How to predict and win trustworthy procurement data using Predictive Procurement Orchestration

Most organizations believe that their procurement system centers on people, processes and technology. What’s missing in this picture? Data. The data that flows through technology and processes is then interpreted by people to make decisions in a distinct domain. Data is often an integral part of the organization that’s overlooked. Despite our best efforts to invest in systems and tools designed to capture, clean and consume data, untrustworthy data still routinely costs organizations tens of millions of dollars of value in unrecognized opportunities. As the famous “Michael Scott” quote goes: “You miss 100% of the shots you don’t take” (Wayne Gretzky), and there are few problems that cause more untaken shots by procurement teams than untrustworthy data. When a procurement team does not trust the data, it doesn’t just lead to missed opportunities, it also creates bottlenecks and miscommunications that can damage customer and supplier relationships. The simple fact is that when procurement teams don’t have confidence in their data — whether that data originates from suppliers or stakeholders — it becomes much harder to make value-creating decisions efficiently and accurately.

The root cause of untrustworthy procurement data is both obvious and ubiquitous: free-text data entry checked using a laborious, manual, error-prone validation process. In fact, most of the procurement leaders interviewed about the topic of data quality for this article stated the data in their supplier quotes received no validation step at all prior to analysis, meaning that costly errors were only identified painfully late in the process.

The new Spend Matters white paper “Predictive Procurement Orchestration is the key to trustworthy procurement data” explores:

• What data challenges procurement practitioners must overcome to create value
• Why traditional methods of managing data using manual validation are harmful
• How Arkestro solves the data problem via Predictive Procurement Orchestration, with a spotlight on free-text Purchase Requisitions and supplier quotes

Procurement’s data problem

Data is at the heart of procurement work. Procurement professionals need to know who’s buying what, who they’re buying it from, what’s being purchased and if it conforms to existing agreements. We need accurate, consistent and dependable data to perform our day-to-day tasks effectively. In other words, we need data we can trust.

What exactly makes data trustworthy? The key litmus test between trustworthy and untrustworthy data is the reliability and speed of the data-validation process to spot and fix errors. Today, most procurement teams perform data validation using a manual validation process — eyeballs and fingers. Even when validated third-party data feeds are used, procurement practitioners often must pull up this data in a separate browser tab and then visually compare their own data with the third-party data to attain validation. This manual data-validation process is not only time-consuming, but it is inevitably prone to human bias and error itself. If you can’t trust the validation process, how are you supposed to trust the data that passes through it?

Data quality is often judged on six main performance criteria:

• Availability: Does the organization have data to begin with?
• Validity: Are the data values acceptable?
• Consistency: Are the values the same across the board regardless of their location? And are they collected in a consistent way?
• Integrity: Are the relationships between the data elements and their respective data sets accurate?
• Accuracy: Is the data a truthful representation of the objects it is supposed to model?
• Relevance: Does the data in question support the objective?

Procurement data seldom meets all six criteria. Procurement teams reliant on manual data validation workflows are perennially unsatisfied by their ability to avoid costly errors. This is a massive problem for any Procurement Analyst, Procurement Enablement, Procurement Excellence and Procurement Operations function, and it puts hard caps on the overall productivity and value attainment of the procurement team. Spend Matters routinely finds that data quality and availability are the two most costly hindrances to optimal performance. In fact, Arkestro’s forthcoming 2022 State of Procurement Data 2022 Report shows that across companies with more than 1,000 suppliers, over 8.25% of quoted prices contained an error that impacted customer’s costs (and, by extension, procurement’s savings). In some cases, these errors are due to Unit of Measure (UOM) misreads or obvious typographical or copy-and-paste errors.

The impact of untrustworthy data isn’t just confined to the cost-savings KPI performance for procurement practitioners: it has far-reaching consequences that can become quite costly for companies. In fact, Gartner’s Research from back in 2018 showed that most organizations believe that poor data quality was responsible for at least $15 million a year in hard financial losses.

Additionally, data volume increases brought on by the digitization of procurement and related business processes further intensifies the problem. The volume of procurement-relevant data created, captured, copied and consumed is growing exponentially considering:

• Supplier data: Procurement teams not only want to know the address, phone number and the point of contact email for their suppliers. Now, supplier data often includes information like financial, ESG and diversity certifications.
• Spend data: Procurement teams surface continuous classification of transactions in specific spend categories and are often held accountable for metadata reports like Purchase Price Variance, Top 20 Suppliers and KPIs like Spend Under Management by category, geography and P&L.
• Process data: Procurement teams must report on metadata about their own processes, especially in a decentralized procurement environment. This includes things like cycle time, “first time right” attainment, exception counts and the productivity and performance of all procurement processes.

