FBFreight Base System
Guides › Implementation

From Raw AP Extract to Savings Pipeline: An 18-Step Spend Analysis Playbook

A spend cube that finance trusts and procurement can act on takes 18 steps, from reconciling the AP extract to ranking categories by savings you can bank. Here's the sequence, plus where to buy instead of build.

Freight Base System editorial team · November 2026 · 8 min read

Scope the Spend Cube Before Anyone Opens the AP Extract

Most spend analysis projects don't fail on technology. They fail because the team pulls three years of data before agreeing on what decisions the output has to support. Months later there's a dashboard nobody uses and a CFO asking where the savings are. Start with scope. The data work comes after.

Step 1: Name the decisions. Write down the three to five questions the cube must answer. Typical ones: which categories go to market next year, where spend is fragmented across too many suppliers, how much tail spend could move to a catalog or marketplace, and which suppliers are getting paid early. If a dimension doesn't serve one of these questions, leave it out of version one.

Advertisement

Step 2: Set the perimeter. Use 12 to 24 trailing months. Twelve shows current run-rate; 24 exposes seasonality and price drift. List the legal entities and ERPs in scope, and decide now what's excluded: payroll, taxes and government levies, intercompany charges, dividends, donations and loan repayments. Teams that skip this end up "discovering" savings in tax payments.

Step 3: Fix the dimensions. A usable spend cube needs four axes: who spent (entity, business unit, cost center), what was bought (category), from whom (normalized supplier) and when (period). Add contract status and payment terms as attributes if you can. Leave the category design for Step 11.

Extract AP and PO Data That Reconciles to the Ledger

Step 4: Pull line-level data from every source. You need AP invoice lines (not headers), PO lines, the vendor master, the GL chart of accounts, and corporate card or P-card transactions. Header-level AP data hides the fact that one distributor invoice can cover cleaning supplies, PPE and spare parts at once.

Step 5: Reconcile to the general ledger. Total the extract by entity and period, then compare it to the GL. A gap above roughly 1 to 2% means something's missing: a second ERP instance, manual journals, or a subsidiary still running on spreadsheets. Get finance to sign off on the reconciliation before you go further. If they don't trust the total, nothing downstream gets used.

Step 6: Normalize currency and tax. Convert to one reporting currency at monthly average rates, and keep the original currency in its own field so category managers can see local price exposure. Strip out recoverable VAT or GST so you compare net spend. Businesses that can't recover input tax, such as many financial services and healthcare providers in the EU and UK, may need gross figures instead. Where withholding tax is deducted at source, as in India and parts of Latin America, check whether AP records the gross invoice or the net payment. Mixing the two skews supplier totals.

Step 7: Flag addressability. Tag each line as addressable, partly addressable (utilities in regulated markets, for example) or non-addressable (rent under a long lease, statutory fees). Don't assume most of your spend is addressable; measure it. The share varies widely by industry, and an inflated figure is the fastest way to overpromise savings.

Normalize Suppliers Into Parents You Can Negotiate With

Your vendor master probably lists the same supplier five ways. Until those records collapse into one, supplier counts are inflated and your leverage is understated.

Step 8: Deduplicate on tax IDs first, names second. Match on registration numbers wherever they exist: VAT numbers in Europe, GSTIN in India, TRN in the UAE, CNPJ in Brazil, EIN in the US, ABN in Australia. Then clean names. Strip legal suffixes (Ltd, GmbH, Pvt Ltd, S.A. de C.V., LLC, Sdn Bhd), punctuation and branch qualifiers, and use fuzzy matching for what's left. Review every fuzzy match above a set spend threshold by hand.

Step 9: Roll up to parent families. A logistics group that invoices through eight national subsidiaries is one negotiation, not eight. Use a commercial firmographic database for corporate hierarchies, or map them manually for the suppliers that make up the bulk of your spend. That's usually a few hundred records, even in companies with thousands of active vendors.

Step 10: Lock down personal data. Sole traders and freelancers in the vendor master can count as personal data under the EU and UK GDPR, India's DPDP Act, Brazil's LGPD and similar laws. Limit who can see bank details and individual names. If an outside classification provider processes the data, put a data processing agreement in place and confirm where the data is hosted. Your privacy counsel should review this before any extract leaves your environment.

Choose a Taxonomy and Prove Classification Accuracy

Step 11: Pick the taxonomy. This choice shapes every report that follows, so make it on purpose.

OptionBest forTrade-off
UNSPSC (standard codes)Multi-country groups, public-sector suppliers, benchmarkingDeep and product-centric; weak on services and logistics. Category managers rarely think in its terms.
Custom three-level taxonomySingle-business companies with a mature category teamMirrors how you buy, but can't be benchmarked and tends to drift without governance
Hybrid (custom L1–L2 mapped to UNSPSC at L3–L4)Most mid-size and large organizationsMapping takes upfront effort, but you get usability plus a standard backbone

For organizations with more than one ERP or country, the hybrid is the better fit. UNSPSC gives you a stable reference, and a custom top layer such as "Freight & Logistics > Ocean > FCL" matches how your category owners actually manage spend.

