Case Study • Post-Merger Data Unification
$680k a year was hiding in two different markup formulas
- Client
- PavingX
- Service
- AI Solutions|Data Analytics
Overview
A field report on Unify, the governed semantic data layer Sphere built for PavingX — a national pavement management and paving contractor — over its post-merger system landscape: several source systems, multiple merged entities, and dozens of branches, unified into one queryable layer with source-system access control enforced on every request.
The Situation
PavingX's 2024 merger and three acquisitions didn't just bring four new AP systems into the company — they brought two ERPs, two CRMs, and two separate estimating systems, each running a different branch's version of “how big is this deal” and “what's our margin.” Unify is Sphere's governed semantic layer over all of them: every mapped branch on one margin definition, most source-system connectors live (Foundation's connector is degraded, traced to a stale credential from the Unigrade handoff), syncing hourly. This morning it merged 3 more account records, normalized 212 more estimates onto one cost-code taxonomy, and answered 214 governed queries without a single access-control exception.
WarehouseCo's account, across three systems
Three separate account records — one Salesforce, two HubSpot — no single point of contact possible.
Comparing margins across multiple merged companies
Impossible — each entity's estimating system computed margin differently.
Who can see what in the unified layer
A single combined database would mean everyone sees everything, merger or not.
Our Solution
Unify resolves duplicate records, normalizes cost definitions across legacy systems, and enforces the same access-control boundaries the original systems had — on every query.
1. Unified Customer 360
Until today, WarehouseCo existed as three different account records — one national record in Salesforce, two regional legacy records in HubSpot dating back to the Meridian acquisition — which meant no one could commit to a single point of contact for the account without checking which system they were looking at. The desk resolves all three into one golden record at 96% confidence. Across the full customer base, 612 duplicate records have become 187 golden records at 94% average match confidence; 46 remain in manual review below the auto-merge threshold, 12 of them already past the 5-business-day SLA and queued for a data steward.
2. Job Costing Reconciliation
The deeper problem was harder to see: HeavyBid (PavingX's legacy estimating system) and B2W Estimate (Meridian's) compute overhead and material markup differently, and 14,000+ estimates were sitting on two incompatible cost models. Normalizing them onto one taxonomy surfaced $2.4M in variance. The single biggest finding: B2W's legacy overhead-allocation formula under-recovers by 4.2% against the HeavyBid standard, silently affecting every one of the 13 Meridian-origin branches. Standardizing the formula recovers an estimated $680k a year — not a hypothetical, a formula bug with a dollar amount attached.
Why This Matters to a Multi-Company Merger
A merger's true margin gets visible, not averaged away
41 branches on one margin definition means the 6.8-point spread is a real number, and the part of it that's a costing artifact versus an operating problem is now distinguishable.
National accounts stop looking like multiple customers
WarehouseCo, PharmacyCo, and HomeCo each resolve to one golden record instead of splitting across the systems their acquired branches happened to use.
A markup bug gets a dollar figure and an owner
$680k/yr isn't a rounding error once it's traced to a specific formula in a specific system affecting a specific 13 branches.
Access control survives the merger
Nobody gains visibility they didn't already have in their original system; the unified layer enforces the same boundaries Salesforce, HubSpot, Viewpoint, and Foundation already had.
The next acquisition has a template
The same six-system pattern (2 ERPs, 2 CRMs, 2 estimating systems) is exactly what the next acquisition will bring — this is now a repeatable playbook, not a one-time cleanup.
Results
Put together, that's what the branch-performance view finally shows honestly: a 6.8-point margin gap between the network's best and worst quartile branches, most of it real, but roughly 3.5 points of the Meridian-origin gap turns out to be a costing-methodology artifact, not an operating difference — Charlotte, Hartford, and Tampa are the clearest examples. Even after correcting for that, three branches still trail the network by 4+ points on a like-for-like basis, which is now a flagged operational question instead of a number nobody trusted. All of it stays governed: 8,240 queries this month, filtered by each user's original source-system permissions, zero exceptions in 90 days — a foundation the SOC 2 review can be built on, not just a faster spreadsheet.
