Smarter Tokenomics Cut Costs 20%, Saved £750K

RESULTS

£750K

savings over two years

~98%

data accuracy

~20%

reduction in blended token cost

Executive Summary

When a global consumer goods company had blind spots in its cloud and AI spending, EPAM built a tokenized FinOps platform that delivered end-to-end spend traceability, stronger AI governance and £750K cost savings over two years.

SERVICES

Artificial Intelligence
Cloud
Data & Analytics

STRATEGIC PARTNERS

Databricks
Microsoft

INDUSTRY

Consumer

THE CHALLENGE

Fragmented Spend, Fractured Accountability

As enterprises become more AI-native, understanding how token consumption affects costs is essential. Without real-time visibility into cloud consumption and token-based AI usage, organizations struggle to assign accountability, control spend, prevent waste and scale AI sustainably.

This was the challenge for a leading consumer goods company with operations in nearly 200 countries. The company lacked clear visibility into Microsoft Azure and Databricks spending, relied on outdated Power BI dashboards and had incomplete cost data. Without effective governance controls, it could not accurately attribute spend, track AI token usage or proactively manage rising cloud costs.

The company faced several challenges:

Unattributed Shared Spend

High-volume, pooled Azure and Databricks resources were operating as anonymous "black boxes" with no way to identify which product team was generating what cost.

Lack of Accountability

Costs landed centrally on IT departments rather than on the product teams generating the spend — leaving those teams without visibility, incentive or ownership to optimize.

Opaque AI & Token Expenditures

Fast-growing consumption of Azure OpenAI models in Microsoft Foundry lacked per-product token tracking, leaving the company without unit-cost benchmarks or governance.

Reactive Financial Control:

The lack of automated policy-driven controls meant that untagged, untraceable or runaway resource provisioning went unnoticed until monthly invoices arrived.

THE SOLUTION

A Tokenized FinOps Platform

To address these challenges, the company turned to its longstanding partner EPAM for its cloud and AI expertise with strong capabilities in Databricks and Microsoft technologies. We worked closely within a tight four-month timeline to build a data and AI FinOps solution from scratch.

The platform was built around a robust, tag-driven FinOps operating model governed by a single operating principle: every unit of data and AI spend must be attributable to an owner. The solution standardizes tag taxonomies across the entire technology estate, ingests these cost signals into a governed lakehouse, attributes the costs directly to product teams and closes the loop with budget guardrails. With real-time cost visibility, accurate financial reporting and automated governance controls, it replaced outdated Power BI dashboards with redesigned reporting and policies that control expenses across the company’s cloud infrastructure.

The solution consisted of the following components:

Cost Visualization

Eight Power BI dashboards show spending by product, region and service. The dashboards track product-level costs, predictive forecasts, budget utilization and automated monthly chargeback statements directly to the owning product streams. EPAM designed and deployed these from scratch to replace legacy systems.

Data Pipelines

Rebuilt transformation logic extracts cost data automatically from Azure Advisor and Databricks into Delta Lake tables while preventing duplicate records and data-quality drift. EPAM optimized SQL code, removed legacy logic and reused 15-20% of existing data transformations.

Governance & Policy

Azure Policy was implemented to enforce standardized tagging and cost controls, and a tag taxonomy was applied enterprise-wide. Automated policy guardrails were configured to block non-compliant or untagged provisioning at the source. Interactive budgets and forecasting tools now translate this clean attribution into proactive, forward-looking financial guardrails. EPAM created policies targeting storage cost reduction, Azure VM series standards and environment-based compute restrictions.

FinOps Metrics & Analytics

37 KPIs were defined for cost tracking and forecasting. EPAM built queries and dashboards to monitor Azure Databricks reservations, virtual machine sizing and consumption patterns. All ingestion records are now pulled into a unified, FOCUS (FinOps Open Cost & Usage Specification)-aligned data model.

Cost Optimization Roadmap

EPAM provided recommendations and a roadmap for sustained savings including policy automation, tagging strategies and prioritized optimization opportunities.

AI Token Attribution

EPAM ensured each product uses a dedicated instance, turning the deployment itself into an attribution signal. Per-deployment token metrics are captured, attributed to the product, translated into a blended cost per 1,000 tokens and charged back precisely like infrastructure.