Angelica Burlaza
(Kunal Barman (Co-Founder, CEO of Faction), Danish Raza (Co-Founder, CTO of Faction))
“Manufacturing doesn’t only have a data problem,” Danish Raza said. “It also has an attention problem — too many folks staring at screens, re-entering the same data, while the real constraints in the system go unmodeled & relationships get left behind.”
The remark captures the moment in which industrial companies now find themselves: awash in information, short on time, and newly exposed to a generation of automation tools that not only log events but also decide what to do next. Raza, co‑founder and CTO of Faction, is betting that a new class of “agentic” AI systems will pull manufacturing out of its clipboard era and into something closer to continuous, software‑mediated optimization.
An Inflection Point for Industrial Automation
Industrial automation is not new; programmable logic controllers and robot arms have been refining assembly lines for half a century. What is new is the software layer sitting above those machines, where thousands of distributors and manufacturers still rely on spreadsheets, email threads, and aging enterprise systems to turn a request for quote into cash in the bank. While the production line may be automated, the business processes surrounding it often remain stubbornly manual.
Market researchers estimate that the global industrial automation and control systems market was worth approximately $200 billion in 2024 and is expected to reach $380 billion by 2030, depending on the methodology and scope. Those forecasts suggest that the next leg of automation will be less about robots wielding torque wrenches and more about software orchestrating work across procurement, sales, inventory, and finance.
From Hedge Funds to Hardware to Factory Floors
Raza’s own trajectory into this world runs through some of the least tangible corners of the modern economy. Before founding Faction, he worked as a quantitative developer at Castleton Commodities International, building high-performance systems to model metal, oil, and gas flows and to forecast commodity prices that ultimately shape how manufacturers price everything from copper tubing to hydraulic fittings. That experience, he said, “forced a very particular way of thinking: the physical world as a network of constraints you can measure, model, and trade around.”
He also spent time at Arm, contributing to the chip designer’s confidential compute architecture. That work on low‑level systems, cryptography, and distributed infrastructure underpins Faction’s pitch to industrial customers who worry about downtime, integration risk, and data leakage.
Raza left Imperial College London, where he had been studying computer science, to work on Faction full‑time. “Manufacturing looked like this enormous, under-modeled system where modern AI could actually move the needle,” he said. “I wanted to move faster and go all-in.”
A Multi‑Modal “Agentic” Platform
Faction, headquartered in San Francisco, describes itself as the AI partner to distributors and manufacturers, offering a platform of software “agents” that slot into existing enterprise resource planning (ERP) and customer relationship management (CRM) systems. The agents read the same documents, emails, and file formats as a human operations team, but take on much of the repetitive work.
One cluster of modules focuses on quote and order processing. Incoming requests arrive as PDFs, spreadsheets, email threads, or text messages; an ingestion engine converts those into structured, ERP‑ready data, maps items to the internal product catalog, and assembles draft quotes for human review or straight‑through processing. Other agents place and track purchase orders with suppliers, cross-match items across a large product graph to identify alternatives, and verify prices and availability across multiple manufacturers simultaneously.
“Multi‑modal is not a buzzword for us,” Raza said. “A distributor lives in documents, in phone calls, in terminal screens, in weird vendor portals. If you can’t operate across all of those, you aren’t truly building for the industry.”
The Data Problem Behind Messy Documents
The deeper problem is not merely that documents are unstructured but that the underlying product data inside many industrial businesses is incomplete or inconsistent. Vendors describe the same fitting or valve in subtly different ways: legacy systems encode attributes in idiosyncratic schemas, and part numbers change across decades of acquisitions. Traditional automation systems treat each of these contexts as separate silos.
Faction’s answer is a proprietary data model that sits across a distributor’s or manufacturer’s catalog, normalizing attributes and relationships across millions of items, then exposing that as a shared substrate for its agents. That product graph enables the system to perform cross-manufacturer item matching, suggesting alternatives when a preferred item is out of stock, and to evaluate trade-offs in price, margin, and lead time.
A Sector Under Pressure
The timing is not accidental. Distributors and manufacturers in sectors such as pipe, valves, and fittings (PVF), electrical, HVAC, packaging, and safety equipment face tight margins, uneven demand, and growing customer expectations for speed and transparency, as they are accustomed to consumer-grade e-commerce. Many operate on thin working capital and cannot easily absorb additional headcount to cope with peaks in quoting, sourcing, and order management.
