Core & Main

Underwriting conclusion

A healthy distributor with a coordination question

Core & Main is a specialty distributor of water, wastewater, storm drainage, fire protection, and smart utility products. The company connects more than 5,000 suppliers to more than 60,000 customers through over 370 branches. Its catalog exceeds 225,000 products, many governed by municipal, state, federal, or engineering specifications.S1

[FACT] Fiscal 2025 net sales were $7.647 billion, up 2.8%. Gross profit was $2.059 billion at a 26.9% margin, operating income was $722 million, net income attributable to Core & Main was $441 million, and operating cash flow was $650 million.S1 [FACT] Fiscal Q2 2026 net sales rose 2.5% to $2.145 billion. Gross profit was $573 million at a 26.7% margin, operating income was $227 million, and net income was $150 million.S2S4

This is not a turnaround. Municipal demand, fire protection, large treatment-plant projects, greenfield branches, and acquisitions are supporting growth. Management reaffirmed fiscal 2026 guidance of $7.8 billion to $7.9 billion in net sales and $950 million to $980 million in adjusted EBITDA.S3S4

The underwriting question is narrower:

Where could Core & Main make specification-bound quotes, orders, and acquired data easier to assemble and inspect without replacing local judgment or duplicating the systems already in place?

Local expertise creates the advantage and the scaling pressure

Core & Main says availability, convenience, reliability, technical expertise, and project management all influence customer purchasing. Branch managers tailor local strategy and offerings, while nearly all branches use an integrated technology platform for procurement, inventory, finance, analytics, and performance reporting.S1

That combination is the operating advantage. It also creates repeated coordination surfaces:

1. A bid can combine plans, specifications, addenda, material lists, supplier pricing, availability, approved brands, alternates, lead times, and product documents.
2. An order can cross customer requirements, branch inventory, supplier acknowledgements, delivery dates, signed tickets, and invoice records.
3. A new branch or acquisition adds item, supplier, customer, pricing, identity, and control records that must become comparable without erasing local knowledge.

The public record proves that these records and roles exist. It does not prove that Core & Main handles them poorly.

The call

The highest publicly measurable pool is inventory: fiscal 2025 carried $986 million of inventory, and inventory reached $1.137 billion at August 2, 2026.S1S2 The best first proof is quote and submittal readiness. It is smaller, but the inputs, accountable reviewer, consequential decisions, and acceptance criteria can be bounded without changing live inventory or customer commitments.

What the first proof should establish

Select one treatment-plant quote team and 5–10 completed bid packages. Include an addendum, an approved-equal decision, a no-substitution item, an incomplete material list, and a package with multiple supplier documents. The designated quote specialist defines the required output before any build begins.

TruePulse tests whether software can:

• identify every required line item, specification, addendum, and document;
• connect each extracted requirement to its source page;
• compare the requirement with the proposed product and supplier evidence;
• flag missing, conflicting, stale, or uncertain information;
• assemble a draft review packet without choosing a product, price, or substitution.

The proof passes only if the specialist judges at least 8 of 10 cases ready for specialist review without rebuilding the packet, all supplied addenda and no-substitution constraints are either reflected or visibly flagged, no unsupported specification or product fact appears, and median preparation-plus-review time falls at least 30% against the historical baseline. If the existing process already clears that standard, Core & Main should not add another layer.

What the reader should leave with

This report gives an operating leader:

1. a financial map of how distribution scale becomes gross profit and cash;
2. a workflow map of where specifications, product evidence, availability, and approvals meet;
3. three bounded opportunities with separate economic and technical confidence;
4. a first proof that can fail without touching a live quote or order; and
5. the diligence questions needed to replace outside-in assumptions with operating facts.

It also shows the TruePulse method: start with the economics, follow the work, compare AI with simpler controls, and make the first test inspectable.

Why the thesis may be wrong

Core & Main may already have the required capability in its integrated platform, Online Advantage, Microsoft Dynamics 365 work, supplier tools, branch procedures, or undisclosed internal software.S1S8S11 The company is profitable, gaining share, opening branches, and reporting successful integrations. A new layer could add duplicate work, create stale copies of product data, or weaken the local accountability that customers value. Public job postings describe individual roles and point-in-time needs; they do not establish a company-wide defect.

