Procure-to-pay control layer

Prove every payable. Don't just process it.

Your ERP records what you ordered. It cannot prove what your supplier agreed to, what actually arrived, which discrepancies were accepted, or who authorised the payment. Procure-to-Pay Platform is the control layer that holds that evidence — and explains every exception well enough to defend it to an auditor.

Built by CloudTech Crafters Limited
Cloud & software engineering since 2017 Nairobi, Kenya AWS Marketplace listing in progress

The evidence chain

PO Purchase order, version-pinned Which version was live when the supplier responded.
ACK The supplier's acknowledged commitment The link nobody else keeps. Line-level accept, partial or decline — bound to an exact order version.
DN Delivery note & receipt evidence Received, rejected and damaged quantities, batch and expiry.
GRN Official goods receipt Reconciled against what your inventory system actually recorded.
INV Invoice, adjustments, payment decision Credit and debit notes recalculate the payable. A human always decides.
The problem

The gap isn't invoice capture. It's evidence.

Most accounts-payable tools solve intake: read the invoice, route it for approval, pay it. That leaves the hard part untouched. The moment a delivery is short, a price has drifted, or an order was amended after the supplier accepted it, the process falls back to email, shared folders and someone's memory.

11%
have automated exception management
The least-automated process in AP, out of ten measured. Supplier onboarding is second-lowest at 22%.
Ardent Partners, The State of ePayables 2025
18.4%
average invoice exception rate
Against an 8.2-day average cycle time. Exceptions are where the cost and the risk concentrate.
Ardent Partners, The State of ePayables 2025
50%+
of occupational fraud cases involve a control failure or override
Median scheme runs 12 months before detection. 2,402 cases, 143 countries.
ACFE, Occupational Fraud 2026: A Report to the Nations
76%
of organisations faced payments fraud in 2025
74% were hit by business email compromise. Only 17% use AI to counter it.
AFP, 2026 Payments Fraud and Control Survey
The uncomfortable question: if a supplier disputes an invoice you paid eleven months ago, can you produce the order version they accepted, the quantities your receiving bay recorded, the variance that was waived, the rule that permitted the waiver, and the name of the person who waived it — in one place, without asking three people to search their mailboxes?
Where we are different

Three things this product does that the category does not.

We reviewed the leading procure-to-pay and AP automation products before building this. These three positions were open.

01

The supplier's commitment is evidence, not a status flag

Elsewhere, "PO acknowledged" is a checkbox, and an advance shipping notice is shipment visibility that never enters the match. Here, the supplier's line-level response — accept, partial with a committed quantity and date, or decline with a reason — is captured against an exact order version and becomes a matched input. A supplier cannot respond to a superseded version, and an order amended after acceptance cannot quietly become the thing you measure the delivery against.

02

Every exception explains itself in full

Not a reason code. Not "invoice blocked". For each dimension the case shows the expected value, the actual value, the variance, the named rule and tolerance that fired, the version of that rule, the evidence versions it was measured against, and the action required. Reviewers of AI-driven AP tools routinely complain about flags they cannot interrogate. A control you cannot explain is not a control.

03

It runs where your data has to live

Nearly every modern product in this category is multi-tenant SaaS only. If your regulator, your board or your data-protection authority requires the financial record to stay inside your own environment, most of the market simply cannot bid. Procure-to-Pay Platform ships as containers and deploys to AWS, Azure, GCP or your own data centre — same build, no customer code fork.

How matching works

What a reviewer actually sees.

Matching is configurable per document class, supplier, entity or process — two-way, three-way or four-way. When a case fails, this is the screen the reviewer opens. Every row is derived, versioned and reproducible.

