Spend Services — User Guide

Spend Services reviews the data you already export from your finance and procurement systems and shows you money to recover, waste to cut, records to clean up, and invoices worth investigating. This guide explains every module in plain language: what it does, how each number is worked out, and the exact steps to use it. No technical knowledge needed.

Start here

Getting started

The left menu of the application lists eight modules. Each one answers a different question about your spend. Every module works the same way, so once you learn one you know them all.

Load sample data — fills the module with a realistic 1,000–5,000 row example so you can see it work instantly. Nothing to prepare.
Upload your CSV — runs the module on your own data. A CSV is the plain spreadsheet file you export from your system (in Excel, choose Save As → CSV). Each module tells you which export it expects.

After it runs you get a big headline number, four summary boxes, an optional signal breakdown, and a table of every individual finding you can export back to CSV.

Your data stays in your browser. The analysis runs on your own device — files are not sent anywhere.

Reading your results

Two colours tell you how solid a number is:

Colour	Meaning
Green	Recoverable / hard dollars. Money you can realistically get back or stop spending — duplicate payments, overbilling, unused software seats.
Amber	Estimated opportunity, or items to act on. Value that needs a decision or investigation — renegotiations, consolidations, data fixes, and fraud flags.

The Dashboard at the top of the menu adds up only the green and estimated-savings modules into one running total. Data Quality and Fraud Signals are kept separate on purpose — they lead to corrections and investigations, not to a savings figure, so mixing them in would overstate the number.

Agents

Sourcing Agent

The Sourcing Agent runs a sourcing event from first request to a recommended award, right inside the ProcBot chat. There is no separate screen. Open Sourcing Agent in the left menu, or just type a request.

Run it

Start an event by typing, for example, start a sourcing event for IT Hardware worth $400K needed in 3 weeks.
The agent reads the value and sets the procurement method (micro-purchase, three quotes, or a sealed bid), picks the RFx type, sets the delegation-of-authority approver, and shortlists pre-qualified vendors.

Work the event

An interactive panel runs six stages: Intake, Approval, Issued, Bidding, Evaluation, Award. One button moves the event to the next stage, and a vendor status board shows where each supplier stands. At Evaluation you get a bid-leveling grid (the RFX Analyser) that scores every bid on weighted technical and commercial criteria. At Award the agent names the recommended supplier and can draft a contract for the winner.

Documents and auto-pilot

Generate the RFP or RFQ and the award memo as Word files, built in the browser.
Agent auto-pilot runs the whole event on its own, with a live activity log.

Other prompts: shortlist vendors for Marketing, show similar RFX runs, status of SRC-XXXXX, list my sourcing events.

Access to the Sourcing Agent is controlled in Admin → Access control, per user, like every other module.

Agents

Contract Agent

The Contract Agent takes a contract from draft to execution inside the ProcBot chat, and keeps the key information, clauses, risks and obligations in front of you the whole way. Open Contract Agent in the left menu, or type a request.

Run it

Create a draft by typing, for example, draft a contract with Acme Corp for IT Hardware, $400K, 2 years. You can also carry over a sourcing award with draft the contract for SRC-XXXXX.
The agent fills the key information (value, start and end dates, term, payment terms, auto-renewal and notice) and lays out the clauses, a risk analysis with a mitigation plan, and the key obligations.

Work the contract

An interactive panel runs five stages: Intake, Review, Approvals, Signature, Executed. At Approvals, three departments sign off in parallel: Legal, Information Security and Risk. Each can approve or return, and all three must approve before the contract moves to signature.

Final contract and auto-pilot

Generate the final contract, with signature blocks for the Organization and the Vendor, as both a Word and a PDF file. Both are generated in the browser, with no internet needed.
Agent auto-pilot runs the whole lifecycle, including the three sign-offs.

Other prompts: list my contract drafts, status of CON-XXXXX.

The clauses, risks and figures are sample content for review. Have Legal, Information Security and Risk review before signing.

Agents

Savings Agent

The Savings Agent captures, validates and tracks procurement savings, and shows how you are tracking against target, all inside the ProcBot chat. Open Savings Agent in the left menu, or type a request.

Run it

Log a saving by typing, for example, log a saving of $150K from renegotiating the Verizon telecom contract in IT.
The agent sets the savings type (hard, soft, cost avoidance, project, automation, contract), the source or lever, the department and supplier, and computes the saving as baseline minus actual, plus any rebate or early-payment bonus.

