.png){kind=logo}
Corporate finance teams are eagerly investing in the technology that promises to make procurement smarter, such as spend analytics platforms, AI-powered spend forecasting, and supplier intelligence dashboards. When executed properly, these tools can add real value – highlighting patterns in purchasing behavior, flagging anomalies before they could turn into real issues, and providing finance executives with a data-backed view of where money is going and why.
The question that rarely gets asked loudly enough here is: “Where does the underlying data come from, and how reliable is it?” For most businesses, the answer to that question lies in the Accounts Payable department – and AP data is often significantly messier than most analytic investments assume.
Spend analytics and AI forecasting tools do not produce data, but they do consume it – usually from the procurement systems, ERP platforms, or the invoice processing workflows that are positioned at the heart of the AP function. The output of any of these tools is only as good as the input they receive. This observation is nothing new on its own, but it does get systematically underweighted when it comes to evaluating an organization’s analytics infrastructure.
Invoice validation is the specific failure point worth mentioning. There are three documents that typically define a transaction in a procure-to-pay workflow: the purchase order that authorized the spend, the goods receipt confirming delivery, and the vendor invoice requesting payment.
When all three of these documents are present and accurate against each other - the transaction that enters infrastructure is classified as "clean" and becomes trustworthy data that can be passed through the spend analytics system. If, however, these documents are not consistent against each other (as they can frequently be in businesses without a formal validation process), the transaction entering the infrastructure will show what is presented on the invoice.
This particular gap is much more significant than some businesses think.
A price-per-unit that differs from the price mentioned in the contract or a PO reference for an invoice that never existed - neither of those are big issues on their own. However, once the number of such issues goes into hundreds or thousands over the span of multiple months, an entire dataset becomes drastically inaccurate, making any predictive work on top of such flawed data ultimately useless.
The downstream effect is obvious by nature, though it is seldom traced all the way back to where the problems began. Model spend categories that have been derived from unvalidated invoice data will be loaded with errors to future baselines. When a supplier has been overcharging for two quarters on a certain product category and those invoices were not validated, the outcome is taken as a new baseline.
Variance alerts won't be able to work if they don't have any clean data to compare the current info against. AI-driven recommendations for supplier consolidation, budget reallocation, or category benchmarking are also going to work with information that is incorrect to begin with, offering completely unrealistic suggestions.
This flavour of a data quality problem is the hardest to identify: not an outright mistake, readily recognizable as being incorrect, but rather an insidious, routine misrepresentation that burrows into models and comes out the other end as a recommendation that's just a tiny bit wrong. So by the time the forecast is sufficiently inaccurate to merit an investigation – the procurement data which had prompted it can be months out of date and incredibly difficult to untangle.
It is at this point that a concept commonly discussed in narrow AP and audit contexts becomes relevant for a much broader question about analytics integrity. The process in question is 3-way matching – the practice of cross-referencing the purchase order, the goods receipt note, and the vendor invoice before approving payment, working as a data validation mechanism at its core.
Every invoice that finishes a proper 3-way match is a confirmed, accurate transaction: with the right goods, at the right price, from the right vendor, and received as documented. Every invoice that doesn’t pass that check, on the other hand, is marked as an unverified assumption.
On its own, a single unverified invoice is nothing but a minor control gap. At scale, any procurement function processing invoices without systematic 3-way matching is just feeding its analytical infrastructure with a dataset filled with unverified assumptions.
Attempting to reframe 3-way matching as a data integrity mechanism instead of being purely an accounting control measure changes the way organizations approach this topic. This process can both catch overbilling and prevent duplicate payments – but it can also ensure that the transactional record of procurement activity is sufficiently accurate to support the analytical layer sitting on top of it.
Procurement intelligence tools can only be as intelligent as the data they’re working with, and 3-way matching goes a long way toward improving data quality by validating the transactional data at the point of its origin.
This sequential problem is not only rampant, but also often underestimated. Many businesses invest heavily into upgrading their analytical infrastructure, with more powerful tools, advanced models, and AI forecasting capabilities. These same businesses, however, rarely bother to ensure that the information the analysis is working with is reliable to begin with.
What makes the problem worse is that the analytical output in such cases can often seem credible while being completely unreliable at its core. It is an issue that is very difficult to resolve retroactively, considering how long it takes for these minor issues to turn into a massive problem.
As such, the only correct approach here for finance and procurement leaders is to address re-engineering of both AP processes and analytical capabilities as the same workflow. Better AP validation with strict 3-way matching enforcement on product-based purchases greatly improves data quality upstream, improving the value of any subsequent investments into analytical capabilities.
A lot of organizations that see the biggest ROI in procurement AP have at least one factor that they share - having a disciplined and validated data foundation that the analytical capabilities can safely rely on when necessary.