Case MC-2026-0041882
Exception — quantity & price
Illustrative match result showing each dimension, its expected and actual values, variance, verdict and the rule that produced it.
Dimension Expected Actual Variance Verdict Rule & evidence
Supplier identity Kilimo Supplies Ltd Kilimo Supplies Ltd Pass IDENT-01 v3 · vendor master v18
Order version LPO-88431 v4 LPO-88431 v4 Pass VERSION-02 v1 · acknowledged 12 Aug
Line identity 18 of 18 lines mapped 18 of 18 lines mapped Pass LINE-04 v7 · confidence ≥ 0.97
Quantity — line 6 480 CTN 432 CTN −48 (−10.0%) Fail QTY-TOL-03 v2 · tolerance ±2% · GRN-55107 v1
Unit price — line 6 KES 1,240.00 KES 1,289.60 +49.60 (+4.0%) Warn PRICE-TOL-01 v5 · warn ≥ 3%, fail ≥ 7%
Tax & compliance Valid electronic tax invoice, buyer PIN present Valid, buyer PIN matched Pass TAX-KE-01 v2 · verified 03 Sep 09:14
Duplicate No prior payable for this invoice No exact or near match found Pass DUP-02 v4 · 4 channels checked
Outcome: case held in EXCEPTION. Routed to Receiving for confirmation of the 48-carton short delivery, then to Procurement for the 4% price variance. Nothing about this case can reach payment review until a named human resolves both dimensions, with a reason, against evidence versions pinned at the moment of evaluation.

Illustrative example using representative data. Dimensions, tolerances, thresholds and routing are configured per customer — they are not fixed in the product.

The controlled journey

Three truths, never overwritten by each other.

The product separates the commercial commitment (the final order and what the supplier accepted), the physical and system-confirmed receipt (delivery note and official goods receipt), and the supplier's financial claim (the invoice and any approved adjustments). The match engine compares those three truths and explains the differences. The workflow decides who has to resolve them.

It is state-led, not document-led. One service owns the lifecycle of a case and the transitions it is allowed to make. Concurrent and out-of-order events — a retry, a late amendment, an invoice that arrives before the receipt — cannot produce a decision against stale or partial evidence.

See the full platform

Order issued and acknowledged

The supplier responds line by line against a specific order version. Reminders, expiry and escalation are automatic. Where a supplier genuinely cannot self-serve, an assisted path records who responded on their behalf and on what evidence — and cannot masquerade as self-service.

Goods delivered and received

Line-level receipt with rejected and damaged quantities, batch and expiry evidence. The platform prepares and records the receiving evidence; where your inventory system remains the official record, the confirmed goods receipt is read back and reconciled.

Invoice arrives on any channel

Supplier portal, email, monitored folder, SFTP or API. Documents are classified, extracted and validated with field-level confidence, and uncertain fields route to human correction. Retries and resubmissions cannot create a second payable.

Matched, explained, routed

A clean case becomes matched and ready for payment review. Anything else becomes an explained exception with a named owner, an SLA and an escalation path.

A human approves the payment decision

Payment preparation is approved or rejected by an authorised finance user, with dual sign-off where you configure it. The platform never executes payment, and no model ever auto-approves one. Late corrections re-open the review and preserve the prior decision.

Product structure

A mandatory core, and modules you license as you need them.

The core establishes security and operability and cannot be switched off. Business modules are individually licensable and can be limited by entity, process or document type. Turning a module off preserves the historical record and its audit trail.

Core

Tenant & entitlements

Licence control, module activation, feature flags and safe rollout.

Core

Advanced IAM & policy

Users, groups, roles, resources, conditions, explicit deny and separation of duties.

Core

Organisation model

Legal entity, business unit, branch, location, currency, time zone and rule scope.

Core

Audit & event ledger

Tamper-evident, append-only history of every action, decision and evidence version.

Core

Workflow & decision engine

Versioned rule sets, human overrides, delegation, escalation and SLA.

Core

Integration runtime

Canonical data model, replaceable adapters, retries, idempotency, reconciliation, dead-letter handling.

Module

Omnichannel intake

Portal, email, monitored folder, SFTP and API, with quarantine and idempotency.

Module

Document AI

Classification, extraction, field-level confidence, human correction. Handwriting, stamp and signature detection; multi-language and multi-script; layout-agnostic extraction for unseen layouts.

Module

Supplier collaboration

Order notification, line-level response, acknowledgement, reminders and delivery notes.

Module

Receiving

Line-level receipt, rejected and damaged quantities, batch and expiry evidence.

Module

Match & decision

Configurable two-, three- and four-way matching with fully explained exceptions.

Module

Exception intelligence

Triage ranking, suggested resolutions and learned tolerance recommendations — advisory, human-approved, never self-applying.

Module

Duplicate detection

File and event idempotency plus exact, semantic and near-duplicate business controls across every channel.