Step 12: Classify in passes. Start with rules on supplier plus GL account, which often cover half or more of spend. Then classify on line-item descriptions. Machine-learning models or a vendor service can handle the long tail. Write down every rule so the next refresh gives the same result.

Step 13: Test accuracy on a stratified sample. Pull 300 to 500 lines, weighted toward high-spend categories but including tail spend, and have category experts mark each one right or wrong. Report accuracy two ways: by spend value and by line count. Reasonable targets are around 95% by spend at level 2 and 85 to 90% at level 3. If a vendor quotes a single accuracy number, treat it as a red flag until they say which level it applies to and how it was measured.

Step 14: Assign owners and a refresh cadence. Each level-2 category needs an owner who signs off on its classification. Refresh monthly or quarterly. A one-off cube is stale within two quarters.

Build or Buy: The Checkpoint After Your First Cube

Step 15: Assess your build capacity honestly. Count the ERPs, the annual invoice lines and the refreshes you need each year, then check whether you have in-house BI and data engineering skills. One ERP, a modest line volume and a capable analyst usually means you can build. Four ERPs across three continents usually means you can't keep a build running.

Step 16: Decide against real costs. The figures below are typical 2026 market ranges. They vary by contract, region and data volume, so confirm them with vendors.

ApproachIndicative annual cost (USD)Time to first cubeWatch for
In-house build (SQL/BI tool plus analyst time)$20k–$80k, mostly staff time8–16 weeksKey-person dependency; refreshes slipping
BI build plus outsourced classification service$40k–$120k6–10 weeksOpaque classification rules; data residency
Dedicated spend analytics software (e.g., Sievo, Coupa, SAP Ariba, JAGGAER, Zycus, Simfoni)$60k–$400k+, based on spend under management and modules8–14 weeksPer-source connector fees; extra charges for reclassification; lock-in on taxonomy

Before you sign for spend analytics software, ask each vendor for three things: accuracy measured on your sample data, the cost of adding another ERP source, and a data export in a usable format if you leave. Vendors that won't run a pilot on your own extract deserve extra scrutiny.

Rank Categories by Savings You Can Actually Bank

A category's size tells you nothing about how much you can save in it. The biggest category might be locked into a three-year contract renegotiated last quarter.

Step 17: Score each category on four factors. Multiply addressable spend by a realistic savings range for the lever you'd use: consolidation, rebid, specification change or demand management. Then adjust for two more factors. The first is timing: when does the current contract allow you to act? The second is feasibility: market competitiveness, switching cost and how much stakeholders will resist. Score timing and feasibility from 0.2 to 1.0 each.

Here's a hypothetical. A distributor has $6M in addressable road freight with 40 carriers. A rebid that consolidates to eight carriers might yield 6 to 10%, or $360k to $600k gross. The contracts roll over in four months (timing 0.9), and the carrier market is competitive (feasibility 0.8). The risk-adjusted midpoint comes to about $345k. Now take $9M of packaging under a contract with 26 months left (timing 0.3, feasibility assumed 1.0). At 4 to 7%, it scores roughly $150k. Freight goes first, even though packaging is the bigger category.

Step 18: Run the results as a stage-gated pipeline. Move each initiative through four stages: identified, validated by finance, contracted, realized in the P&L. Finance has to agree on the baseline before anything counts as validated. That one rule does more for the credibility of category management savings than any dashboard.

A decision rule for the first cycle: launch the top three categories by risk-adjusted score that also have a contract event in the next six months. Re-score everything when the next quarterly refresh lands.

Quick answers

How long does it take to build a first usable spend cube?

For a single-ERP company with clean AP data, roughly 6 to 10 weeks is realistic. Multi-ERP, multi-country groups should plan on 10 to 16 weeks, and most of that time goes to reconciliation and supplier normalization rather than tooling. Getting finance to sign off on the GL reconciliation early is the most reliable way to keep the project on schedule.

Should we use UNSPSC or our own category taxonomy?

Most organizations with more than one country or ERP do best with a hybrid: a custom top two levels that match how category managers work, mapped to UNSPSC at the lower levels. Pure UNSPSC is hard for business users to read and weak on services. A purely custom taxonomy is easy to use but can't be benchmarked, and it drifts without governance.

What classification accuracy should we demand from a spend analytics vendor?

Ask for accuracy by spend value and by line count, reported separately at each taxonomy level. Around 95% by spend at level 2 and 85 to 90% at level 3 are reasonable targets. Insist the numbers come from a stratified sample of your own data, checked by your category experts, not from the vendor's reference dataset.

Written by the Freight Base System editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.