Analysts forecast that the broader industrial automation market will grow at a rate of roughly 8 to 11% annually between 2024 and 2030, with the Asia-Pacific region retaining the largest regional share as it expands its manufacturing base. Within that, software‑centric offerings that orchestrate workflows across machines and business systems are expected to capture a growing portion of value. For many mid‑sized firms, the question is no longer whether to automate, but how far to push automation without alienating customers or losing control over core processes.
Supporters See a New Operational Layer
Supporters of platforms such as Faction argue that agentic automation can serve as a kind of operational mesh, knitting together legacy ERPs, home‑grown tools, and modern AI into a coherent whole without requiring a multi‑year systems overhaul. Because the agents work through existing interfaces and APIs, they say, companies can experiment with narrow workflows before scaling successful patterns.
For Raza, the conceptual leap is to treat the entire quote-to-cash and procure-to-pay continuum as a single system, rather than a chain of loosely coordinated departments. “If your quoting logic doesn’t understand your real inventory constraints, you end up with promises you can’t keep,” he said.
He describes Faction’s role less as replacing workers than as changing the shape of their days. Instead of copying data from emails into forms, inside sales teams focus on exceptions, negotiations, and customer relationships; procurement staff focus on supplier strategy rather than status checks; finance teams focus on disputes instead of routine reconciliations. “The constraint in these businesses is not willingness to work hard,” Raza said. “It’s cognitive bandwidth.”
Critics Warn of Hype and Fragility
Not everyone is convinced that agentic AI is ready to shoulder that bandwidth. “The industry has seen waves of automation hype before, from early MRP systems to ERP to RPA, and each time the reality is messier than the marketing,” said Laura Chen, a fictionalized composite of consultants who advise manufacturers on digital transformation. “AI agents that can ‘figure it out’ sound compelling, but factories are unforgiving environments. If a model misinterprets a spec or miskeys a part number at scale, the downstream costs can be enormous.”
Those concerns echo broader debates about AI’s role in high‑stakes domains. Studies of automation adoption stress that productivity gains depend not only on the technology’s capabilities but on how organizations redesign processes, reallocate labor, and invest in skills. Without that work, automation projects can stall or even erode performance, as partial implementations create new failure modes.
Building for Reliability and Governance
Raza does not dismiss these objections. In fact, he argues that his team’s background in security, systems engineering, and quantitative finance makes them unusually attuned to operational risk. At Arm, he worked on secure execution environments that had to run correctly across millions of machines under adversarial conditions.
“That experience makes you deeply conservative about what you let software do without human eyes,” he said. In Faction’s deployments, he emphasized, customers can choose levels of autonomy: in some cases, agents prepare drafts for human approval; in others, they execute end‑to‑end but with detailed logging, anomaly detection, and controls tied back into the customer’s existing governance frameworks.
Human Capital in the Loop
Underlying the dispute is a more basic question about the future of industrial work. McKinsey and other research groups estimate that, by 2030, automation technologies could reshape or replace a significant share of work activities, with midpoint estimates suggesting that half of today’s tasks could be automated sometime between 2030 and 2060. For operators and line workers who have already lived through earlier waves of lean initiatives and offshoring, yet another promise of “productivity” can sound ominous.
Raza insists that Faction’s agents are designed to extend, not erase, human expertise. In practice, he said, the first beneficiaries are often experienced staff who find themselves freed from rote work to focus on training newer colleagues, refining business rules, or working directly with key accounts. “There’s a lot of tacit logic in how a great inside salesperson thinks,” he said. “If you can encode that once and distribute it, you’re not replacing them; you’re scaling them.”
A Cautious Path to 2030
As manufacturers and distributors weigh their options, the path to 2030 is unlikely to be a straight line. Early adopters will test agentic systems in narrow domains. Yet the structural forces driving interest — aging workforces, tight labor markets, volatile supply chains, and rising expectations for responsiveness — show little sign of easing.
For Raza, that creates both opportunity and responsibility. “If we’re going to wire AI this deeply into the operational fabric of these companies, we have to treat it with the same seriousness as a risk system at a hedge fund or a kernel in a data center,” he said. “It’s not a chatbot; it’s infrastructure.”
As the conversation wound down, Raza returned to the question he began with: attention. “The machines on the floor are already pretty fast,” he said. “The bottleneck is how quickly a business can notice what’s changing and respond coherently. If automation can buy humans back that attention, then it’s not about replacing work at all. It’s about finally seeing the system we’ve built, and steering it with a clearer head.”