Order Assembly Intelligence
Predicting Multi-Phase Delivery Failures 5 Days Before They Happen
Industry
HVAC & Plumbing
Scale
$120M Revenue
Duration
20 Weeks
Location
Montréal, Quebec
Engagement
AI Consulting
Executive Summary
The Branch Operations Manager at a 5-branch HVAC and plumbing distributor in Montréal was watching the same pattern repeat on every commercial project: staged deliveries arrived at jobsites out of sequence, ahead of schedule, or missing components that the contractor needed for the current phase. 26% of commercial deliveries required rework — return trips, emergency shipments, or warehouse re-staging — because the coordination between the sales desk, the warehouse, and the contractor’s construction schedule broke down somewhere in a chain of phone calls, emails, and manual scheduling. We embedded predictive coordination intelligence into the distributor’s workflow on Infor SX.e.
Business Impact
91%
On-time phase-aligned delivery rate, up from 74%
67%
Reduction in coordination-driven delivery errors
$210K
Annual emergency freight cost eliminated
14%
Improvement in contractor satisfaction scores
The Situation
The distributor served mechanical contractors across Greater Montréal, Laval, the South Shore, and Québec City — commercial HVAC projects for hospitals, schools, office buildings, and multi-residential developments. Commercial work represented 58% of revenue and carried higher margins than residential counter sales, but only when deliveries arrived at the right jobsite, at the right construction phase, with the right material.
A commercial HVAC project didn't arrive as a single order. It arrived as a drawing that became a BOM, then a quote, then a purchase order, then 4-8 staged deliveries over 3-6 months — each delivery timed to a construction phase that shifted constantly.
Operational complexities:
- A typical commercial project generated 4-8 staged releases over the project lifecycle — rough-in piping and hangers first, then equipment (RTUs, air handlers, boilers), then trim and controls, then test and balance materials. Each release required different products pulled from different warehouse zones
- The sales desk scheduled releases based on the delivery dates the mechanical contractor provided at project kickoff — dates that were stale within weeks as the GC’s schedule shifted due to other trades, inspections, weather, and material delays on the structural side
- Contractors communicated schedule changes through a mix of phone calls to their sales rep, emails to the branch, and text messages that often didn’t reach the warehouse coordinator until after the original delivery was already being staged
- The warehouse team staged material 24-48 hours before the scheduled delivery. When a contractor’s phase slipped by 3 days after staging had begun, the material sat on the warehouse floor consuming space, or was loaded onto a truck and delivered to a jobsite that wasn’t ready — generating a return trip at $800-$1,200 per occurrence
- The 5 inside sales reps managing commercial accounts each carried 15-20 active projects with 4-8 open releases per project — 60-160 pending deliveries tracked through a combination of SX.e delivery dates, personal spreadsheets, and memory
- No system predicted which deliveries were likely to fail. The team reacted to failures after they occurred — scrambling to reroute trucks, re-stage material, and manage contractor frustration
The distributor had the product knowledge and inventory to serve commercial contractors well. What it couldn’t do was keep deliveries synchronized with construction schedules that changed faster than its manual coordination process could track.
The Challenge
The Branch Operations Manager described a $1.8M hospital mechanical project that illustrated the pattern. The contractor had provided a phase schedule at project award showing rough-in starting in March, equipment set in May, and trim in July. By April, the GC had pushed equipment set to June because the structural steel contractor was 3 weeks behind. The contractor’s project manager mentioned the delay in a phone call to the sales rep on April 4th. The sales rep noted it in his personal spreadsheet. The warehouse didn’t learn about it until May 2nd — when the staging team pulled $42K in RTUs and air handlers for a delivery the jobsite wasn’t ready to receive.
The return trip, re-staging, and eventual redelivery in June cost $3,400. But the real cost was the warehouse space consumed for 4 weeks by staged equipment that couldn’t ship, the 2 other projects whose staging was delayed because floor space was occupied, and the contractor’s comment to the branch manager: “Your guys showed up with equipment we told you wasn’t needed yet.”
- The sales rep had the information on April 4th. The warehouse didn’t get it until May 2nd. 28 days of signal delay between the person who knew and the person who needed to know
- Across all 5 branches and all active commercial projects, this pattern generated the 26% coordination failure rate — not from incompetence but from information moving too slowly through too many handoffs
- The Branch Operations Manager had tried a shared Google Sheet for commercial project tracking. Adoption lasted 2 months before reps stopped updating it because the manual entry duplicated what they were already doing in SX.e and their own tracking tools
The Solution
We spent 5 weeks in discovery across 3 branches — observing how inside sales reps managed commercial project releases, how the warehouse staged deliveries, and how schedule change information actually flowed (or failed to flow) between them.
