Spend analytics software that classifies every dollar and pins every saving to the invoice lines behind it
Every dollar classified, and every saving pinned to the invoice lines behind it.
See one case, screen by screen ↓Spend Analytics is spend analytics software that classifies and cleans all spend by category and supplier to find savings for procurement. Seven agents pull every invoice, card and travel line and reconcile each source to the ledger, merge duplicate vendor records into parent companies, place every line in your category tree with a reason in words, and pin every saving to the invoice lines behind it. Category owners sign each rule and saving, and finance signs what is banked.
How to choose AI solutions for back-office and enterprise work →
Updated
Spend sits in millions of invoice, card and travel lines across several ERPs, coded to whatever ledger account was to hand when the invoice was paid. A notebook bought by free-text requisition lands in office supplies, repairs or sundry; a card purchase lands under the merchant code of an electronics store. The vendor master hides the same company under several records — one per currency, a subsidiary, a former name, a card merchant that pays into the same bank account — so a strategic relationship looks like several small ones.
So the spend cube becomes a project instead of a habit. The analyst spends the first week of the month rebuilding it, category managers do not trust the classification and rebuild their own numbers in a spreadsheet, and corrections made line by line vanish at the next refresh. A savings number on a slide cannot be traced to the invoices behind it or checked against the contract, and savings found are rarely savings finance has banked.
Estimated split for one monthly refresh in one large category, by hand and with the solution. Days are working days for the spend analyst and the category manager; decisions still take a person.
Seven agents load, merge, classify, check and size savings; the category manager signs every rule and saving, and finance signs what is banked.
Kestrel & Vane Group spends $6.20 billion a year with outside suppliers, in 4,182,640 lines across six spend systems. On Wednesday 14 October the agents publish the October refresh before the working day starts. Anouk Verhoeven, category manager for IT hardware and devices, opens it. Here is what happens next, screen by screen, in the working solution.
Her home is two things: a conversation with the spend agents on the left and the spend ladder on the right, from all spend ($6.20B) through Indirect and IT & telecoms to IT hardware ($118.4M, 41,870 lines). The morning note says the refresh is reconciled to the ledger, that the agents moved $45.2M in IT hardware to the right category, and that three things wait for her: 9 supplier families, 6 rules (1 held below target) and a new saving, OP-0412, at $612.0k a year.
Every rung shows how much the agents moved from the ledger coding and the accuracy the validator measured — IT hardware: 38.2% moved, 96.9% accurate, 93 of 96 sampled.
She splits the IT hardware rung by supplier and the normaliser shows its work: every family is one company by registry number, ownership filing, tax ID or bank account. The 41 records collapse into 9 parent suppliers on the ladder. One candidate, Arden Digital Media, is held for her rather than merged.
“41 records → 9 parent suppliers. Proposed by the normaliser, waiting for your review.”
The classifier answers in the thread and the ladder follows it down to End-user devices. 1,284 lines name the Aster 14 G5 notebook but were coded to Office supplies (612 lines), Repairs & maintenance (301), Sundry (214) and card “electronics stores” (157). It proposes moving all $5.31M to End-user devices › Laptops under rule RL-118; the validator checked 60 of them blind and agreed on all 60.
“Nothing moves until you accept the rule.”
Each number in the answer carries a citation. The first opens the invoice line itself: Vantor Direct Ltd, invoice VD-88214, line 1, 4 Aug 2026, delivered to Brno, free-text request REQ-55021, coded to GL 6990 Sundry, $2,824.00 — “2 × Corvel Aster 14 G5 notebook … Unit price $1,412.00”, with the words the agent read highlighted.
“Nothing on this page was typed by an agent; it is the record as it sits in ERP Europe.”
Her October review has five steps — supplier families, rules, unsure lines, disagreements, savings — with 107 items waiting. Northgate Technology comes first, 99% confident: eight records in four systems, $31.6M. Two share the registry number of Northgate Technology B.V., five are subsidiaries in the parent's ownership filing, and the card merchant “NGT*ONLINE STORE” settles into the same bank account. She can approve the family, split a record out, or ask the normaliser before she does.
The normaliser never merges below 90% confidence, and never merges two registry numbers without a parent link — Arden Digital Media Ltd is held for her: the name is close, but it is a marketing agency with its own registry number and no ownership link.
RL-118 reads “Lines that name the Aster 14 notebook belong in End-user devices › Laptops”: 1,284 lines, $5.3M, replacing four codings, with the classifier's reason — buyers at nine plants raised free-text requests or paid by card, and the words on the line, the supplier and the unit price all point to a laptop. The validator: 60 of 60 agree. An accepted rule applies to every line, past and future, with a reason code.
