Deterministic Agents, Not Chatbots: Field Materials' Take on Construction AI Procurement
Why Construction AI Procurement Can't Afford to Hallucinate.
The original vision sold by construction ERP (enterprise resource planning) vendors was that one software solution would run every part of the business value chain for visibility, control and efficiency.
But while this approach helped ERP vendors win a larger share of the customer wallet, functions like procurement may not have received much investment or attention. In the meantime, project owners expect to drive efficiencies and protect margin in the business using artificial intelligence (AI). And AI procurement software that can streamline operations and improve margins on materials has a great return on investment (ROI) story to tell.
Field Materials AI is one of the more interesting players in the construction AI procurement space, and I first delved into their product a few years back. At the time, the company was expanding from its original focus on wall and ceiling contractors into additional trades including mechanical, electrical and plumbing (MEP) contractors and self-performing GCs. Since that early conversation, MEP has become the lion’s share of Field Materials’ customer base.
In 2023, the company launched AI extraction agents that read and interpret supplier documents like quotes, delivery slips, and invoices. This helped dramatically reduce the time contractors spend on data entry and manual verification steps in procurement and AP.
“We're not just bolting AI onto an existing platform as a chatbot,” Sadikov said. “AI is the essence of our platform.”
The process starts with large language model (LLM) inference applied to any data input. Rather than traditional optical character recognition (OCR), Field Materials interprets such data as voice, text, or PDFs by mapping it to structured data in your ERP like purchase orders, delivery slips, and AP invoices. This capability also reduces the need for EDI (Electronic Data Interface) integrations with material supplier systems and simplifies the user experience for contractors.
“You can dictate, upload a document, take a picture of a form … you should never have to enter any information by hand,” Sadikov said.
LLMs vs Deterministic AI for Construction
More recently, Field Materials has gone beyond LLM inference on user input like PDF documents to AI agents, which automate recurring tasks.
Automation Agents running in Field Materials act as virtual users in the software, executing workflows according to configurable business rules. Field Materials has a broad selection of pre-defined agents for contractors to work with. But users can also create their own agents using natural language that Field Materials translates into deterministic logic built around triggers and conditions.
For example, AP processing is a big pain point in procurement. Many invoices are processed the same way. Overhead invoices, for example, are coded against the same GL accounts. This highly repetitive task is easily automated by AI agents. However, a typical LLM may easily hallucinate and introduce errors in the process, something that the accountants or purchasing managers in the office will not tolerate.
Field Materials solves this by making automation agents completely deterministic. They follow structured logic and don’t hallucinate, so given the same invoice, they will always execute the same set of steps.
This approach aligns with this blog’s repeated focus on the importance of deterministic AI for construction, not only for explainability but to conserve compute power and AI tokens.
“The idea is to take any recurrent procurement or AP task and automate it using the system,” Sadikov said. “And the cool part is that you can combine automation with inference.”
Material Pricing Insights
“When you buy lots of materials, how do you track the prices you pay throughout the year on different jobs?” Sadikov asked. “How does the price fluctuate? With the Pricing Intelligence module that we launched last year, you can see the best time to buy materials throughout the year. The data shows, for example, that buying materials in the spring is almost always better than buying materials in the summer because the demand is the highest in the summer and the prices are as much as 25% higher.”
Other price fluctuations may follow different patterns, and Field Materials will use these insights to time larger buyouts or hold-for-release orders that enable a supplier to keep the materials on their site after purchase, and deliver them when contractor requests materials to be released.
BOTTOM LINE: In categories where advanced AI capabilities deliver clear and measurable wins, extending a system of record like ERP or Procore can drive significant ROI. In selecting a vendor, the functionality and how well it meets immediate and future needs should be top-of-mind. And when dealing with emerging technologies and venture-funded software companies, the staying power of the organization is critical. And while competitor Kojo has raised nearly five times the venture capital ($94M) and processes $5 billion in orders annually, Field Materials manages over $2.5 billion in annual purchasing volume with $21M in venture capital, with customers benefiting from procurement and AP processes automated by AI agents. Technology-forward companies like Field Materials AI will have an easier time winning and retaining contractors. Both companies are privately-held, so we cannot compare typical software-as-a-service (SaaS) metrics like churn rate, but Field Materials maintains a very high ARR-to-headcount efficiency ratio typical of lean AI-native SaaS companies.
At Rathmann Insights, we see LLMs as essentially a tool for human-computer interaction rather than intelligence, and Field Materials users will have a more intuitive time with the platform as it does not require them to train on complex user workflows or work through lengthy electronic data interface (EDI) supplier onboarding processes.
Contractors should do due diligence and ask for demos and proof points for claims software vendors make about their predictive AI capabilities, particularly when vendors expect the customer to structure their own AI workflows. Do they have a roadmap for the product, and how well have they conformed to their public statements about product direction in the past? And is your business truly a fit for the product?
Comments