Maverick spend hid in Enterprise Resource Planning (ERP) POs; Snowflake–Power BI classification exposed contract gaps

Overview

A global manufacturer’s visibility into spend was limited because analytics were pieced together in ad hoc tools. Purchase orders (POs), invoices, and contracts lived in different systems with inconsistent supplier naming and category mappings, so maverick and off?contract spend surfaced late and category managers argued from different numbers. Intelligex ingested POs, invoices, and contract data into Snowflake, applied AI?assisted classification governed by a finance taxonomy, and delivered curated dashboards in Power BI. Outliers were flagged early, sourcing decisions leaned on shared evidence, and off?contract purchasing declined under consistent, finance?owned rules—without replacing the ERP, procurement tools, or contract repositories.

Client Profile

• Industry: Industrial manufacturing (global plants, direct and indirect spend)
• Company size (range): Enterprise with regional procurement and centralized Controllership
• Stage: ERP for P2P and invoicing; contracts stored across CLM and SharePoint; spend analytics maintained in spreadsheets and point tools
• Department owner: Finance & Accounting (Procurement Finance and Controllership)
• Other stakeholders: Strategic Sourcing, Plant Procurement, IT/Integrations, Legal/Contracts, FP&A, Internal Audit, Business Unit Leaders

The Challenge

Spend data originated in different sources and rarely lined up. PO lines and goods receipts lived in the ERP, invoices carried vendor remittance details and tax, and contracts sat in a mix of contract lifecycle management (CLM) tools and shared drives. Supplier names were spelled differently across systems, categories varied by region, and contract price lists were not consistently referenced on POs. Analysts pulled extracts into spreadsheets and BI point solutions, but transformations were bespoke, undocumented, and hard to reproduce.

Category managers needed to spot maverick spend, price creep, and missed consolidation opportunities. Instead, they reconciled basic totals and debated which extract was correct. Off?contract buys and tail suppliers slipped through because contract coverage and price compliance were not evaluated consistently. Dispute conversations relied on screenshots rather than on a shared model, and audit requests required rebuilding the lineage from invoice to contract and category assignment.

Time and people were constrained. The team could not maintain multiple homemade models across regions and plants, and integrations to procurement tools were brittle. There was no central taxonomy to normalize categories or suppliers, and rule changes lived in notebooks rather than in a governed registry.

Why It Was Happening

Root causes were fragmented inputs and ungoverned classification. Vendor masters were inconsistent across entities, category codes differed by region, and contracts were stored as documents without structured terms for price and scope. Ad hoc models used different keys and assumptions, so mappings and filters were rebuilt each time. Without a canonical schema and a finance?owned taxonomy, AI or rules could not reliably classify spend or identify off?contract purchases.

Ownership was diffuse. Procurement set category strategies, Finance owned reporting, IT moved files, and Legal maintained contracts. No shared workflow defined how supplier normalization, category assignment, contract coverage, and price compliance would be evaluated and approved before insights hit dashboards. As a result, outliers were found late and disputes centered on data assembly rather than on action.

The Solution

Intelligex delivered a spend pipeline anchored in Snowflake with AI?assisted classification governed by a finance taxonomy and surfaced through Power BI. POs, receipts, and invoices were ingested from the ERP; contract metadata and price lists were extracted from the CLM and shared repositories; supplier identities were normalized to a single master. A classification engine assigned categories and contract coverage using a rules layer aligned to standards such as the UNSPSC model, augmented by AI to resolve ambiguous lines. Human?in?the?loop review handled borderline cases, and approved mappings fed back into the rules. Dashboards highlighted maverick spend, price variances, and consolidation opportunities. The platform leveraged Snowflake for scalable data processing and Power BI for curated analytics, with the ERP and CLM remaining systems of record.

• Integrations: ERP extracts for POs, receipts, vendor masters, and invoices (for example, SAP or Oracle); contract metadata and price lists from CLM or SharePoint; data pipelines and models in Snowflake; dashboards in Power BI.
• Canonical spend schema: Standard fields for supplier, parent supplier, category, contract ID, item/description, quantity, unit price, tax/freight, plant, business unit, and buyer; identity crosswalks across regions and entities.
• Taxonomy and mappings: Finance?owned category taxonomy aligned to UNSPSC concepts; supplier normalization and parent/child roll?ups; effective?dated mapping tables under change control.
• AI?assisted classification: Model to classify free?text PO lines and invoice descriptions to categories and detect likely contract coverage; confidence scores route low?confidence cases to review; approved outcomes retrain the model.
• Contract and price compliance: Price list matching and variance detection; flags for off?contract buys and unapproved suppliers; tolerance thresholds governed by Finance.
• Exception workflow: Maker?checker reviews for reclassification, supplier merges, and large price variances; reason codes and attachments required; audit trail preserved.
• Dashboards and alerts: Maverick spend by category and plant, vendor consolidation opportunities, price variance patterns, contract coverage views, and tail spend trends; subscription alerts for threshold breaches.
• Audit and lineage: End?to?end traceability from dashboard tile to source PO/invoice and contract clause; immutable logs of mapping changes, approvals, and model versions.

