How DEMA Engineering turned decades of knowledge into an AI risk assessment system

DEMA Engineering Company builds chemical dispensing systems and industrial valves for customers who can't afford a part to fail in the field. That standard shows up in a process called Design Failure Mode and Effects Analysis, or DFMEA: a line-by-line audit of every component in a new product, done before it ever ships. DEMA took that process seriously. They just hadn't found a way to make it fast.

The opportunity

Every new product meant the same routine: eight to ten engineers, six to eight meetings, and a blank spreadsheet that had to be filled in from memory. The team would sit together and work backward through every possible failure, cause, and effect, component by component, almost entirely from recall.

The strange part was that none of this was new information. DEMA had been building similar products for years, and most of what the team was reconstructing already lived in old DFMEA files scattered across SharePoint.

"We were essentially starting from scratch every time. We had decades of institutional knowledge, but it lived in people's heads and in past DFMEA documents scattered across SharePoint."

— DEMA Operations Lead

That disconnect was costly in two ways. It burned hours of senior engineering time on work that was mostly re-derivation, not new analysis. And it left the process exposed: when the quality of a safety review depends on who's in the room and what they happen to remember, something eventually gets missed. DEMA wanted to grow its product portfolio without stretching its engineering team thinner on every launch.

Why Cassidy, built by SoftSnow

DEMA partnered with SoftSnow, an AI strategy and execution firm, to tackle the problem at its root: not the meetings themselves, but the fact that none of DEMA's historical knowledge was set up to be reused. SoftSnow began with discovery, running sessions with DEMA's team to understand exactly how DFMEAs got built and where the underlying data actually lived.

From there, SoftSnow took DEMA's archive of past DFMEA documents and restructured it into organized, product-specific databases inside SharePoint. To actually put that knowledge to work, they chose Cassidy as the engine behind it. Cassidy could connect to SharePoint directly, understand DEMA's data well enough to reason over it, and carry out a full multi-step process rather than just surface a document when asked.

What they built

SoftSnow designed the agent around DEMA’s existing DFMEA process, giving engineers a populated starting point instead of a blank spreadsheet. The workflow starts simple and gets more sophisticated under the hood:

• The engineer provides three things: a product name, a product type, and an Excel file listing the stakeholders, feature elements, and component elements involved in the new design. No new format to learn, no separate tool to open.
• Cassidy reads the file component by component. For each one, it searches DEMA's historical product databases for similar parts, looking across feature descriptions, component types, and prior failure records rather than doing a simple keyword match.
• It pulls the most relevant precedent for each component, searching the relevant product database for similar entries and surfacing the closest-matching examples.
• It drafts the new DFMEA, populating failure modes, failure effects, root causes, and estimated risk priority numbers into a spreadsheet built in DEMA's own existing template format, so nothing about the output looks or feels foreign to the team.
• The finished file lands back with the engineer, ready to walk into a review meeting with rather than build from scratch.

The effect on DEMA's actual meetings was immediate. Engineers stopped opening blank templates and started opening drafts, pre-populated analysis to react to instead of a blank page to fill in from memory.

"It gets us 80% of the way there. We still review everything, adjust language, validate the risk scores. But instead of staring at a blank spreadsheet, we're refining analysis that's already grounded in what we've learned."

— DEMA Operations Lead

The impact

The nature of the work itself shifted. Instead of spending meeting time generating ideas about what could fail, engineers spend it stress-testing a draft that's already comprehensive, which means more attention goes to the edge cases that actually need human judgment.

The system also doesn't stay static. When a new failure mode turns up during an assessment, it gets logged back into the SharePoint database, and the agent folds that knowledge into every assessment that follows.

"We're not just saving time. We're designing a system that gets smarter with every product we develop."

— DEMA Operations Lead

That compounding is really the headline result. Every assessment now draws on the entirety of DEMA's institutional knowledge, not just what the people in the room remember, which has meant fewer safety-review delays and more consistent risk coverage across their full product line.

Why this matters beyond DEMA

DFMEA is a manufacturing term, but the problem it solves isn't industry-specific. Any team sitting on years of hard-won expertise, whether that shows up as compliance reviews, QA checks, or testing protocols, faces the same opportunity DEMA found. Getting there took real work: discovery, workflow mapping, data analysis, restructuring historical records into organized databases, and designing an agent that could act on all of it.

DEMA Engineering turned years of engineering judgment into a system that gets sharper with every product it touches. Whether you're looking to automate technical risk assessments, turn historical documentation into a living knowledge base, or give your team a faster starting point for judgment-heavy work, Cassidy, implemented with partners like SoftSnow, can help you build an AI solution that fits your operations.

Book a demo to see how Cassidy can work for your team.