By Craig Webb
President, Webb Analytics
Construction supply dealers have reason to be both intrigued and confused by the hubbub over Artificial Intelligence (AI). Whether LBM dealers should care depends both on what you’re doing now and whether AI’s biggest recent advances will be worth the cost.
Here’s my take:
AI’s innovations in marketing and data analysis hold the greatest short-term promise for construction supply.
Big dealers will benefit by using AI to achieve more consistency across far-flung operations.
Most small dealers should continue improving their key processes—particularly dispatch/delivery, inventory management, e-commerce, and ERP utilization—before spending any time thinking about higher-level AI innovations. Smaller dealers also will have to move their data to the cloud.
Except for better pricing, AI’s benefits are more likely to show up on the expense rather than the revenue line.
“I think it’s starting to become more important,” said John Carrico, a VP at Epicor specializing in product management, distribution, and building supply. Earlier in the interview, he stressed: “We by no means are saying AI is the end-all and be-all. But we see the use of AI, and it comes down to AI helping a user make the best decision and make an informed decision—and sometimes make a decision for them.”
A Fuzzy Definition
Arthur Duffy, Founder and Managing Director of Excenta Ltd., a U.K.-based IT consultant, cautions that lots of processes cited as examples of AI strike him as just data processing. You could argue that a cash register that tells how much change to give on a purchase is “intelligent.” But to Duffy, that’s not AI.
As he defines it, “AI is a computer system that is able to perform tasks that ordinarily require human intelligence. These artificial intelligence systems are powered by machine learning.” Another way to put it is that AI’s value arises when the problem comes too frequently, with too many variables, and from too many sources for humans to solve the task at a profitable speed.
Consider direction-finding software like Uber, and Waze. A lumberyard’s dispatcher can do manually much of what the apps do so long as only a few trucks are involved. But if a dispatcher had to handle Uber’s15 million rides daily, that person would be overwhelmed. Beyond that, these route-finding apps are constantly “learning” which routes work best by collecting data from lots of resources, such as the crowdsourcing contributions that help power Waze.
8 Forms of AI, Plus One
Christina Stathopoulos, a former analytical lead at Waze, calls that type of AI “prescriptive intel.” In a speech last month in Berlin at a summit for DIY dealers worldwide, Stathopoulos broke AI into eight types:
Descriptive Intel. Here, AI creates relationships between data to understand past patterns, particularly if the connection was unexpected. The classic example is the early 1990s finding of a close correlation between sales of beer and diapers. Humans stumbled on that one. Today, AI can do a much better job finding patterns.
Predictive Intel. This is where you use past patterns and current demand to predict the future. Most of construction supply’s ERP systems already do this when they suggest stocking levels.
Prescriptive Intel. Here, AI sorts through multiple options to figure out the next course of action, as with routing software.
Autonomous Decision-Making Systems. Here you give the computer a problem and then let it “learn” the best ways to resolve the challenge. So-called smart buildings do this now when they strive to reduce energy use while maximizing comfort.
Hyper Personalization. The more AI knows about you, the better it’ll be able to find and suggest products that meet your need. Thirty-five percent of Amazon’s sales involve software-generated recommendations, Stathopoulos said.
Artificial Vision. This is when AI learns to read an image and convert it into computer-readable data. Starbucks has removed its name from its logo, but AI can enable a scanner to know when a Starbucks mug is on the shelves just by spotting the mermaid.
Natural Language Processing. Think of Siri, Alexa, and “Hey Google.” All of them record human speech and convert what’s said into information that a computer can handle. Natural-language processing also helps computers scan e-mails and then sidetrack the ones it thinks are spam.
Generative AI. Here you find ChatGPT and software that creates images. In this case, the AI collects information from across the Internet and applies it to the request in a format that fulfills the request. This is the most recent form of AI to win attention.
To these eight, add Quality Control. John Maiuri of ECI believes one of AI’s biggest potential benefits for his customers will be how it can spot inconsistencies across large volumes of data, such as when the same customer appears twice in the system because one of the entries has a spelling error or an address change. Poor data hygiene mucks up systems and leads to bad results.