Validate and approve

The panel runs four stages: Identified, Submitted, Approved, Realized. You can approve or reject at Submitted. Only approved and realized savings count toward the annual target. The panel shows the full calculation (baseline, actual, price savings, bonus, total, savings percent, variance to plan) and the saving's contribution to the target.

Dashboard and reports

Ask for the savings dashboard to see realized savings, attainment versus target, ROI, pipeline, and a breakdown by type and by department.
Generate a savings validation memo as a Word file.

Other prompts: savings by department, savings by type, savings register, status of SVG-XXXXX.

Access to the Savings Agent is controlled in Admin → Access control, per user. Figures shown are sample data for review; confirm evidence before booking a saving.

Savings module 1

Duplicate Payments

What it is. Finds the same bill paid more than once, and early-payment discounts you were entitled to but missed. This is the fastest money to recover because it is a clear, hard dollar amount.

How the numbers are calculated

Duplicate payments. The module groups invoices that share the same vendor, the same amount, and the same invoice number after tidying it up — spaces, dashes and symbols are removed and leading zeros dropped, so INV-0045 and 45 are treated as the same number. If two or more invoices fall in the same group, the first is the real one and every extra copy is flagged as a duplicate payment.

Recoverable (duplicates) = sum of the amounts of every extra copy

Missed early-pay discount. Payment terms like 2/10 Net 30 mean "take 2% off if you pay within 10 days, otherwise the full amount is due in 30." If the invoice was paid after the discount window, the discount was lost.

Missed discount = discount % × invoice amount (only if paid late)

Headline = duplicate payments + missed discounts

Worked example

Vendor A's $8,000 invoice "INV-1023" appears twice → the second copy is a $8,000 duplicate. A $10,000 invoice on 2/10 Net 30 paid on day 25 → missed discount of 2% = $200. Recoverable so far: $8,200.

Step by step

Open Duplicate Payments from the left menu.
Click "Load sample data" to see a live example, or "Upload your CSV" and choose your accounts-payable / paid-invoice export.
Read the headline — the green figure is the total recoverable.
Scan the table — each row shows the vendor, invoice number, amount paid, and the recoverable amount, tagged as a duplicate or a missed discount.
Click "Export findings" to send the list to your AP team or recovery vendor.

Example screen — Duplicate Payments

Recoverable — identified

$466,022

1,000

Records analyzed

Duplicate payments

120

Missed discounts

$466,022

Total recoverable

Finding	Vendor	Invoice #	Recoverable
Duplicate payment	Nordic Steel Works	INV-48210	$18,400
Missed discount	Cobalt Office	INV-77341	$212

Savings module 2

Contract Leakage

What it is. Finds invoice lines where you were charged more than the price in your contract — from price creep, expired discounts, or off-contract buying.

How the numbers are calculated

For every line, the module compares the price you actually paid per unit to the price in the contract. If the paid price is higher, the difference is overbilling. That per-unit gap is multiplied by the quantity on the line.

Leakage on a line = (price paid − contract price) × quantity (only when paid > contract)

Headline = sum of leakage across all lines

Worked example

Contract price $10.00/unit, you paid $12.00/unit, quantity 500 → ($12 − $10) × 500 = $1,000 recoverable on that line.

Step by step

Open Contract Leakage.
Load sample data, or upload your invoice lines with the contract price alongside each line.
Read the green headline — total recoverable overbilling.
Sort the table by leakage to see your biggest overcharges first, then take them to those vendors.
Export findings for your category managers.

Tip: this module needs the contracted price next to each invoiced price. If your invoice export doesn't include it, join it to your price list first (your analyst can do this in Excel).

Savings module 3

Renewals & Rebates

What it is. Two things at once: contracts about to renew automatically (so you can renegotiate or cancel before they lock in), and rebates you have already earned but probably haven't claimed.

How the numbers are calculated

Auto-renewals at risk. Any contract set to auto-renew with 90 days or fewer until its renewal date. Their combined yearly value is shown as "value at risk." Because renegotiating or cancelling typically saves a portion rather than the whole, the estimated recoverable is a conservative 10% of that value.

