How AI Routes RFQs Across Multi-Plant Manufacturers

The Montanstahl Group operates three production sites - Stabio in Switzerland, Schwerte in Germany, Talamona in Italy. Each site has its own production processes, its own metallurgical specialty, its own institutional knowledge built up over decades. When a customer in California sends an RFQ asking "can you produce this specialty steel profile in this alloy at this tolerance," the answer might live at any of the three sites. Often it lives across all three - Stabio handles cold-rolling, Schwerte does hot extrusion in nickel and titanium, Talamona produces hot-extruded seamless profiles in superalloys. A single feasibility question can route through three countries before being answered.

Before Mavlon AI was integrated into their workflow, that routing was largely a series of phone calls and emails between senior engineers at each site. The question “can we make this” became a multi-day discovery process. A sales engineer in one country asked a senior engineer in another country whether a similar profile had been produced before. The senior engineer dug through their site's historical records. The answer came back in hours or days. The customer, meanwhile, was waiting - and sometimes the competitor responded first.

This pattern - cross-plant feasibility routing through human coordination - is the structural reality of every multi-site manufacturer I have spoken with. Specialty steel groups. Aerospace tier-1s with multiple plants. Automotive tooling builders with engineering centers across continents. Heavy machinery groups with regional production. The specific cross-plant configuration differs. The cost of internal coordination on feasibility questions does not.

The Multi-Site Manufacturing Reality

Multi-site manufacturing exists for good economic reasons. Different sites specialize in different processes. Regional sites serve regional customers with lower logistics costs. Acquired sites bring unique capabilities that the parent organization could not have built organically. The Montanstahl Group exists in three countries because the three sites' combined capabilities are greater than any single site could offer.

But multi-site structure creates a coordination overhead that single-site competitors do not face. When a single-site manufacturer receives an RFQ, the answer to “can we make this” lives in the building. The senior estimator walks down the hall and talks to the process engineer. The decision happens in hours.

When a multi-site manufacturer receives the same RFQ, the answer often lives across multiple buildings, multiple time zones, and multiple business units. The estimator in Switzerland does not know what the production engineer in Italy knows. The production engineer in Italy does not have the historical database from Germany. The discovery process to assemble a complete answer takes hours or days instead of minutes.

At the Montanstahl Group, this coordination overhead was the explicit problem statement when they engaged with AI. Senior engineers across the three sites were spending thousands of hours per year on the same recurring question: "Can we actually make this?" The answer required combining knowledge from Stabio's cold-rolling history, Schwerte's hot-extrusion records, and Talamona's superalloy production data. No single engineer had complete access to all three.

How the Coordination Cost Compounds

The cost of cross-plant coordination compounds in four ways.

First, response latency. A multi-site manufacturer's feasibility response time is structurally longer than a single-site competitor's. If the multi-site response takes two days and the single-site response takes two hours, the customer often has already committed to the single-site supplier by the time the multi-site quote arrives.

Second, knowledge attrition. Each routing step risks losing nuance. The sales engineer in Switzerland asks the production engineer in Germany whether a similar profile has been produced before. The production engineer says yes, summarizes the key facts in three sentences, and gets back to their primary work. The three sentences leave out the actual reason that profile was challenging to produce - a context that lived in the production engineer's head but was not in their summary. The sales engineer does not know to ask the follow-up question. The quote goes out without the context. Execution surfaces the issue. Margin is lost.

Third, internal demand on senior engineers. Senior engineers at each site become the routing nodes for cross-plant questions. Their time gets fragmented by phone calls and emails from sister sites asking "have we seen something like this before." The senior engineer in Italy spends 4-6 hours per week answering feasibility questions from Switzerland and Germany. That time is not available for their own site's program work.

Fourth, decision opacity. When a feasibility decision routes across three sites, no single person can later reconstruct why the decision was made the way it was. The senior engineer in Italy answered "yes, doable" based on their interpretation. The sales engineer in Switzerland interpreted that as confirmation. The customer interpreted the quote as a commitment. When execution surfaces a problem, the chain of attribution has been lost - no one remembers who actually decided what.

Specialty Steel: The Montanstahl Pattern

At the Montanstahl Group, the cross-plant feasibility question had a specific structure. Each customer RFQ specified a profile geometry, an alloy grade, a tolerance band, a surface treatment requirement, and a delivery timeline. The feasibility analysis had to answer: which of the three sites can produce this, what process route is best, what is a realistic lead time, and what alternatives can we offer if the exact request is not feasible.

