It seems impossible to escape the words artificial intelligence in discussions about procurement technology. Hopeful innovators see it as a way of blowing the doors off the value their teams can provide. Procurement tech companies are putting the label on everything they market. Executives are handing down “AI incorporation” mandates that could mean almost anything.
But in the middle of the buzz, there are several long-proven uses for AI in Procurement. Here, we’ve explored how a specific one—machine learning in spend analytics—and how teams can use it to reimagine the ways they relate to and use their data.
What is spend analytics?
Spend analytics is software that centralizes an organization’s spend data and provides insights into where the business is spending, how much of that spending is under control, and where the business has opportunities to lower costs.
What is machine learning?
In technology, AI is an umbrella term that encompasses advanced computing processes that do work that would normally require humans. It’s a broad term, but there are several specific processes that fall underneath it. Machine learning (ML) is one of them. At a high level, it’s a set of algorithms that allow an AI model to learn and progressively improve its pattern recognition abilities, similar to how humans learn.
Role of machine learning in spend analytics
Procurement deals with a tremendous amount of data that comes from a mix of different ERPs, suites, and P-card solutions. Because of the volume and complexity of this data, assembling it into any analysis-ready form manually is practically impossible.
Machine learning, especially its ability to learn patterns, provides a major leap forward in how Procurement can analyze spend and use the findings to create organizational value. Below, we’ve briefly described the most important benefits that machine learning in spend analytics can provide.
Data cleaning
Procurement’s data isn’t just massive; it’s also messy. Even if a team could consolidate its files, the result would be a conglomeration of different formats, vendor naming conventions, and missing information. But because machine learning becomes more specialized and capable with each set of information it processes, it’s the perfect way to clean this mess and turn it into an asset.
Normalizing vendor names is a perfect example. A machine learning model can recognize vendor naming conventions that go together, like Fed-Ex, FedEx, and Federal Express, and suggest a consolidated version: Fed-Ex. In just a few minutes, it can turn a jumble of spreadsheet cells into the foundations of spend visibility.
Data categorization
Using AI to clean data might make sense, but categorization probably seems like something your team would want to do manually. You know your categories better than anyone else, after all, and getting them right is crucial.
However, there’s just as much data to deal with in this step as in the previous one, and categorization is even more involved than cleaning. By the time you have everything in place, it will be time to refresh the data for the next month or even quarter, and that’s ignoring the possibility of human error. The good news is that machine learning is especially skilled at categorization.
Because a properly trained model will have seen so much spend before encountering your data, it will be able to see patterns in your data’s purchase descriptions, vendor names, purchasing departments, etc. Using these patterns, a model can rapidly and accurately organize your data, setting the stage for comprehensive category management.
Learn more about why your category taxonomy is just as important as your categorization accuracy.
Anomaly detection and data hygiene
One reason for the notorious quality of Procurement’s data is that anomalies and poor hygiene are hard to detect and control across an enterprise. When someone mistypes in one row of a spreadsheet, that error can go unnoticed for weeks or months. But because machine learning in spend analytics is trainable, flagging these instances is easy. Machine learning also nearly eliminates the possibility of human errors in spend analytics simply by removing humans from everything but the output review process.
Audits
There’s a lot that can hide in an organization’s spend data—maverick spend, supplier pricing discrepancies, inappropriate purchasing. These issues are often invisible because someone has to look through a spreadsheet line by line to find them. But machine learning in spend analytics can make these issues impossible to ignore by flagging them for specific dashboards, like the one pictured below.
Supplier categorization and performance management
On the surface, categorizing suppliers may seem like a simple exercise once you have a list of the companies you buy from. But some vendors have diverse product and service offerings like hardware consulting, and software. Categorizing these manually would be too time intensive to be feasible, even though it’s extremely important.
As we’ve discussed, variation among vendor naming conventions is another common data issue. Even a vendor with a single product offering can have 10+ names across your various data sources. Without machine learning, there’s no way to fix these issues except by hand.
KPIs and benchmarking
Procurement KPIs are foundational for mature organizations because they make important information easy to track, report, and keep from falling through the cracks. Likewise, benchmarking gives teams perspectives on how they’re improving over time and performing compared to leading organizations.