1. Why this underwrite exists

This underwrite is for a Core & Main operating, finance, or technology leader deciding whether one AI-enabled workflow deserves capital and management attention.

The decision is not whether AI could parse a specification or summarize an exception. The decision is whether a bounded intervention can change a measured operating result while preserving accountable product, pricing, inventory, and customer decisions.

The company-specific question is:

Can Core & Main reduce the work required to turn project documents and operating exceptions into review-ready decisions, or do its current systems and branch practices already solve the problem at lower risk?

Evidence scope

The evidence cutoff is September 19, 2026. Sources include Core & Main filings, earnings materials, company service and portal pages, and official job postings. No management interview, customer record, supplier file, quote package, order history, or system access was available.

Research and drafting used AI-assisted source discovery, extraction, calculation, and synthesis. Material claims cite the primary source used. Official job postings are point-in-time evidence and may disappear after a role closes. One public bid page contains both a 24-hour turnaround statement and an unfilled [X–Y hours] placeholder; this report uses the page only as workflow evidence, not as proof of a service-level commitment.S7

Five labels separate evidence from analysis:

• [FACT] Direct disclosure from a cited source
• [CALCULATION] Reproducible math from cited facts
• [ASSUMPTION] An editable input unavailable publicly
• [HYPOTHESIS] A falsifiable operating explanation
• [UNKNOWN] Information that cannot responsibly be inferred

Analytical boundaries:

• No stock recommendation or price target
• No claim that a job posting describes every branch
• No assertion that a public workflow is defective
• No conversion of capacity into savings unless cash expense falls
• No presentation of working-capital release as EBITDA
• No addition of overlapping revenue, capacity, cash, and risk sensitivities
• No recommendation for autonomous product, price, supplier, credit, or inventory decisions

2. The business is winning, but growth raises the cost of coordination

2.1 Industry structure

Core & Main distributes essential but specification-sensitive products into municipal, non-residential, and residential infrastructure work. The company estimates its United States and Canada addressable market at $44 billion and says it is one of two national distributors in a fragmented field.S1S3

The distributor earns its position between thousands of manufacturers and tens of thousands of municipalities, utilities, contractors, and other customers. Breadth alone is not enough. Product availability, local specifications, technical advice, delivery, and project support all affect the purchase.S1

Several structural conditions matter to this underwrite:

• Products often must meet engineering and government specifications.S1
• Demand spans new construction and repair or replacement in a near-equal mix.S1
• Municipal work is supported by water and wastewater infrastructure spending, while residential and non-residential work remains cyclical.S1S3
• Weather can move construction, repair, and delivery activity between quarters.S1
• Supplier availability, raw-material conditions, transportation, tariffs, and product terms can change price and lead time.S1
• Local service matters because requirements, competitors, customers, and branch inventory differ by market.S1

These conditions make current, traceable evidence valuable. They also make confident automation dangerous.

2.2 Business model

Core & Main makes money by sourcing specialized products, holding or locating inventory, advising customers, pricing and winning orders, coordinating projects, and delivering the right material under the required terms.

[FACT] Fiscal 2025 product sales were:S1

[CALCULATION] Shares equal category sales divided by $7.647 billion. Product categories sum to the reported total.S1

[FACT] Management estimates that 44% of fiscal 2025 sales came from municipal customers, 38% from non-residential markets, and 18% from residential markets. It also estimates a near-equal split between new construction and repair or replacement.S1

This mix matters. Municipal and repair work can provide steadier demand, while project specifications and local requirements increase the cost of an incorrect product or late clarification. Construction exposure adds schedule and weather sensitivity.

2.3 Financial profile

Sources: fiscal 2025 Form 10-K, fiscal Q2 2026 Form 10-Q, and fiscal Q2 2026 earnings release.S1S2S4

[CALCULATION] Fiscal 2025 gross margin was $2.059 billion divided by $7.647 billion, or 26.9%. Operating margin was $722 million divided by $7.647 billion, or 9.4%.S1

[FACT] Fiscal 2025 sales rose 2.8%, gross profit rose 4.0%, operating income rose 0.4%, and adjusted EBITDA was $931 million compared with $930 million in fiscal 2024.S1 [FACT] Q2 2026 sales rose 2.5%, gross profit rose 2.3%, operating income rose 6.6%, and adjusted EBITDA rose 3.0%.S2S4

The condition is healthy, but the operating question has changed. At low-single-digit sales growth, better coordination and mix can matter more than a volume wave. That does not prove a software gap. It explains why small execution gains deserve measurement rather than assumption.