Estimated saving = 10% × (yearly value of auto-renewals due within 90 days)

Unclaimed rebates. If a contract is rebate-eligible and your year-to-date spend has already passed the rebate threshold, you have earned a rebate that is usually sitting unclaimed.

Earned rebate = year-to-date spend × rebate rate (when spend ≥ threshold)

Headline (estimated) = 10% of at-risk value + earned rebates

Worked example

A $100,000 contract auto-renewing in 45 days → estimated saving $10,000. A contract where you've spent $120,000 against a $100,000 threshold at a 2% rate → earned rebate $2,400.

Step by step

Open Renewals & Rebates.
Load sample data, or upload your contract / CLM register export.
Check "Auto-renewals ≤90d" — these are the ones to act on now, before the notice window closes.
Check "Capturable rebates" — the green figure you can claim back from suppliers.
Export findings and prioritise by value.

The estimated saving is a planning figure, not a promise — it assumes you act on the renewal. Treat the rebate figure as the harder number.

Savings module 4

License Waste

What it is. Money wasted on software, SaaS, cloud, and server licenses that are unused, under-used, or still assigned to people who have left the organization. Licenses like these quietly auto-renew year after year for users who never touch them.

The four kinds of waste it finds

Waste type	What it means	Saving
Assigned to former staff	The licensed user has left the company or their contract ended, but the paid license is still active in their name.	Full annual cost (cancel or reassign)
Never / unused	An active employee with no logins in 90+ days (usage "None"). Paying for software nobody opens.	Full annual cost
Under-used	Low usage, or idle 31-90 days. Often sitting on a premium tier they do not need.	About 50% (downgrade to a cheaper tier)
Auto-renewing waste	Any of the above set to auto-renew within 60 days - money about to be spent again unless you act.	Reclaim before the renewal date

License types covered

Each license is tagged Desktop (e.g. Photoshop, AutoCAD), SaaS (e.g. Salesforce, Zoom), Cloud (e.g. Microsoft 365, GitHub Enterprise) or Server (e.g. SQL Server, VMware, Oracle). The results break recoverable savings down by each type, so you can see exactly where the money is.

How the numbers are calculated

Recoverable = annual cost of every license assigned to former staff or never used

Under-used saving (estimate) = 50% x annual cost of each low-usage license

Headline = recoverable, also split by Desktop / SaaS / Cloud / Server

Worked example

A $1,200 Salesforce SaaS license still assigned to someone who left the company gives $1,200 recoverable. A $2,400 premium desktop license used only lightly saves about $1,200 by downgrading. A server license idle for four months and auto-renewing in three weeks should be reclaimed before it renews.

Step by step

Open License Waste.
Load sample data, or upload your license inventory - one row per assigned license, with the user status and last-used date.
Read the green headline and the "Recoverable by license type" panel.
Start with "Assigned to former staff" - the clearest waste to cancel today.
Reclaim auto-renewing licenses before their renewal date so you are not billed again.
Export findings for your IT / software asset management team to action.

Example screen - License Waste

Recoverable - identified

$789,366

2,000

Licenses analyzed

170

Former staff

381

Never / unused

$789,366

Recoverable

Waste type	Software	License	Reclaimable
Former staff	Salesforce	SaaS	$1,650
Never / unused	SQL Server Enterprise	Server	$7,400

Savings module 5

Tail Spend

What it is. The scattered, low-value spend spread across many small suppliers, plus buying done without a purchase order ("maverick" spend). Both are ripe for consolidating and bringing under contract.

How the numbers are calculated

The tail. Suppliers are sorted from smallest to largest. The smallest ones that together add up to the bottom 20% of total spend are the fragmented "tail."

Maverick spend. Any transaction not backed by a purchase order.

Estimated opportunity = 12% × tail spend + 8% × maverick spend

The percentages are conservative planning assumptions for what consolidation and putting spend under contract typically save.

Worked example

Tail spend of $250,000 → $30,000. Off-PO spend of $500,000 → $40,000. Estimated opportunity ≈ $70,000.

Step by step

Open Tail Spend.
Load sample data, or upload your spend / general-ledger transaction export.
Read "Off-PO (maverick) spend" and "Fragmented tail vendors" to size the problem.
Use the table to pick vendors to consolidate.
Export findings for a sourcing plan.

This is an estimated opportunity (amber): the money is real but only lands once you act — consolidate suppliers or move maverick spend onto contracts.