Stabio in Switzerland handles cold-rolled and laser-welded profiles. Geometry is the primary constraint - cold-rolling has specific shape limitations that an experienced Stabio engineer recognizes immediately. Schwerte in Germany handles hot extrusion in nickel and titanium. Alloy is the primary constraint - hot extrusion of high-temperature alloys requires equipment capability that an experienced Schwerte engineer can validate. Talamona in Italy produces hot-extruded seamless profiles in carbon, stainless, and superalloys. Process route is the primary constraint - Talamona's extrusion capabilities span a wider alloy range than the other sites.

A single customer RFQ might require expertise from all three sites: the geometry needs Stabio's interpretation, the alloy needs Schwerte's confirmation, and the tolerance achievability needs Talamona's process route knowledge. Without unified knowledge access, the feasibility question routes through three senior engineers in three countries.

The pre-AI workflow was: sales engineer receives RFQ, calls or emails senior engineer at the most likely production site (based on the sales engineer's best guess), waits for response, sometimes follows up with second or third site if the first site says "we can't but maybe Talamona can," eventually assembles a unified answer to send to the customer. Average cycle time: 2-5 days per feasibility question. Senior engineer time consumed across sites: 6-10 hours per question.

With AI-assisted unified knowledge access, the workflow compresses dramatically. The AI ingests the RFQ, queries the unified historical database across all three sites, and returns a feasibility recommendation: this site for this profile, this process route, this realistic lead time, this alternative if the exact request is constrained. The sales engineer reviews and validates the recommendation. Cycle time: under an hour. Senior engineer time consumed: zero on most questions, with senior engineer review reserved for the 15% of genuinely novel questions where AI confidence is low.

The 85% Pattern

Across the Montanstahl Group's incoming RFQ stream, AI-assisted feasibility routing achieved 85%+ accuracy. For 85 of every 100 RFQs, the AI's recommended feasibility answer was either directly usable or needed only minor adjustment. The remaining 15% were genuinely novel questions - profiles, alloys, or tolerance combinations not represented in the historical database - that correctly routed to senior engineers for human judgment.

This 85/15 split is what makes the structural shift work. The 85% that AI handles is the volume of routine feasibility questions that previously consumed senior engineer bandwidth. The 15% that AI flags for human review is exactly the cases where senior engineering judgment has the most value - novel situations where pattern matching against the historical database is not sufficient.

At the workflow level, the senior engineer's job changes from "answer recurring feasibility questions from sister sites" to "make judgment calls on the genuinely novel cases the AI surfaces for review." The time consumed shifts from 6-10 hours per question across all questions to 2-4 hours per question on only the 15% that need human judgment. Total senior engineer time on feasibility routing drops by approximately 85%.

The capacity that gets freed up is substantial. At the Montanstahl Group's scale of RFQ volume across three sites, the estimated annual senior engineer time savings ran to thousands of hours per year. The estimated annual business impact from capacity gains and higher win rates ran into the low seven-figure range.

Beyond Specialty Steel: The Same Pattern Across Industries

The cross-plant feasibility routing problem is not unique to specialty steel. The same pattern appears across multi-site manufacturing:

Aerospace tier-1 subcontractors with multiple plants face the same question. A Boeing component RFQ might be best produced at one of three plants depending on the specific geometry and certification requirements. The feasibility decision currently routes through plant-level engineering managers. The same 85/15 compression applies when AI mediates the routing.

Automotive tooling builders with engineering centers across continents face it. A Stellantis program for a European plant might be quoted by the European engineering center while the Detroit engineering center holds the relevant past-cell history. The feasibility question routes across the Atlantic. The same cycle compression applies.

Heavy machinery manufacturers with regional production face it. An order from a Japanese customer might be best produced at one of two regional plants depending on the specific configuration. The feasibility decision routes between regional sales engineering and global production engineering.

Medical device contract manufacturers with multiple facilities face it. Regulatory constraints often require specific certifications at specific sites. A feasibility question routes through quality engineering at each candidate site before the commercial team can respond to the customer.

In each case, the structural problem is identical: distributed knowledge across multiple sites, coordination overhead in the routing process, senior engineer bandwidth consumed by recurring feasibility questions, and competitive disadvantage versus single-site competitors who can answer faster.

What AI Changes About the Routing

The shift that AI enables for multi-site manufacturers is the creation of a unified knowledge surface that spans all sites. Each site's historical jobs, process capabilities, and production constraints become queryable through a single interface. The senior engineer's mental model of "what we have produced before and how" is externalized into structured data.

Specifically, AI enables four shifts.

First, instant cross-site feasibility matching. The AI queries all sites' historical records simultaneously and returns the best-matching past jobs ranked by similarity. No phone calls. No emails. No waiting for sister sites to respond.