2.4 Operating model

[FACT] At February 1, 2026, Core & Main had over 370 branches, about 5,600 associates, more than 60,000 customers, more than 5,000 suppliers, and more than 225,000 products.S1S3

Branch managers tailor local strategy, marketing, and product and service offerings. District, regional, and company resources support them. Nearly all branches operate on an integrated technology platform used for procurement, inventory management, finance, analytics, and reporting.S1

Online Advantage already gives customers pricing and availability, job summaries, invoices and statements, signed tickets, delivery photos and GPS coordinates, online payment, and support for customer part numbers.S8

[FACT] The fiscal Q2 2026 presentation reported seven greenfield openings in the first half and a plan for eight to ten openings for the full year. It also reported 43 acquisitions since 2017 and described its acquisition pipeline as active.S3

A current job posting for a Microsoft Dynamics 365 Finance and Operations solution architect shows ongoing ERP design work with business-process owners, information-systems teams, technical architects, and ERP product owners.S11 [UNKNOWN] The deployment scope, timeline, modules, branch coverage, and relation to legacy systems.

The evidence points to a technology-enabled operator, not a blank slate. New work should sit inside the existing operating model or stop.

2.5 The public record shows repeated document and exception work

The strongest ground-level evidence comes from Core & Main's own service descriptions and roles:

• Treatment-plant services include customized specifications, takeoffs, pipe-layout drawings, value analysis, project management, quotes, procurement, and delivery coordination.S6
• The Supply bids page accepts a bill of materials, takeoff, spreadsheet, PDF, or screenshot; asks for part, quantity, brand, no-substitution, location, date, and configuration details; and describes scope confirmation, pricing, availability, alternatives, order conversion, and fulfillment.S7
• A Treatment Plant Quote Specialist II posting describes RFQ management, formal quotations, submittals, addenda, pricing coordination, bid deadlines, project logs, and customer questions.S9
• A Purchasing Coordinator posting describes purchase-order processing, specification and price checks, vendor acknowledgements, delivery dates, and discrepancy resolution. One posting names Mincron experience as preferred.S10
• Online Advantage exposes transaction evidence to customers, including pricing, availability, signed tickets, invoices, and delivery photos.S8

These sources establish real workflows and decision points. They do not establish their frequency, cost, error rate, or consistency across branches.

3. Follow the economics before choosing the technology

3.1 Revenue and gross profit

Distribution revenue
= addressable customer demand
× share of requests reached
× quote or order conversion
× average order value
× fulfillment rate
× price realization
− cancellations and credits

Gross profit
= revenue
− product cost
− inbound freight and related cost of sales
− product and fulfillment leakage included in cost of sales

The later opportunities may affect quote capacity, conversion, fill rate, price accuracy, rework, or credits. They do not get credit for all of those at once.

Publicly visible drivers:

• Sales and gross profit by period
• Product-category sales
• End-market mix
• Branch, supplier, customer, and product counts
• Greenfield and acquisition activity

Material unknowns:

• Quote volume, win rate, and average quoted value
• Margin by quote type, customer, product, and branch
• Lost orders caused by incomplete or late packages
• Requote and submittal-rework rate
• Credit, return, and cancellation causes
• Existing branch and enterprise automation

3.2 Inventory and fulfillment

Fill-rate revenue
= demand for in-scope products
× available-to-promise accuracy
× fulfillment rate
× retained order value

Inventory investment
= unit cost
× quantity held
× time held
+ transfer and handling cost
+ obsolescence and damage risk

[CALCULATION] Fiscal 2025 cost of sales was $5.588 billion. Average annual cost of sales per day was therefore about $15.3 million: $5.588 billion ÷ 365.S1

[CALCULATION] Ending inventory of $986 million represented about 64.4 days of fiscal 2025 cost of sales: $986 million ÷ $5.588 billion × 365.S1 This is a rough period-end ratio, not management's disclosed inventory-days metric and not a diagnosis of excess stock.