Implementation

• Discovery: Cataloged ERP fields and regional differences, supplier master quality, contract repositories, and price list formats; reviewed existing category structures and sourcing priorities; gathered examples of maverick spend and recurring disputes; captured audit feedback on evidence and lineage.
• Design: Defined the canonical schema and identity crosswalks; authored the finance?owned taxonomy and mapping governance; designed AI features for description and supplier context; specified contract coverage and price variance rules; planned exception workflows, dashboards, and access controls.
• Build: Implemented ingestion to Snowflake for POs, invoices, and contracts; built normalization for suppliers and categories; developed the AI classifier with human?in?the?loop review; configured contract/price compliance checks; assembled Power BI datasets and dashboards; added maker?checker approvals and logging.
• Testing/QA: Ran in shadow mode: produced draft classifications and maverick flags while legacy reports continued; reconciled totals to ERP; tested contract price matches and variance thresholds; tuned model features and mapping rules with Procurement and Finance reviewers.
• Rollout: Launched dashboards for selected categories and plants first; retained spreadsheet reports as a controlled fallback; expanded coverage by region and category after a stable cycle of classifications and approvals; enabled alerting and stricter governance as adoption grew.
• Training/hand?off: Delivered sessions for category managers, plant buyers, and Finance on reading dashboards, requesting reclassifications, and approving supplier merges; updated SOPs for contract uploads and price list maintenance; transferred ownership of taxonomy, mappings, and dashboards to Procurement Finance under change control.
• Human?in?the?loop review: Established a monthly forum to review classification accuracy, mapping changes, and threshold updates; decisions recorded with rationale and effective dates.

Results

Spend analytics moved from reconciled snapshots to a governed, shared model. Category managers saw off?contract purchases, price creep, and tail supplier fragmentation as they formed, not after quarter?end. Reclassifications and supplier merges flowed through approvals, and dashboards traced every insight to the underlying PO or invoice and the contract clause or price list that applied.

Disputes and cycle friction fell. Procurement and Finance reviewed the same taxonomy and mapping logic, so conversations focused on actions: consolidating suppliers, enforcing contracts, and adjusting thresholds where policy warranted. Audit requests drew from the same lineage the business used daily. ERP, CLM, and reporting tools remained in place; the addition was a Snowflake?based pipeline with AI?assisted classification and Power BI views that made spend both visible and defensible.

What Changed for the Team

• Before: Each analyst built one?off transforms. After: A canonical model in Snowflake fed governed Power BI datasets.
• Before: Suppliers and categories differed by region. After: Normalized masters and a finance?owned taxonomy aligned views globally.
• Before: Maverick spend surfaced late. After: Off?contract and unapproved supplier flags appeared with evidence and reason codes.
• Before: Price checks were manual. After: Price lists drove automated variance detection with thresholds and approvals.
• Before: Reclassifications lived in email. After: Maker?checker approvals logged mapping changes with lineage.
• Before: Audits required re?creation. After: Dashboards linked to source POs, invoices, and contract citations.

Key Takeaways

• Normalize suppliers and categories first; a finance?owned taxonomy is the foundation for credible analytics.
• Centralize data; use a warehouse like Snowflake to align POs, invoices, and contracts under one schema.
• Use AI with guardrails; combine model?based classification with human review and effective?dated mappings.
• Make contract coverage explicit; attach price lists and contract IDs to spend and monitor variance with governed thresholds.
• Treat exceptions as workflow; maker?checker approvals and reason codes reduce rework and strengthen auditability.
• Integrate, don’t replace; keep your ERP and CLM while adding governed models and Power BI dashboards.

FAQ

What tools did this integrate with? The pipeline ingested POs, invoices, vendor masters, and receipts from the ERP (for example, SAP or Oracle), extracted contract metadata and price lists from CLM or SharePoint, processed and modeled data in Snowflake, and delivered curated analytics through Power BI. Category design aligned to concepts in UNSPSC.

How did you handle quality control and governance? Supplier and category mappings lived in a finance?owned registry with effective dating. AI classifications carried confidence scores; low?confidence cases went to a review queue. Reclassifications, supplier merges, and threshold changes required maker?checker approval with rationale. Every mapping, model version, approval, and dashboard refresh was immutably logged with linkbacks to source records.

How did you roll this out without disruption? The model ran in shadow mode first, generating draft classifications and maverick flags while teams used existing reports. Results were reconciled to ERP totals, and mappings and thresholds were tuned. Rollout started with a handful of categories and plants, then expanded across regions once stability and adoption were established. Legacy reports remained as a controlled fallback during early cycles.

How were categories and suppliers standardized? Supplier identities were normalized with parent/child relationships, aliases, and remittance details. Category assignment used a finance?owned taxonomy aligned to UNSPSC concepts, with rules and AI signals from descriptions, item codes, and contract references. Approved mappings updated the master and improved future classifications.

How did you detect maverick or off?contract spend? The engine matched PO and invoice lines to contract IDs and price lists; gaps flagged likely off?contract buys. Price variance checks compared unit prices to agreed rates with tolerance by category. Purchases from unapproved suppliers or outside preferred categories generated alerts and entered a review workflow with evidence for category managers.

How was sensitive commercial data protected? Access to supplier pricing and contracts was role?based. Power BI datasets exposed only necessary fields, and row?level security limited views by region or business unit. All data pulls, mappings, and dashboard views were logged for audit.