Second, automatic process route recommendation. Based on the geometry, alloy, tolerance, and surface treatment requirements, the AI recommends which site's process is best matched. The sales engineer no longer has to guess which site to ask first.

Third, alternative routing surfacing. When the exact customer request is constrained by any single site's capabilities, the AI surfaces alternative configurations that the customer might accept. This is the equivalent of asking the most experienced senior engineer in the company "if we can't do exactly this, what's the closest thing we can do" - except the answer comes from structured knowledge rather than a single person's memory.

Fourth, decision audit trail. Every AI-recommended feasibility answer carries a traceable reasoning chain - which past jobs were matched, which capabilities were validated, which constraints were checked. When execution surfaces an issue, the chain of attribution can be reconstructed. The decision opacity problem is eliminated.

Zero-Friction Integration

The challenge for any multi-site manufacturer evaluating AI- assisted cross-plant routing is that the existing workflow lives in entrenched tools. Sales engineers work in their existing email, CRM, or ERP. Production engineers work in their existing process documentation systems. Forcing them onto a new platform breaks the workflow and triggers adoption resistance.

At the Montanstahl Group, the integration was designed specifically to avoid this. The AI screening engine was embedded directly into the team's existing communication tools. When an RFQ arrived in the sales engineer's Outlook inbox, the AI's feasibility suggestion appeared in the same workflow - through a sidebar that sits inside the email client, not a separate application requiring login and navigation.

This integration approach drove 100% adoption from day one. Sales engineers did not have to learn a new tool. They saw the AI's suggestion in the same place they were already working. The friction that typically kills enterprise software rollouts was structurally eliminated.

For multi-site manufacturers evaluating similar deployments, the integration pattern matters more than the AI capabilities. An AI that answers feasibility questions correctly but lives in a separate tool that engineers must navigate to will not drive the workflow compression. The compression requires the AI to sit inside the existing workflow, surfacing answers where engineers already are.

Data Sovereignty in Multi-Site Deployments

Multi-site manufacturers, particularly in regulated industries, face data sovereignty constraints that public-cloud AI cannot always meet. European sites operate under GDPR. Aerospace tier-1s operate under ITAR or EAR. Defense suppliers operate under classification regimes.

The deployment pattern that works for multi-site manufacturers is one where the customer retains full ownership and control of their proprietary data. The AI processes data within the customer's security boundary - either on-premise, in a VPC, or in the customer's own cloud account. Nothing is used to train external models. Full regulatory compliance is preserved.

At the Montanstahl Group, full GDPR compliance was a requirement of the deployment. The AI's training and inference both happened within the customer's security boundary. The cross-plant knowledge unification did not require any data to leave the customer's control.

What to Demand From AI Vendors

If you are evaluating AI for cross-plant RFQ routing at a multi-site manufacturer, the questions that actually matter are not the typical vendor capabilities pitch.

Ask: "How does this integrate with our existing email, CRM, or ERP without forcing our engineers onto a new platform?" The adoption pattern matters more than the AI sophistication.

Ask: "What is your measured cross-site similarity matching accuracy on a sample of our historical data?" Not the vendor's reference customers. Yours. Real numbers on real data.

Ask: "What does deployment look like for our data sovereignty requirements?" On-premise, VPC, or customer-owned cloud should be options. If the vendor only offers public-cloud SaaS, they cannot serve multi-site manufacturers with regulated data.

Ask: "How does the system handle novel cases that AI cannot confidently answer?" The 15% that flag for human review needs to route to the right senior engineer with the right context - not just produce an "AI uncertain" flag.

The Industry Trajectory

Multi-site manufacturers are at a competitive disadvantage relative to single-site competitors specifically because of cross-plant coordination overhead. AI-assisted unified knowledge access is the structural intervention that closes that gap. The multi-site manufacturer with AI-mediated routing can respond as fast as a single-site competitor - and bring the combined capability advantage of multiple sites to the customer relationship.

Over the next 36 months, the multi-site manufacturers that integrate AI-assisted cross-plant routing will see their RFQ response time compress to single-site competitive levels. They will retain the multi-site advantage of broader capabilities. They will gain market share from competitors who cannot match their response speed.

The multi-site manufacturers that do not integrate this will continue to compete with a structural latency penalty that is invisible on their P&L but visible in their win rate against faster competitors. The penalty compounds over time as customers begin to expect faster responses across the entire market.

Cross-plant knowledge fragmentation has been the structural cost of multi-site manufacturing for decades. AI-assisted unified knowledge access is the first technology capable of eliminating it without consolidating sites. The companies that recognize this shift first will compound an advantage that, by the time the laggards catch up, will be very hard to close.