Marc Kermisch is Chief Technology & AI Officer at Protolabs, a digital manufacturing service for accelerated production of quality parts.
Few topics are getting as much focus in manufacturing as artificial intelligence. Which tooling should you adopt? Which models work best? How do you maximize ROI? How quickly can it all happen?
It’s exciting to be part of an industry going through such significant change, and it’s easy to see why manufacturers are eager to invest in this new technology. However, it’s also easy to assume that AI can simply be unleashed and expected to work.
Before they jump in, companies need to build a strong foundation first, prioritizing connectivity, process and, most importantly, data management.
Clean Data: What Is It, And How Do You Get It?
I’ve found it’s best to start with process and data governance. A data catalog can define data elements, explain how they are categorized and describe how data flows into the organization. A data catalog can provide the context for AI to give accurate answers consistently. Without that context, the same question may produce different answers on different days. If AI receives poorly structured data, the outcome is unlikely to be clear.
On the factory floor, one machine might measure productivity by hours running, another might measure by parts produced and another might measure by a different metric altogether. If AI is asked about plant productivity, it may not know how to reconcile those definitions and could return inconsistent or inaccurate answers. In this example, if the data is clearly defined, well-managed and “clean,” AI can use it consistently and produce repeatable answers. With work, it can do this across disparate systems.
Manufacturing Data Is More Challenging
Clean data can improve the effectiveness of AI regardless of the industry you’re working in, but it’s especially relevant in manufacturing, where the data informs physical outcomes. The industry relies on, in part, geometry, manufacturability recommendations, quoting accuracy and more. Failing to feed AI with consistent data can mean rework, machine downtime, higher costs and missed delivery schedules. This is compounded by the fact that manufacturing is closely aligned with high-stakes industries like aerospace and defense, medical technology or industrial equipment, where every part must meet strict criteria.
Manufacturing also differs because it requires so many different data sets from different systems. Customer order data may come through CRM or ERP systems, while labor, machine, quality and on-time delivery data may come from separate plant-floor systems. Plant floors are not always well-connected, and operators are often focused on producing parts rather than considering how their data might support AI. Over time, I’ve found that many manufacturers have also underinvested in technology, data and standardized processes.
Manufacturing companies often run on lean margins and have significant capital tied up in raw materials, machines and labor. As a result, the technology teams supporting plants tend to be smaller. If you compare this with industries such as banking, which are more asset-light, it’s very different. Those industries, in my experience, often have more modern and integrated systems. They typically have larger technology teams and cleaner, centralized data warehouses.
Enter The Digital Thread
To reap the benefits of AI-connected systems, manufacturers need a clean digital thread that connects design intent to production execution and quality outcomes. When you can connect a part’s design to how it actually performs in machining, molding or inspection, this often results in a feedback loop that AI needs.
You can think of the digital thread like the contrail an airplane leaves that shows where it has traveled; this is a digital trail an order leaves as it passes through systems. By capturing and defining that thread, you can identify where friction exists in the system. Once those friction points are understood, it’s possible to introduce system improvements, automation or AI to reduce friction for customers.
Digital threads can help you track different aspects of your orders, from the first CAD file upload to the time completed parts are delivered. Along the way, you can also collect various data points about manufacturability, conformity, machine performance and quality. This contrail of information can lay the groundwork for applying AI.
Putting Improved Data Systems Into Practice
I’ve found that a hybrid approach, combining in-house contextual knowledge with third-party software, is a solid strategy for improving data management. Leaders can centralize their data warehouses using a cloud solution that collects data from different source systems and features them in one place.
This process involves three stages, or “layers,” as I like to call them. First is the bronze layer, where raw data is collected without applying quality rules. The next stage is the silver layer, where data is transformed, cleaned and aligned with master data management rules. This includes defining elements such as the customer, product and part names. The gold layer includes the highly trusted consumer data used in enterprise dashboards. For us, that includes things like daily sales, shipments, website activity, file uploads and quotes. These dashboards can combine enriched data sets into trusted views that can, in turn, help train your in-house AI tools.
If you’re still wondering where to start, I’ll give you a few examples from my company’s work. Our pilot AI projects include experimenting with technical drawing assessments, a search functionality that scans comparable parts that have been manufactured before, a feature that recommends materials based on specifications and a project that helps sales teams by collating customer and company information. The goal with these projects is to demonstrate how manufacturing can see returns from improved data management. The work could result in faster quoting and design for manufacturability (DFM) analysis, a shorter path from design to production, better predictive capabilities for quality control and improved customer outreach.
It’s common to expect immediate, measurable returns from AI investment. That said, manufacturing companies need to instead aim to become a better-connected ecosystem, where part data and processing data can more accurately inform AI-powered workflows. This transformative technology has the potential to drastically overhaul the way that we manufacture products, but it all comes down to the quality of the information we feed it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