The critical finding: 71% of delivery failures were predictable 3-7 days in advance. The signals existed — in contractor communications, in the pattern of which project types and which contractors historically experienced phase delays, in weather patterns that consistently pushed exterior work, and in the GC schedule changes that contractors mentioned in passing during routine calls. No system captured these signals or connected them to pending delivery schedules.
The signal that a delivery would fail almost always existed 5-7 days before the truck was loaded. It was in an email, in a weather pattern, in the contractor’s history of schedule accuracy. No human tracking 60-160 pending deliveries could process those signals. The ML could.
The system analyzed signals including:
- NLP analysis of contractor communications — emails, logged phone call notes, and text messages parsed for schedule change language (“pushed back,” “delayed,” “not ready until,” “GC is behind”) that indicated a pending delivery was at risk, flagged to the warehouse coordinator before staging began
- ML models trained on 3 years of delivery outcome data predicting which upcoming releases were most likely to fail based on project type, contractor history, construction phase, season, and the time elapsed since the contractor last confirmed the schedule
- Agentic workflows that automatically triggered contractor confirmation requests 5 days before each scheduled release — and escalated to the sales rep when the contractor didn’t respond within 48 hours, because non-response was itself the strongest predictor of a schedule change
- Weather-adjusted delivery risk scoring — correlating historical delivery failures with weather patterns by season and project type (exterior equipment sets failed at 3x the rate during Montréal’s November-March period)
- Warehouse staging optimization that learned which product combinations and delivery sequences generated the fewest errors for each project phase type — rough-in staged differently than equipment set, trim staged differently than controls
The signal that a delivery would fail almost always existed 5-7 days before the truck was loaded. It was in an email, in a weather pattern, in the contractor’s history of schedule accuracy. No human tracking 60-160 pending deliveries could process those signals. The ML could.
The Challenge
The Branch Operations Manager described a $1.8M hospital mechanical project that illustrated the pattern. The contractor had provided a phase schedule at project award showing rough-in starting in March, equipment set in May, and trim in July. By April, the GC had pushed equipment set to June because the structural steel contractor was 3 weeks behind. The contractor’s project manager mentioned the delay in a phone call to the sales rep on April 4th. The sales rep noted it in his personal spreadsheet. The warehouse didn’t learn about it until May 2nd — when the staging team pulled $42K in RTUs and air handlers for a delivery the jobsite wasn’t ready to receive.
The return trip, re-staging, and eventual redelivery in June cost $3,400. But the real cost was the warehouse space consumed for 4 weeks by staged equipment that couldn’t ship, the 2 other projects whose staging was delayed because floor space was occupied, and the contractor’s comment to the branch manager: “Your guys showed up with equipment we told you wasn’t needed yet.”
- The sales rep had the information on April 4th. The warehouse didn’t get it until May 2nd. 28 days of signal delay between the person who knew and the person who needed to know
- Across all 5 branches and all active commercial projects, this pattern generated the 26% coordination failure rate — not from incompetence but from information moving too slowly through too many handoffs
- The Branch Operations Manager had tried a shared Google Sheet for commercial project tracking. Adoption lasted 2 months before reps stopped updating it because the manual entry duplicated what they were already doing in SX.e and their own tracking tools
The Solution
We spent 5 weeks in discovery across 3 branches — observing how inside sales reps managed commercial project releases, how the warehouse staged deliveries, and how schedule change information actually flowed (or failed to flow) between them.
The critical finding: 71% of delivery failures were predictable 3-7 days in advance. The signals existed — in contractor communications, in the pattern of which project types and which contractors historically experienced phase delays, in weather patterns that consistently pushed exterior work, and in the GC schedule changes that contractors mentioned in passing during routine calls. No system captured these signals or connected them to pending delivery schedules.
The signal that a delivery would fail almost always existed 5-7 days before the truck was loaded. It was in an email, in a weather pattern, in the contractor’s history of schedule accuracy. No human tracking 60-160 pending deliveries could process those signals. The ML could.