Module

Fraud risk intelligence

Configurable indicators, statistical anomalies, an explainable 0–100 score, bank-detail tamper detection, graph analysis for related-party rings, and an investigation workflow.

Module

Supplier management

Onboarding, compliance, contacts, terms, maintenance, suspension and performance.

Module

Adjustments & returns

Credit notes, debit notes, returns and payable-position recalculation.

Module

Remittance advice

Payment-status intake, advice generation, dispatch, bounce, reissue and supersession.

Module

Analytics & assistive AI

Operational, executive, supplier, exception and audit reporting. Natural-language query is read-only and scope-enforced; generated summaries are labelled and human-verified.

Module

AI/ML data & model enablement

Isolated training store, lineage, evaluation, drift monitoring, rollback and anonymised export. Selectable OCR engine and governance substrate. No shared-model training on your data.

Integration

We connect to whatever you already run.

Every source system talks to the platform through the same canonical business objects and versioned contracts. Differences between one customer's ERP and another's live in a replaceable adapter, not in a fork of the product. That is the difference between an integration and a rebuild — and it is why a second customer does not cost what the first one did.

  • ERP, accounting, CRM, document management, mailbox and folder sources
  • Reconciliation, retries, idempotency and dead-letter handling as first-class behaviour
  • Source-system freshness and sync health visible to operators, not buried in logs
  • Typed OpenAPI contract; a breaking change fails the build, not your month-end
SAPSAP Business One OracleMicrosoft Dynamics 365 NetSuiteQuickBooks OdooSage XeroSalesforce SharePointSFTP & network folders IMAP / ExchangeREST & webhook APIs

Adapters are built against the same canonical contract. Systems not listed are an adapter, not a product change — bring yours to the working session.

Tax & compliance control

Clearance proves the invoice is valid for tax. It does not prove the goods arrived.

Tax authorities across Africa, the Gulf and Europe are moving to clearance, pre-clearance and five-corner e-invoicing. Those systems validate the seller's document. They say nothing about whether the quantity matches your receipt, whether the price matches what your supplier accepted, or whether you have already paid this invoice once. That remains a buyer-side control problem — and it is now a buyer-side tax exposure.

Kenya, from 1 January 2026: KRA validates income and expenses declared in income-tax returns against TIMS/eTIMS data, withholding records and customs data. An expense that is not supported by a valid electronic tax invoice — correctly transmitted with the buyer's own PIN — is disallowed. Holding a PDF is not enough. Your AP ledger has to reconcile to your KRA purchase schedule before you file.
Kenya — KRA eTIMS
Validation live 1 Jan 2026
Expenses unsupported by a valid, correctly-PIN'd electronic tax invoice are disallowed for deduction. OSCU (always-online) and VSCU (batch, offline-capable) integration paths both require KRA vetting and certification. Buyer-initiated invoicing is available where supplier turnover is at or below KES 5 million.
UAE
Wave 1 go-live Jan 2027
Five-corner Peppol model (DCTCE) under the PINT AE specification, mediated by Accredited Service Providers. Structured XML with 51 required fields for a standard tax invoice. Taxpayers above AED 50m go live January 2027; below, July 2027; B2G October 2027.
Saudi Arabia — ZATCA
Phase 2 waves ongoing
Real-time clearance of B2B invoices with cryptographic stamps and compliant XML, rolled out in waves by revenue threshold. Thresholds have descended into the low millions of riyals, so effectively the whole supplier base is in scope. Confirm your current wave and deadline directly with ZATCA.
Nigeria — NRS / FIRS
Phasing through 2027
Pre-clearance model on the Merchant Buyer Solution: B2B and B2G invoices must be cleared before delivery to the buyer, returning an IRN, cryptographic stamp and QR code. Six-year retention with signatures that must remain verifiable.
European Union — ViDA
Intra-EU DRR from Jul 2030
Mandatory structured e-invoicing and digital reporting for intra-Community transactions from July 2030, with all pre-2024 domestic regimes converging by January 2035. National mandates land sooner — France from September 2026, Belgium January 2026, Poland KSeF from February 2026, Germany issuing from 2027.
Rwanda, Uganda, Tanzania, Ghana, Zambia, Egypt
In force
Real-time or near-real-time transmission to the revenue authority, most with a certified integration requirement. Rwanda's OSDC/VSDC split mirrors Kenya's; Uganda's EFRIS expanded to twelve further sectors; Egypt operates a full clearance model.
Deployment

Same build. Your environment.