Machine learning speeds up both. KPI tracking becomes an automated process that doesn’t require someone to look for and report the information. Often, it lives in pre-populated configurable dashboards. Machine learning can also facilitate benchmarking, especially internally, by comparing past data points to current ones, providing comparisons, and highlighting specific areas that deserve attention.
Implementing AI in spend analysis
The good news is that implementing machine learning analyzing spend is a mostly plug and play decision. However, there may be some foundational work you have to do before AI slots into your tech stack seamlessly. Below, we’ve outlined a few factors to consider before choosing AI for your spend analytics.
Data quality
Usually, implementing AI before you fix your data issues will only make those problems worse. Machine learning in spend analytics is the one exception; it’s designed specifically to fix data quality and fragmentation issues. However, there are two caveats. First, you should do everything you can to improve data quality at the source levels. Second, you should know that more generalized GenAI use cases can crumble if they have a foundation of bad data.
See why our data experts say that feeding AI bad data is like cleaning a lens with a muddy cloth.
Training and skill development
Another great benefit of machine learning for analyzing spend is that it requires little training and no additional headcount. Machine learning models built into spend analytics solutions like SpendHQ run completely on the vendor’s side. Other than quality assurance and final approvals, you and your team won’t need to do anything until it’s time to use the intel.
Integration with existing systems
Implementing a machine learning model can be a light lift, but you still need to get the data out of your systems and into the model before anything else can happen. You can export files and upload them manually, but this can be a time consuming and frustrating tasks, especially if access to ERPs and suites isn’t consolidated.
The fastest option is to choose a spend analytics and machine learning solution that integrates with your existing systems. This option can automate the process of ingesting data, not just during implementation but also during refreshes. The result: faster initial time to value and updated spend intelligence on a quarterly or even monthly basis.
Vendor selection
Unless you decide to build your own machine learning model, the vendor you select will be the single most important factor to your project’s success. Remember, every AI or machine learning model is separate and the quality of how it was built will determine how well it works for you.
Before you make a decision, ask your vendor the hard questions like:
- How expansive was or is the model’s training data set?
- What is the accuracy rate of its suggestions for different processes?
- What is the model’s primary objective or function?
- How quickly can it process data and requests?
- How long have you been using machine learning in this capacity?
Data security
Equally important, you need to ensure that the model you’re using is secure. The rise in popularity of artificial intelligence has raised critical security concerns from IT experts—feeding sensitive data into a model can have disastrous results if the model doesn’t follow the right security protocols. Before you select any piece of software, bring your IT department into the conversation and make sure you understand how the model anonymizes and protects personally identifiable data (PII).
The future of spend analysis in Procurement
The pace of Procurement at both the tactical and enterprise strategy levels will continue to force decision making to speed up. Infusing AI into spend analytics gives teams a golden opportunity to not just keep up but to transition into a predictive and proactive role.
The faster teams can turn their raw data into a comprehensive picture of organizational spend, the broader and deeper their impact can be. Instead of managing category spend as it develops, they can guide the direction of sub-categories based on developing market trends. They can piece together risk-averse supply chains. They can build vendor profiles that are simultaneously eco-friendly, diverse, and profitable.
Most importantly, a future where machine learning is at the heart of spend analytics doesn’t have to take human ingenuity out of the equation. It can empower it. By taking the manual, repetitive work off of humans, AI allows our best capabilities to shine once again: problem solving, cross-functional collaboration, and creativity.
SpendHQ’s role in driving data-driven decisions in spend analytics
SpendHQ has placed machine learning at the heart of its spend analytics solution for years. Born out of a boutique strategic sourcing consulting firm, our earliest product leaders understood that offloading Procurement’s infamous, looming data chaos was one of the greatest services we could provide.
Since then, our model has analyzed $8+ trillion of enterprise spend and developed an intimate understanding of the vendor names and categories that make up a modern organization’s spend cube. As a result, it can typically provide actionable spend intelligence in just a few weeks.
To learn more about how you can use SpendHQ’s machine learning in spend analytics model to bridge the data gap left by fragmented suites and ERPs, download our guide Don’t Fear the Reaper: How AI in Procurement: How AI Empowers Procurement.