Case Study • Institutional Knowledge
The 31 years in one foreman's head, captured before he retires
- Client
- PavingX
- Service
- AI Solutions|Data Analytics
Overview
A field report on KnowledgeAI, the institutional-memory layer Sphere built for PavingX — a national pavement management and paving contractor — over 50+ years and 35,000+ archived jobs: turning three decades in one foreman's head into something the next estimator can actually use, and cutting estimate turnaround from 6.1 hours to 42 minutes.
The Situation
PavingX has 52 years of job history, 41 branches, and 614 team members — and a great deal of what makes an estimate accurate or a crack-repair judgment call correct lives in a relatively small number of long-tenured heads, not in any searchable system. KnowledgeAI is Sphere's answer: a search layer over 38,600 archived job files, 24,880 of them photo-linked, answering in an average 6.2 seconds what used to take a 35-minute manual file pull. This morning it's already saved 38.6 estimator hours and auto-drafted 12 estimates from matched precedent.
A 31-year foreman retires in 90 days
Decades of freeze-thaw crack-repair judgment leave with him, uncaptured.
Drafting an estimate for a new bid
Starts from a blank sheet, or whatever the estimator remembers — avg 6.1 hours.
A recurring site-failure pattern
Noticed job by job, warranty claim by claim, rarely connected across branches.
Our Solution
KnowledgeAI captures expert judgment before it retires, surfaces the closest historical comparable for every new bid, and connects recurring failure patterns across the full branch network.
1. Crew Expertise Directory
96 tenured foremen and estimators are profiled by expertise area, tenure, and retirement timeline — and 34% of the network's senior estimators are eligible to retire within five years. Top of the list: a senior foreman in Indianapolis, 31 years, retiring in 90 days, carrying freeze-thaw crack-repair judgment across 214 jobs that's largely uncaptured and protects an estimated $2.1M a year of Indianapolis volume. The desk doesn't just flag the risk — a 47-minute interview with him is already scheduled and transcribed this afternoon, capturing his crack-seal timing heuristics before the clock runs out. Three other senior estimators are next in the queue, each with a different kind of expertise and a different runway.
2. Estimator Copilot
Every draft this quarter — 312 of them, across 41 branches — cites the historical jobs it pulled from, not just a number. An AutoCo estimate for a new Aurora, CO lot auto-drafts from Job #41102, a 2022 AutoCo job, at a 96% spec-and-soil match, saving an estimator roughly 5.5 hours. On a Houston bid for a HomeCo distribution center, the copilot flags something a rushed estimator might miss: 11 of the 12 comparable Houston jobs included a storm-drainage line item this draft omits — caught before submission, not after a change order.
Why This Matters to PavingX's Bid Pipeline
Retiring expertise gets captured, not lost
Knowledge-capture interviews are prioritized by what's actually at risk — dollars, tenure, backup coverage — ahead of whoever happens to mention retiring first.
Estimates start from precedent, not a blank sheet
89% faster turnaround, with every draft citing the comparable jobs it drew from, so an estimator can sanity-check the AI's math rather than take it on faith.
Bid-oversight catches happen before submission
The same comparable-job matching that speeds up drafting also flags line items a rushed estimator might have missed.
Recurring site failures get connected across branches
A pattern that shows up once in Houston and once in Tampa gets clustered and addressed once, not fixed twice and warrantied a third time.
M&A integration extends to institutional knowledge, not just systems
Newly acquired branches get profiled into the same knowledge base, so a 62%-onboarded branch like Downingtown, PA shows up as a visible gap well before it becomes a blind spot.
Results
None of this replaces the estimator or the foreman — it gives the next one somewhere to start instead of a blank sheet, and gives the retiring one's judgment a way to survive the retirement. Rolled up over 12 months: 9,400 estimator hours reclaimed, roughly 4.5 FTE worth of time; $1.8M in warranty rework avoided by catching recurring site-failure patterns like the 14-site Gulf Coast clay cluster before the next claim; and 78% of the network's at-risk institutional knowledge captured on interview before it walks out the door.