Procure-to-Pay Platform is built as modular services with hard internal domain boundaries, shipped as containers. It runs on AWS, Azure or GCP in your own account, or entirely on-premises in your data centre. There is no customer-specific fork: what differs between deployments is configuration, entitlements and adapters.

  • Dedicated on-premises — for organisations whose financial record cannot leave the building
  • Dedicated private cloud — your VPC, your account, your region, your keys
  • Multi-tenant SaaS — where you want us to run it

Tenant isolation is enforced at the database with row-level security scoped by tenant, not by application code alone — and cross-customer isolation is a tested control, verified by an adversarial negative-access suite, not an assertion in a brochure.

Deployment, security & data residency

Data residency

Saudi NCA cloud controls, UAE and Kenyan public-sector expectations, EU GDPR, and internal board policy all push financial records toward a specific jurisdiction. Choose the region, or choose the building.

No shared-model training

Your documents, corrections and outcomes are never used to train a model that serves another customer. Every dataset, queue, cache key, audit event and ML export is isolated by default.

Selectable AI engine

Self-hosted vision-language OCR by default; cloud Document AI or a classical fallback as an opt-in profile. The model registry and serving substrate are equally selectable — nothing is hard-wired into the core.

Explainability by construction

Every automated recommendation records feature provenance, rule or model version, confidence, evaluation status and the human disposition that followed it.

Who this is for

Organisations where receiving is distributed and the evidence is not.

Multiple entities, business units, branches or sites. High supplier counts. Goods that physically arrive somewhere other than where the invoice is approved. A regulator, an auditor or a board that expects the control to be demonstrable.

CFO & Financial Controller

You are accountable for a payable position you cannot fully evidence, and for a tax deduction that now depends on documents your suppliers control. You need the exception rate down and the audit answer immediate.

Head of AP & Shared Services

Volume grows, headcount does not, and roughly a fifth of your team's time goes on supplier queries. The tools you have automated the easy invoices and left you the hard ones.

Head of Procurement

You negotiated terms that AP cannot see and suppliers do not honour. You need the accepted commitment to be a record, not a conversation.

Head of Internal Audit & Risk

You are testing controls that are applied inconsistently across sites, and reconstructing context from mailboxes. You need an append-only trail with the rule, the variance and the authoriser attached.

CIO & Head of Architecture

You are being asked to add an AP tool to an estate you are already trying to rationalise, under a data-residency constraint most vendors cannot meet. You need a system-agnostic integration boundary, not another silo.

Sectors

Healthcare, retail and distribution, manufacturing, hospitality, FMCG, services and the public sector. Industry differences are configuration — entities, document classes, tolerances, tax rules and traceability fields — not code.

How to buy

Directly today. Through AWS Marketplace shortly.

Procure-to-Pay Platform is licensed by module, with implementation, configuration and adapter work scoped separately. Pricing depends on the modules you activate, your entity and site count, document volume and deployment model — so it is quoted, not published.

We are listing the platform on AWS Marketplace as a container product with a companion professional-services listing. When that is live, customers will be able to transact against their existing AWS agreement on standard contract terms — and, where an AWS Private Pricing Agreement is in place, Marketplace spend typically draws down committed spend, subject to the terms of that specific agreement.

Being straight about status: the AWS Marketplace listing is in progress, not live. We are also selecting a small number of design partners for the first deployments. If you would rather wait for a reference customer than be one, say so in the working session — we will tell you honestly where the product is.

What the working session actually is

Forty-five minutes, no slideware. Bring a real problem and we will work it.

Your evidence chain, mapped

Where orders, acknowledgements, receipts and invoices live today, and where the chain breaks.

One real exception, walked through

Take a case that cost you a week. We show what the match result would have said and who it would have gone to.

Deployment and residency constraints

What your regulator, board or DPO requires, and which deployment profile that leads to.

An honest answer on fit

Including "not yet" or "not us". A control platform is a poor fit for an organisation that mostly needs invoice capture, and we would rather say so now.

Straight answers

The questions a serious buyer asks.