A one-day enterprise change would equal about $15.3 million of cash timing at fiscal 2025 cost levels. No opportunity receives that value by default. The proof must define the eligible product and branch cohort, distinguish seasonal build from avoidable inventory, and show that service levels do not fall.

3.3 Quote and submittal capacity

Quote capacity
= available specialist hours
÷ preparation and review hours per package

Incremental gross profit from better conversion
= annual quoted value
× addressable document-heavy share
× verified conversion change
× gross margin on won work
− added operating cost

Recovered time is capacity. It becomes profit only if the company handles more valuable work, avoids external cost, or removes cash expense.

3.4 Cash timing

Operating cash
= operating earnings
− cash taxes and interest
− change in receivables
− change in inventory
+ change in payables
± other working capital

Faster order resolution or lower inventory may release cash. Neither automatically increases EBITDA. Q2 and H1 balances are seasonal; annual and interim figures should not be mixed into a false trend.S1S2

3.5 Acquisition value

Acquisition value
= acquired gross profit and cash generation
+ cross-branch growth
+ purchasing and inventory benefits
− purchase price and financing cost
− integration cost
− customer or talent loss
− control and data risk

The Q2 presentation reports 43 deals since 2017. Fiscal 2025 included two acquisitions adding five locations with $76 million of aggregate transaction value.S1S3 The company also opened branches organically.

[HYPOTHESIS] Faster record mapping could reduce integration capacity and make shared inventory or reporting available sooner. [UNKNOWN] Whether record mapping is a binding integration constraint, how it is performed today, and whether any measurable delay remains after current ERP work.

4. Follow the work from request to cash

4.1 Bid to approved quote

The critical control is provenance. A reviewer must see which page, addendum, supplier record, and product document supports each output.

4.2 Approved quote to fulfilled order and cash

A useful intervention would not predict demand in the abstract. It would show a buyer or branch operator the small set of orders where existing evidence conflicts, is stale, or requires a choice.

4.3 Demand to inventory deployment

4.4 Target to integrated acquisition or greenfield

4.5 Requirement to smart-utility project evidence

Core & Main's meter and smart utility work can include hardware, network infrastructure, software installation, training, and long-term service contracts.S1 This chain may carry meaningful lifecycle and compliance records.

It is not selected for deep underwriting because public evidence does not show a specific repeated bottleneck, accountable record owner, or acceptance baseline. A generic proposal to summarize utility documents would be too loose.

5. Value-pool map

5.1 Company-wide map

5.2 Priority ranking by separate dimensions

The first proof favors attribution over theoretical value. Inventory may be larger, but a quote packet can be judged without changing stock, supplier commitments, or customer terms.

6. Priority opportunity underwrites

Opportunity 1: Quote, addendum, and submittal readiness

Thesis

[HYPOTHESIS] For document-heavy treatment-plant bids, a source-cited requirement and evidence packet can reduce preparation and review time without increasing unflagged specification, addendum, or substitution errors.

Financial driver

• Additional quote capacity if recovered time is used on valuable work
• Incremental revenue and gross profit only if measured conversion improves
• Avoided rework and schedule risk

Capacity is not counted as savings. Win-rate value is not counted until a comparable baseline exists.

Public evidence

[FACT] Core & Main advertises customized specifications, accurate takeoffs, pipe-layout drawings, value analysis, project management, competitive quotes, procurement, and delivery coordination for treatment-plant work.S6

[FACT] Its Supply bid process accepts BOMs, takeoffs, spreadsheets, PDFs, and screenshots. It requests product, quantity, brand, no-substitution, destination, date, and configuration details, then confirms scope before pricing, alternatives, ordering, and fulfillment.S7

[FACT] A Quote Specialist II role manages RFQs, quotations, submittals, addenda, supplier and internal pricing, deadlines, logs, and customer questions.S9

Current-state hypothesis

[HYPOTHESIS] A specialist assembles project requirements from heterogeneous documents, coordinates supplier evidence and price, and revises the package when addenda or availability changes.

[UNKNOWN] Whether the work is mostly manual; which quote, document, product, and supplier systems are used; how often packages are incomplete; whether a native tool already compares versions and assembles submittals.