It’s not just the volume of the information, either. It’s the volatile nature of the data. It’s constantly evolving, and its “use-by-date” expires quickly. This is why applying machine learning and big data to the validation task is so critical. We need to move from a single-threaded queue of one-at-a-time data validation to a continuous monitoring paradigm where many data-validation tasks can be performed simultaneously across a whole portfolio of fast-moving, complex and option-constrained processes.

A new approach to managing procurement data is needed

Traditional methods of managing procurement data have never worked very well because procurement data includes both internal data (from stakeholders and ERP systems) and external data (from suppliers), creating translational challenges across item and service codes, descriptions and other numerical identifiers.

In fact, traditional manual methods for validating and improving procurement data quality apply only to a small subset of the data that most procurement teams need in order to be effective. Here are a few of the drivers for endemic and persistent data-quality issues in procurement:

• Data entry tasks are often fragmented, making the data prone to errors. With numerous internal and external stakeholders involved in creating and curating procurement data, the result is often a high variance on data quality. This is especially true for commodity and GL code classifications.
• Data entry and data consumption are often siloed, blocking validation. Siloes between systems and processes make cross-referencing the same value in multiple tables super challenging. When individuals who don’t work in procurement enter in data (e.g., in a free-text Purchase Requisition) they aren’t motivated to follow processes and guidelines regarding data entry.
• Data quality controls that rely on human validation are not scalable. Instead, manual validation makes the number of hours in a day a hard cap on procurement data quality. With the amount of procurement data growing exponentially, the number of errors and bottlenecks grow right along with it, rendering the traditional approach to mastering data harmful.

This makes the procurement data problem a chicken-and-egg situation. We know there’s incredible opportunity to be had if we can analyze every transaction across the procurement spend ecosystem, but we cannot do so if there isn’t trust in the quality and ease of access of the data we’re working with. The most productive approach is to address the barriers to data quality head-on and commit to a Six Sigma-like continuous improvement process to improve data trustworthiness. The best products are iteratively built when we don’t let ‘perfect’ be the enemy of good.

Predictive procurement orchestration is the key to trustworthy data

The reality is that procurement is moving from a world of backwards-looking static reports to a world of continuous always-on data feeds that update in real time without human intervention. As more enterprise data becomes subject to continuous updates, this creates the opportunity to move from an intermittent “data-cleaning” approach to data quality to a continuous improvement approach to “live” data quality. Here’s where:

• Predictive procurement embeds “always-on” validation loops using external sources to check internal records and generates prioritized lists of exceptions. Validation loops run continuously to cross-reference data elements between tables in disparate systems. Arkestro predicts and checks the value using external data sources and continuously delivers recommendations to fix.
• Predictive procurement embeds version control and anomaly detection. Arkestro embeds version control within the existing solutions and processes, showing trends, spikes and outliers in a particular value over time. By identifying and ranking exceptions, Arkestro enables real-time anomaly detection before an incorrect purchase or quote is approved and impacts the P&L.

Predictive Procurement Orchestration (PPO) leverages technology to do what traditional methods of procurement data management have so far been unable to do.

PPO is characterized by three main features:

1. A no-login, seamless user experience across multiple systems: Users access data via an embedded agent designed to orchestrate data that applies to the same entity, but lives siloed and unsynced across multiple systems. End users are not required to log in to an application for the agent to run.
2. Pre-embedded and always-on data validation: An embedded platform uses live data to predict inputs and optimal attributes before any process begins. Live data is extracted from email-only or text-only workflows and is updated automatically and checked for accuracy, field-mapping and expiration.
3. Alerts, notifications and delegated actions-by-default: An embedded platform, operating without human intervention, self-triggers recommendations to human users and then learns from the acceptance rate of those recommendations, thereby improving that acceptance rate over time.

The whole premise of Predictive Procurement Orchestration is to change to a new approach to data quality: a self-improving and self-healing data process that applies an “always-on” approach to procurement data monitoring. PPO enables organizations to leverage the power of predictive models for real-time error handling by simulating an action (such as a data entry) before it happens. The same methodology can be used to rank exceptions and anomalies to improve a process in real time as the process runs.

Read more here to discover why Arkestro was named Value Leader in Spring 2022 Spend Matters SolutionMap for Sourcing.