RL-121 is held: 37 of 40 agree (92.5%), below the 95% target — two printers and a toner recycling service would have moved.
The validator classifies its sample blind, before it sees the classifier's answer: 1,174 of 1,200 lines agree across all spend (97.8%), 93 of 96 in IT hardware. Where they differ, both make their case. On D-07, an extended warranty for 85 Aster 14 units at $9,350.00, the classifier says End-user devices › Laptops — bought on the same order and priced per unit — and the validator says IT services › Hardware maintenance, because a warranty is a service contract.
She picks “Classifier is right”, “Validator is right” or “Split the line”; the decision is logged and the agents learn from it.
The group bought 6,308 Corvel Aster 14 G5 notebooks in twelve months from six resellers. Northgate, the contracted reseller, charged $1,128; the others charged up to $1,412. The agent shows the price by reseller and the arithmetic — 1,180 × $68 + 1,020 × $116 + 860 × $161 + 610 × $219 + 498 × $284. Before it counted, the contract matcher checked clause 4.2, that the contract is in force to 31 Mar 2027, the item code on all 1,284 lines and Northgate's catalogue, lead time 5 days.
“Not counted: clause 4.2 gives $1,104 above 5,000 units a year; negotiating it on all 6,308 units would add $151.4k.”
She approves OP-0412 with OA-01, “Saving confirmed against contract”, signed with her name and role. The savings tracker freezes the baseline at $7,727,466 for 6,308 units, and the plan the agent drafted becomes ready: route Aster 14 requests to the Northgate catalogue, a notice to 23 buyers at 9 plants, a card rule for IT resellers above $600, and a letter asking Northgate to confirm the price hold to 31 Mar 2027.
The notice and the letter wait for her: the agents never send anything to a supplier or a buyer.
Across the group the agents have found $33.45M; $24.27M is approved and $4.76M realised this year. The tracker measures realised savings from new invoices against each frozen baseline, price effect only, and Ruben validates or returns each one with a reason. $2.67M is banked so far; $2.09M waits for him. Ruben will validate OP-0412 from November's invoices.
“A saving counts as banked only when finance validates it against the baseline frozen on the day it was approved.”
The same refresh, seen by five of the people who carry it — what their month looked like, and what it looks like now.
Seven agents load, merge, classify, check and size savings; the category manager signs every rule and saving, and finance signs what is banked.
Pulls lines from six spend sources, converts them to US dollars at treasury month-end rates, removes intercompany lines and reconciles each source to its ledger control total.
Merges vendor records into parent companies by tax ID, registry number, ownership filings, bank account and name, and explains each merge with its evidence.
Reads each line's words, supplier, ledger account, material group and card code, places it in the four-level category tree with a reason in words, and groups patterns into rules.
Draws 1,200 lines weighted by value, classifies them blind, measures accuracy on each rung of the ladder and each rule, holds rules below the 95% target and argues each disagreement.
Matches lines to contracts and price lists, marks each on or off contract, and checks every saving claimed against its clause and price.
Looks for consolidation, off-contract buying, price variance and tail spend, sizes each saving with its method, and drafts the plan, the buyer notice and the category brief.
Freezes the baseline when a saving is approved, measures realised savings from new invoices every month and routes them to finance.
Signs rules and savings; finance signs what is banked. The agents propose; a named person decides.
Ask about any number, supplier, rule or saving in plain words. The agent that did the work answers, cites every figure, moves the ladder to the rung it means and offers the next step.
Why did laptops move out of office supplies?
1,284 lines name the Aster 14 G5 notebook but were coded to Office supplies (612 lines), Repairs & maintenance (301), Sundry (214) and card “electronics stores” (157). A notebook is a device, so I propose moving all $5.31M to End-user devices › Laptops under rule RL-118. The validator checked 60 of them blind and agreed on all 60. Nothing moves until you accept the rule.
Who sells us the Aster 14 G5?
Six resellers sold you 6,308 units in twelve months. Northgate, the contracted reseller, charged $1,128; the other five charged $1,196 to $1,412. Buying all 6,308 at the contract price saves $612,042 a year. The contract matcher confirmed the price and that the contract runs to 31 Mar 2027.
Show off-contract buying in IT hardware
In IT hardware, $14.6M of $118.4M went around a contract that exists: free-text requests $7.9M, cards $3.4M, and purchase orders to a supplier without a contract $3.3M. The laptops are the largest single item: $5.31M across 9 plants and 23 buyers.