The system analyzed signals including:
- NLP analysis of contractor communications — emails, logged phone call notes, and text messages parsed for schedule change language (“pushed back,” “delayed,” “not ready until,” “GC is behind”) that indicated a pending delivery was at risk, flagged to the warehouse coordinator before staging began
- ML models trained on 3 years of delivery outcome data predicting which upcoming releases were most likely to fail based on project type, contractor history, construction phase, season, and the time elapsed since the contractor last confirmed the schedule
- Agentic workflows that automatically triggered contractor confirmation requests 5 days before each scheduled release — and escalated to the sales rep when the contractor didn’t respond within 48 hours, because non-response was itself the strongest predictor of a schedule change
- Weather-adjusted delivery risk scoring — correlating historical delivery failures with weather patterns by season and project type (exterior equipment sets failed at 3x the rate during Montréal’s November-March period)
- Warehouse staging optimization that learned which product combinations and delivery sequences generated the fewest errors for each project phase type — rough-in staged differently than equipment set, trim staged differently than controls
The signal that a delivery would fail almost always existed 5-7 days before the truck was loaded. It was in an email, in a weather pattern, in the contractor’s history of schedule accuracy. No human tracking 60-160 pending deliveries could process those signals. The ML could.
Implementation
Deployment occurred over a Month 1-2 – Month 5-6 period.
Delivery Failure Prediction Engine
ML model scoring every pending release for failure probability based on contractor history, project phase, schedule age, weather, and communication patterns.
Schedule Signal Detection
NLP parsing contractor emails and logged call notes for language indicating phase delays, flagging at-risk deliveries to the warehouse coordinator before staging begins.
Proactive Confirmation Workflow
Agentic system triggering automated schedule confirmation requests 5 days before each release, with escalation logic when non-response indicates elevated risk.
Staging Sequence Intelligence
ML-optimized staging sequences by project phase type, reducing warehouse picking errors and ensuring deliveries are loaded for the correct installation sequence.
SX.e Integration
Prediction scores and schedule risk flags surfaced within the existing delivery scheduling workflow — the warehouse coordinator sees risk before committing staging labor.
Delivery Failure Prediction Engine
ML model scoring every pending release for failure probability based on contractor history, project phase, schedule age, weather, and communication patterns.
Schedule Signal Detection
NLP parsing contractor emails and logged call notes for language indicating phase delays, flagging at-risk deliveries to the warehouse coordinator before staging begins.
Proactive Confirmation Workflow
Agentic system triggering automated schedule confirmation requests 5 days before each release, with escalation logic when non-response indicates elevated risk.
Staging Sequence Intelligence
ML-optimized staging sequences by project phase type, reducing warehouse picking errors and ensuring deliveries are loaded for the correct installation sequence.
SX.e Integration
Prediction scores and schedule risk flags surfaced within the existing delivery scheduling workflow — the warehouse coordinator sees risk before committing staging labor.
Strategic Impact
Delivery Reliability as Competitive Advantage
On-time phase-aligned delivery rate rose from 74% to 91% across active commercial projects. Three mechanical contractors independently cited delivery reliability as the reason they consolidated purchasing with the distributor — shifting volume from competitors who were still operating on manual coordination and reactive rescheduling.
Emergency Freight Elimination
$210K in annual emergency shipment costs eliminated — the expedited deliveries, return trips, and re-staging labor that occurred every time a delivery failed coordination. The Branch Operations Manager noted that the savings understated the real impact: “Every emergency shipment also disrupted the next day’s planned deliveries because it pulled a truck and a driver off schedule.”
Warehouse Productivity Recovery
The warehouse staging team went from spending roughly 30% of their time on rework — restaging material that shipped to jobsites that weren’t ready, breaking down loads that were staged for deliveries that got rescheduled — to spending that time on planned work. No headcount was added. Throughput increased because the team stopped doing the same work twice.
Key Takeaway
In commercial HVAC distribution, the delivery isn’t a logistics problem — it’s an intelligence problem. The material is in the warehouse. The truck is available. The crew is ready. The question is whether the jobsite is actually at the construction phase that needs what you’re about to send — and the answer to that question changes every day on every active project. This distributor didn’t need a better warehouse management system or a more sophisticated dispatch platform. It needed the signals that predicted delivery failures — contractor communications, schedule confirmation patterns, weather data, historical accuracy by project type — captured, processed, and connected to the staging decision before the truck was loaded. The 26% failure rate didn’t come from bad execution. It came from good execution against bad information.
Related Engagements
Contractor Pricing Intelligence
The VP of Commercial Sales at a 10-branch HVAC and plumbing distributor in Nashville discovered that the same 3/4" ProPress fitting sold for 11.4% less in Chattanooga than in Nashville — to comparable...
Seasonal Demand Intelligence
The Director of Supply Chain at a 6-branch HVAC and plumbing distributor in Portland was trapped in the same annual cycle: over-purchase heating equipment and hydronic products ahead of the October-Ma...
Mechanical Takeoff Intelligence
The Commercial Sales Manager at an 8-branch HVAC and plumbing distributor in Denver had a 6-person estimating team declining 30-40% of commercial RFQs during peak season because each mechanical takeof...