Workflow contract

• Trigger: A treatment-plant RFQ or addendum arrives.
• Inputs: Plans, specifications, BOM or takeoff, addenda, customer rules, supplier quotes, price and availability, approved product documents.
• Required output: A review-ready requirement matrix and draft quote/submittal evidence packet with source citations and visible exceptions.
• Accountable owner: Designated Treatment Plant Quote Specialist II or equivalent team lead.
• Human decision: Product, approved equal, price, margin, supplier, freight, lead time, terms, and customer submission.

Intervention alternatives

1. Process change: Standard intake form, naming convention, addendum checklist, and owner matrix. Use this if missing inputs cause most delay.
2. Source-data improvement: Require structured BOMs and supplier document identifiers. Use this if extraction is not the problem.
3. Native configuration: Configure the current quote, ERP, product-information, or document system to require fields and compare versions. Prefer this if available.
4. Deterministic rules: Validate mandatory fields, dates, part identifiers, and package contents.
5. AI extraction and retrieval: Extract requirements from plans and specifications, retrieve approved product evidence, compare versions, and cite sources.
6. Copilot: Draft the packet and exception list for specialist review.
7. Agent or autonomy: Not recommended. The system should not choose products, accept substitutions, set price, or send a bid.

Recommended reference architecture

Use an offline evidence-assembly pipeline for the proof:

Historical RFQ + specifications + addenda + BOM
→ document and table extraction
→ deterministic required-field and version checks
→ retrieval from an approved product-document set
→ line-level requirement/evidence matrix
→ conflict, missing-data, and uncertainty queue
→ draft packet
→ specialist review and score

Every output points to the source document and page. The system abstains when product identity, revision, approval status, price, or availability is uncertain. A failed extraction remains visible rather than being filled with a plausible answer.

Required data and access

• 5–10 completed bid packages, including exceptions
• Original RFQ, plans, specifications, all addenda, BOM or takeoff
• Supplier pricing and availability evidence as it existed at the decision date
• Product data, certifications, drawings, and approved-equal evidence
• Final quote and submittal packet
• Revision history and known rework
• Review time and preparation time if available
• Read-only access for proof; no live customer or ERP write access

Value equation and sensitivity

Use the fiscal 2025 gross margin of 26.9% only as an outside-in placeholder. Actual quoted-work margin must replace it.S1

Incremental gross profit
= annual quoted value
× addressable document-heavy share
× verified conversion change
× 26.9% placeholder gross margin
− annual operating cost

All non-margin inputs are [ASSUMPTION]. The table is a threshold test, not a forecast. If a team cannot support the quoted-value and conversion inputs, do not use the output.

Capacity is separate:

Annual capacity value
= eligible packages
× hours removed per package
× loaded hourly cost

Do not add capacity value to gross profit if the same recovered time causes the modeled conversion change. Do not call capacity savings unless cash cost falls.

Proof design

• Examples: 5–10 historical packages, including one addendum, one no-sub item, one approved-equal decision, one incomplete BOM, and one multi-supplier packet
• Hidden cases: 2–3 unseen completed packages if volume permits
• Expected output: Requirement matrix, evidence links, version comparison, missing/conflict list, and draft packet
• Judge: One designated quote specialist, with a branch or treatment-plant sales leader resolving disputed criteria
• Baseline: Preparation time, review time, rework count, missed requirement count, and specialist readiness rating on the same cases
• Pass: At least 8 of 10 cases ready for specialist review without reassembly; 100% of supplied addenda and no-sub constraints represented or flagged; zero unsupported product/specification facts; at least 30% lower median preparation-plus-review time
• Unacceptable errors: Invented specification; stale or wrong product revision; silent substitution; omitted addendum; unsupported price or availability; customer submission without approval
• Human boundary: The specialist approves every product, price, supplier, exception resolution, and outbound package

Production controls

• Read access only to approved project and product sources
• Source document, page, revision, and retrieval time on each extracted fact
• Immutable evaluation and decision log
• Version comparison before every draft refresh
• Confidence threshold and mandatory abstention
• Separate queues for missing evidence, conflicts, and business judgment
• No customer send, price update, ERP write, or supplier commitment without approval
• Monitoring for document drift, extraction failure, and reviewer overrides
• Exportable requirement matrix and packet; no lock-in to a proprietary record format

Kill criteria

Stop if:

• native software already meets the acceptance test;
• fewer than 20% of historical cases require material reassembly or review work;
• source documents cannot be versioned or linked reliably;
• any hidden case contains an unflagged no-substitution, addendum, or material specification error;
• preparation and review time fall by less than 30%;
• annual volume and economic value do not cover operation and change management;
• specialists must recreate the packet to trust it.