Does this replace our ERP?
No, and you should be suspicious of anything that says it does. Your ERP remains the system of record for orders, inventory and the general ledger. Procure-to-Pay Platform holds the evidence and the control that sit between those systems and your suppliers, and hands back a decision-ready case. Where your inventory system owns the official goods receipt, it keeps owning it — we read it and reconcile against it.
What exactly do you mean by four-way matching?

Matching is configurable: two-way, three-way or four-way, selectable by document class, supplier, entity or process. In our four-way configuration the fourth input is the supplier's acknowledged commitment — their line-level response bound to an exact order version — alongside the purchase order, the goods receipt and the invoice.

Worth knowing: parts of the industry use "four-way" to mean a quality inspection or acceptance document as the fourth leg instead. The term is genuinely ambiguous across vendors. When you evaluate anyone, including us, ask which four documents are actually compared and which of them can block a payment.

Our suppliers will never log into a portal. Does this fall apart?
This is the single most common reason AP automation programmes fail, and pretending otherwise would be dishonest. Ardent Partners calls supplier enablement the "Achilles heel" of these deployments. Two things follow. First, we survey supplier readiness before design, not after go-live, and classify each supplier as self-service-ready, assisted, or not ready. Second, there is a deliberate assisted path: your team can record a supplier's response on their behalf, and the system marks it as assisted with the source evidence and the internal actor recorded. It cannot masquerade as a self-service response, and it cannot bypass approval. Invoices still arrive by email, folder or SFTP regardless.
Can it really run on-premises, or is that a sales answer?
It runs on-premises. The product is containerised with hard internal domain boundaries and no dependency on a hosted control plane. Tenant isolation is enforced in PostgreSQL with row-level security; a customer that contractually requires physical isolation can have a dedicated database. The trade-off is honest: on-premises means you own the infrastructure, the upgrade window and the operational monitoring, and it is not the cheapest way to run this. It is the only way for some organisations, which is why it exists.
Will an AI approve payments?
No. Autonomous approval of a payment is a permanent, deliberate non-goal of this product — not a feature we have not built yet. AI is used for classification, extraction, duplicate and fraud signals, triage ranking and suggested resolutions. Every one of those is advisory, labelled, versioned and human-disposed. A matched case still requires an authorised finance user to approve payment preparation, and the platform does not execute payment at all.
How long is implementation, and what does it depend on?
The honest answer is that it depends on things we have to look at first: how many source systems and entities are in scope, the quality of your vendor master, whether your suppliers can respond electronically, and whether a tax-authority integration needs certification (Kenya's OSCU/VSCU and Rwanda's OSDC/VSDC both do, and certification has a lead time). We scope with a paid discovery phase that produces a bottom-up estimate and a go/no-go gate before any build commitment. Anyone who quotes you a timeline before seeing your vendor master is guessing.
What happens to our data, and do you train models on it?
No shared-model training on customer data, by default and by design. Every dataset, document, queue, cache key, audit event and ML export is isolated per customer. Where you enable the AI/ML module, the training store is yours, the lineage is recorded, and any export is anonymised and access-controlled. The OCR engine and the model registry are both selectable per customer, including fully self-hosted options, so no document has to leave your environment.
You have no public reference customers. Why should we take the risk?
Because we are telling you that up front rather than putting invented logos on this page. CloudTech Crafters is an established cloud and software engineering firm delivering into regulated sectors from Nairobi — that is the track record you can check. The P2P product itself is early, and the first deployments are design-partner engagements with the commercial terms that implies. If you need a reference customer in your own industry before you can move, we would rather you told us that in the first conversation than in month four.
Next step

Bring one exception that cost you a week.

Forty-five minutes. We map your evidence chain, walk a real case through the match engine, and give you a straight answer on whether this fits — including if it does not.

Sources for figures on this page

Exception management, supplier onboarding and exception-rate figures: Ardent Partners, The State of ePayables 2025. Occupational fraud: ACFE, Occupational Fraud 2026: A Report to the Nations. Payments fraud: Association for Financial Professionals, 2026 Payments Fraud and Control Survey. Kenya expense validation: Kenya Revenue Authority, Validation of Income and Expenses in the Income Tax Returns. Regulatory dates are summarised in good faith from public sources and change frequently — confirm your own obligations with the relevant authority or your tax adviser. Figures cited are third-party research about the market and are not performance claims for Procure-to-Pay Platform.