Control module 6

Data Quality

What it is. A health check of your vendor list (the "master data"). Bad vendor records cause payment mistakes and open the loopholes that fraud slips through, so cleaning them up protects both accuracy and security.

What it checks (11 rules)

Check	What it means in plain terms
Duplicate vendor master	Two records for the same company (same tax ID) — risks paying the same supplier twice or hiding activity.
Missing tax ID	No tax identification number on file.
Malformed tax ID	A tax ID that doesn't match the valid format.
Missing / invalid VAT	No usable VAT number where one is expected.
Missing UNSPSC	No standard product/service classification code — hides what you actually buy.
Missing NAICS	No standard industry code for the supplier.
Non-verifiable address	A PO Box or residential address instead of a real business premises.
Invalid email	The contact email isn't a real address format.
Missing bank details	No bank account on file to pay against.
Dormant vendor	No transaction in over 12 months — an unused profile that could be misused.
Unverified bank change	Bank details were changed without a second, out-of-band (2FA) verification — a classic fraud entry point.

How the score is calculated

The headline is a count of issues found, not a dollar figure — these are things to fix. The health score tells you how clean the list is overall.

Health score = 100% − (records with at least one issue ÷ total records × 100)

Worked example

If 700 of 3,140 vendor records have one or more issues, the health score is about 78%.

Step by step

Open Data Quality.
Load sample data, or upload your vendor / supplier master export.
Read the health score and the "Records with issues" box.
Use the breakdown chips to see which problems are most common.
Work the table — each row names the record, the field, and the exact problem. Fix the duplicates and unverified bank changes first.
Export findings as a clean-up worklist for your master-data team.

Control module 7

Fraud Signals

What it is. Screens every invoice for patterns that often point to fraud, ranked by severity. Think of it as a smoke detector: it raises flags for a person to review — it does not accuse anyone or confirm fraud on its own.

The signals it looks for

Signal	Plain-English meaning
Ghost vendor	A supplier whose bank, address, or phone matches an employee's — a sign of a fake vendor set up by insiders. (Needs the employee file, below.)
Bank redirection (BEC)	Bank details changed within 14 days of a payment without 2FA, and/or the bank country doesn't match the tax country — the fingerprint of a payment-diversion scam.
Bank shared across vendors	The same bank account used by several "different" suppliers.
Phantom goods / short receipt	Invoiced for more than was actually received, with the 3-way match manually overridden.
Split purchase / smurfing	One big buy chopped into several smaller invoices to stay under an approval limit — same vendor, approver, and day.
Just below approval limit	An invoice priced suspiciously close to, but under, the approval threshold.
Duplicate / double-dip	Same vendor, amount, and date billed twice — often across invoice-number variants or paid by both invoice and card.
Rubber-stamp approval	Approved in under 5 seconds — too fast for a real review.
Off-hours / weekend entry	Entered late at night or on a weekend, when oversight is low.
Brand-new vendor	An invoice on a supplier created less than 7 days ago.
Round-dollar amount	Suspiciously round figures (e.g. exactly $10,000) that rarely occur naturally.

Benford's Law — the number-pattern test

In genuine spend, far more amounts start with the digit 1 than with 9 — a natural mathematical pattern called Benford's Law. When someone invents numbers, the pattern breaks. The chart shows the actual first-digit spread against the expected one and flags the digit that is most over-represented. A big gap is a reason to look closer, not proof of fraud.

How the totals are calculated

Spend under review = combined value of the flagged transactions

This is exposure to investigate, not confirmed loss. Findings are sorted by severity, so the most serious signals (ghost vendor, bank redirection, shared bank, phantom goods) sit at the top of the table.

Turning on the employee cross-match (for ghost vendors)

The ghost-vendor check compares supplier details to your staff records. On the module page there is an optional "Upload HR file" button — add your employee master export and the check switches on. The sample data already includes matching records, so the demo shows it working without any upload.

Step by step

Open Fraud Signals.
Load sample data, or upload your invoice / payment activity export.
(Optional) Upload your HR file to switch on ghost-vendor detection.
Read "High-severity flags" — start your review there.
Check the Benford chart for an over-represented digit.
Work the table top-down — each row explains why it was flagged. Investigate; don't accuse.
Export findings as a case list for your audit or controls team.