How sure are you about IT hardware?
I classified 96 IT hardware lines blind, weighted by value, and agreed with the classifier on 93 (96.9%), above the 95% target. Across all spend it is 1,174 of 1,200, 97.8%. The three IT hardware disagreements wait for you to settle.
13 screens from the working solution, on its sample data. Pick one to see it large.
The tree, the confidence and accuracy thresholds and who may approve what are settings, owned by the systems administrator; every change is saved as a version and logged.
| Setting | Default | Choose from |
|---|---|---|
| Category tree | 4 levels · 212 sub-categories · 864 items | Your own tree, an owner per category, optional mapping to standard codes (UNSPSC v26); add a sub-category, save as a version |
| Apply a classification at | 92% | Any confidence level, saved as a version |
| Send a line to a person below | 80% | Any confidence level, saved as a version |
| Accuracy target (hold rules below) | 95% | Any measured accuracy, saved as a version |
| Lines sampled each refresh | 1,200, value-weighted | Sample size, weighted by value |
| Merge suppliers at | 90% | Any confidence level, saved as a version |
| Flag price variance · smallest saving shown · tail suppliers | 5% · $50k a year · under $25k a year | Each set separately, saved as a version |
| Who approves | Rules: category owner (spend analyst under $250k) · Savings: category manager to $5M, CPO above · Realised savings: finance | Eight roles, each with its own rights to see, accept rules, approve merges and savings, validate, change settings, export and comment |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample spend data · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (a supplier-data vendor — 10–20% duplicate vendor records before a cleanse, a vendor figure; The Hackett Group 2025, via Procurement Magazine — 60% less savings leakage for world-class teams that cut off-contract buying). People, companies, suppliers, products and amounts named on this page are fictional — characters and sample data in the working solution.
Spend Analytics is spend analytics software that classifies and cleans all spend by category and supplier to find savings for procurement. Seven agents pull every invoice, card and travel line and reconcile each source to the ledger, merge duplicate vendor records into parent companies, place every line in your category tree with a reason in words, and pin every saving to the invoice lines behind it. Category owners sign each rule and saving, and finance signs what is banked.
Accuracy is measured, not self-reported. Every refresh a validator agent classifies a value-weighted sample blind — 1,200 lines in the demo — and measures agreement on every rung of the spend ladder and every rule: 97.8% across all spend and 96.9% in IT hardware in the demo. A rule below the target, 95% by default, is held with the lines it got wrong.
The supplier normaliser merges vendor records into parent companies by tax ID, registry number, ownership filings, bank account and name, and shows the evidence for each record. It never merges below 90% confidence or two registry numbers without a parent link. A person approves each family, splits a record out or keeps it separate, with a reason.
Yes. Instead of correcting lines one by one, the classifier proposes a rule in words with its reach and the validator's sample. Once the category owner accepts it, the rule applies to every line, past and future, with the reason code recorded; it can be withdrawn from the audit trail.
The opportunity agent looks for consolidation, off-contract buying, price variance and tail spend, and sizes each saving with its method on its own invoice lines. The contract matcher checks every saving against its clause and price list before it counts, and anything not yet negotiated is shown as not counted. Finance validates realised savings against the baseline frozen on approval.
Yes. The contract matcher marks every line on or off contract, and the solution shows spend that went around a contract that exists by channel, plant and buyer. The opportunity agent drafts the notice to the buyers; the category manager edits it and sends it.
Yes. Nothing moves until a person accepts it: the agents propose rules, merges and savings, and named owners approve, edit or reject each with a reason code. The agents never send anything to a supplier or a buyer, and only finance can mark a saving banked. Every step is on a hash-chained audit trail.
Yes. The tree is yours — four levels in the demo, 212 sub-categories and 864 items — with an owner per category and an optional mapping to standard codes. Thresholds, the sample size and approval limits are settings too, and every change is saved as a version.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your category tree, sources, contracts and a sample of past spend — and it goes live once every quality gate has passed. We will walk you through it on twelve months of your own spend first.
Hidden costs priced before award, and every score backed by the page it came from.
1-min demoRFP CreationHalf the bidder questions, because every gap past bidders asked about is closed first.
1-min demoSupplier Negotiation PreparationPrice increases held to what costs support, every number in the brief cited.
1-min demoWe'll run Spend Analytics on twelve months of your own invoice, card and travel lines.