Expect to Start by Automating Words
One of the first places you’re likely to see ChatGPT-style AI enter LBM is through communications. AI makes it easier to write standard versions of many e-mails that dealers send out today, such as reminders to pay, requests for information, and thank-you notes. You might have many of these sitting around already, but AI helps you use employ ever-stronger language when, say, your customer’s invoice is getting further and further past due. It also helps automate the sending of these reminders, including customized info.
Think of the opportunities for your website, too. AI word-generating tools can make it easy to produce a first draft of how-to articles that you then can customize to your needs. You can pull together information faster, thus allowing you to reach customer segments more regularly. (But again, you should review the AI-created content before sending it out to make sure you’re not passing along factual errors, fake information, and even disinformation.)
Marketing also is likely to get more sophisticated. Bulk e-mail systems such as Constant Contact have begun incorporating ChatGPT into their offerings. An April 2023 survey of public relations officials conducted by WE with the University of Southern California’s Annenberg Center for Public Relations found 88% saying AI will improve their speed and efficiency and 72% believe it’ll reduce workloads. But it’s still early days for these companies: Only 16% said they are extremely knowledge about AI applications, and just 23% said their organization is making changes to how they work due to new AI tools.
All of these advances theoretically could boost sales, but it's just as likely that AI's main benefit here is that it'll let you do more work with the same or fewer people and resources. As in many other areas, AI's first benefit probably will be reduced costs and thus higher profit margins.
Does Anyone in LBM Use High-Level AI?
Depending on whether you subscribe to Stathopoulos’ loose or Duffy’s tight definition of AI, you could say lots of dealers or virtually no dealers use AI today. One of the rare Duffy-level examples is at The Home Depot, which has created a program that gives store associates an optimal order for restocking items. Such an algorithm’s suggestions can be based on a myriad of constantly changing factors, including the weather forecast (‘Snow coming? Get out the shovels.”), recent sales, whether a promotion is about to launch, and how much product is on the shelves.
Big dealers also are likely to find AI can help them monitor whether branches are doing what the central office wants. Latin America’s biggest DIY chain, Sodimac, uses robots to scan the aisles for stocking levels, out-of-date price tags, and fidelity to the planogram. The collected data then can be converted into charts (see graphic at top of page) showing out-of-stock levels overall and by store, as well as zoom in on a particular branch to see how well that branch is keeping up with price changes over time.
Stathopoulos also cited the German DIY giant Obi, which has introduced a “hey Obi” service on its app that analyzes voiced requests to route the customer to the most suitable product expert on its help desk. She also likes how Lowe’s has created a digital replica of its stores so that product leaders in the central office can—with AI help—can rearrange a planogram and predict how the moves would affect particular products’ sales.
Better Pricing. Better Utilization
Duffy sees some of AI’s greatest promise in the pricing advice it can give. Britain’s Bubo.ai claims it can boost gross margins three percentage points (e.g., to 18% from 15%) by examining all the discount rules established for every product and every customer. Bubo says its system “then ‘learns’ what worked (profit improved) and what didn’t (no impact on profit) and makes improvements for future transactions. Ultimately, it has the ability to optimize the profit of every customer purchase.”
Naturally, the more customers and products you have, the more complicated your current discounts are likely to be. Thus, big dealers are more likely than smaller ones to benefit here, just as big dealers also can use AI for managing far-flung operations.
For smaller dealers, the next steps are already in your possession. Most dealers’ ERP systems are dramatically underused, officials at software companies say.
Coming Attractions
At a conference earlier this year, Carrico put into one slide 21 different ways AI could help product distributors. Many involve gathering, organizing, and analyzing large amounts of data based on something as simple as a verbal request—no code-writing or intense research needed. Others work with software like optical character recognition to “learn” where to find and transcribe the same needed information that appears in different places on different companies’ invoices.
Hyper-personalization goes beyond pricing with AI. By pulling together both in-store and online information about a customer, AI can help with the sales pitch, tone, and timing of efforts. It also will be easier to manage customer relations across media rather than having to sort through records now held in separate silos for text, e-mail, online, and in-person contacts.
“We see ourselves helping with non-value-added activities, doing work faster and more accurately than a human can do,” Carrico said. “That's where we’re getting our feet wet first.”