Example screen — Fraud Signals

Flagged spend under review

$3.1M

5,270

Invoices analyzed

1,480

Flagged invoices

472

High-severity flags

$3.1M

Spend under review

Signal	Vendor	Why flagged
Ghost vendor	Delta Components	vendor bank = HR bank acct
Bank redirection	Beacon Telecom	changed 4d before pay, no 2FA

Fraud Signals flags patterns worth reviewing, not confirmed fraud. Every flag needs a human to investigate and clear or escalate it.

Control module 8

Compliance Audit

What it is. An automated audit of the whole source-to-pay lifecycle against your policies, delegation of authority (DOA), and contract terms. It scores how compliant your spend is and lists every breach, ranked by severity, for your audit team. It is designed to complement Fraud Signals and Data Quality, not repeat them.

What it checks, by lifecycle stage

Stage	Violation it flags	Policy
Sourcing	Purchase above the quote threshold with fewer than 3 quotes; sole-source buy with no approval on file.	Sourcing thresholds; Competition policy
Contracting	After-the-fact contract (signed after work started); unauthorized signatory (value above the signer's DOA limit); missing mandatory clause (IP, SLA, liability).	DOA; Legal & Contracting
Change orders	Cumulative amendments above 20% of the original contract value.	Contract Amendment policy
P2P	Confirming order (PO raised after the invoice date); DOA approval breach (PO value above the approver's limit).	PO approval thresholds; No PO, No Pay
Vendor	Expired insurance certificate (COI) or lapsed ISO / SOC 2 certification.	Third-Party Risk Management

Severity: what happens with a finding

Each violation is ranked so the right ones get stopped first:

Sev 1 (high) - sole-source without approval, after-the-fact contract, unauthorized signatory, DOA approval breach. Action: block payment and escalate to internal audit.
Sev 2 (medium) - missing quotes, missing clause, confirming order, value creep, expired certificate. Action: route for manager ex-post approval.

How the score is calculated

Compliance rate = 100% - (non-compliant records / total records x 100)

Spend under review = total value of the transactions that failed a check

Worked example

A $250,000 purchase with only 1 quote on file breaches the 3-quote rule (Sev 2). A $900,000 contract signed by a Category Manager whose DOA limit is $100,000 is an unauthorized signatory (Sev 1) - block and escalate.

Not duplicated here

Transactional fraud signals (split POs, bank redirection, phantom goods) live in Fraud Signals; duplicate, dormant, and unverified-bank-change vendor records live in Data Quality; off-PO maverick spend lives in Tail Spend. Compliance Audit focuses on policy, DOA, and contract-term adherence.

Step by step

Open Compliance Audit.
Load sample data, or upload your S2P transaction log joined to contract, DOA, and certificate data.
Read the compliance rate and the Sev 1 (high-severity) count.
Use the breakdown to see which policies are breached most often.
Work Sev 1 findings first - these are payment-blocking.
Export findings as an audit worklist, each mapped to its policy.

Example screen - Compliance Audit

Policy violations found

419

2,200

Transactions analyzed

418

Non-compliant records

170

High severity (Sev 1)

81%

Compliance rate

Severity	Violation	Policy
Sev 1	Unauthorized signatory	Delegation of Authority
Sev 2	Confirming order	No PO, No Pay

Glossary

Recoverable	Hard dollars you can realistically get back or stop spending.
Estimated opportunity	Value that depends on you taking an action (renegotiating, consolidating).
PO (purchase order)	An approved order raised before buying. Spend without one is "maverick."
Maverick spend	Buying done outside the proper process / without a PO.
Master data	Your core reference lists — here, the vendor master (list of suppliers).
UNSPSC / NAICS	Standard codes that classify what you buy and what industry a supplier is in.
VAT / Tax ID	Government identifiers used to confirm a supplier is a real, registered business.
2FA / out-of-band	A second, separate confirmation (e.g. a phone call) before a sensitive change like new bank details.
3-way match	Checking the purchase order, the goods received, and the invoice all agree before paying.
Approval threshold	The dollar limit above which a purchase needs higher sign-off.
Ghost / shell vendor	A fake supplier, often set up by an insider to send false invoices.
BEC	Business Email Compromise — a scam that redirects payments by changing bank details.
Benford's Law	The natural pattern where real numbers start with 1 most often; breaks when numbers are faked.