human-on-the-loop with stronger monitoring and audit trails.

• Prepare for ecosystem scale where agents interact with external systems. Align contracts, standards and security controls accordingly.

What Leaders Should Do Next

• Set the direction: publish a short AI charter tied to business objectives and risk appetite.

• Pick 3 journeys: such as credit and collections, cash and working capital and procure-to-pay.

• Set up a hybrid team: business product owner, data scientist, MLOps engineer and change manager.

• Instrument value: define baseline metrics, build dashboards and review quarterly.

• Tighten governance: codify data use, approvals and model risk management.

Looking Ahead

Enterprise vendors are embedding agents into core platforms and new protocols will enable richer machine to machine collaboration. As these capabilities mature, the balance will tilt from augmentation to orchestration. The imperative for leadership is to build muscles today in workflows, governance and measurement so that the organisation is ready to take advantage