Confidence

Opportunity 2: Inventory and order exception briefs

Thesis

[HYPOTHESIS] For one product cohort and branch group, a deterministic exception layer with source-cited summaries can help buyers and branch operators resolve late, conflicting, or incomplete supply signals while preserving existing inventory and purchasing decisions.

Financial driver

• Working-capital release from verified inventory-day reduction
• Retained revenue and gross profit from better fill rates
• Lower transfer, expedite, and coordination burden

Working-capital release, revenue, gross profit, and capacity remain separate.

Public evidence

[FACT] Fiscal 2025 inventory was $986 million against $5.588 billion of cost of sales. Inventory was $1.137 billion at August 2, 2026.S1S2

[FACT] Nearly all branches use an integrated platform for procurement and inventory management.S1 [FACT] A Purchasing Coordinator role checks purchase-order specifications and pricing, vendor acknowledgements and delivery dates, and order discrepancies.S10 [FACT] Online Advantage exposes pricing, availability, signed tickets, delivery evidence, invoices, and customer part numbers.S8

Current-state hypothesis

[HYPOTHESIS] Buyers and branch operators already receive structured order and inventory data but may still reconcile exceptions across customer requirements, supplier acknowledgements, promised dates, transfers, and delivery evidence.

[UNKNOWN] Whether the existing platform already prioritizes these exceptions; which systems are authoritative; the current inventory policy; fill rate; transfer and expedite volume; forecast accuracy; obsolete inventory; and supplier-data latency.

Workflow contract

• Trigger: A sales order, PO, transfer, supplier acknowledgement, delivery date, or inventory status conflicts with the expected state.
• Inputs: Order lines, product and customer identifiers, stock and commitments, PO, acknowledgement, lead time, promised date, transfer, delivery, and invoice state.
• Required output: A ranked exception brief stating the evidence, conflict, age, financial exposure, and permitted next action.
• Accountable owner: Designated purchasing or branch operations lead.
• Human decision: Buy, transfer, allocate, substitute, expedite, cancel, credit, or change a customer promise.

Intervention alternatives

1. Fix supplier acknowledgement discipline and exception ownership.
2. Improve product, lead-time, and customer-part data at source.
3. Configure native ERP alerts and work queues.
4. Use deterministic rules for missing acknowledgements, date conflicts, stale status, and duplicate commitments.
5. Add AI only to summarize evidence and retrieve applicable customer or product constraints.
6. Do not allow autonomous inventory movement or substitution.

Recommended reference architecture

The proof should join read-only snapshots from the authoritative order, inventory, purchasing, and delivery sources. Deterministic rules identify candidate exceptions. Retrieval supplies the customer, product, or project requirement. AI may write the short evidence brief, but it does not create the exception and cannot execute an action.

If one system already contains the complete exception and supporting evidence, configure that system instead.

Required data and access

• 5–10 resolved exceptions from one product cohort and branch group
• Sales order, PO, acknowledgement, inventory, transfer, delivery, and invoice snapshots
• Current exception codes and ownership rules
• Historical resolution, elapsed time, touches, and outcome
• Product and customer constraints
• Read-only access for the proof

Value equation and sensitivity

One day of enterprise inventory at fiscal 2025 cost
= $5.588B cost of sales ÷ 365
= $15.3M

Cohort working-capital release
= $15.3M
× eligible inventory share
× verified days removed

Eligible share and days removed are [ASSUMPTION]. The calculation is working capital, not EBITDA. It ignores service-level loss, operating cost, tax, financing, and any inventory moved elsewhere. A proof must measure the selected cohort, not claim enterprise value from this table.

Proof design

• Examples: 5–10 historical late, conflicting, or incomplete order/PO cases, including a transfer, supplier date change, constrained item, and customer-specific requirement
• Hidden cases: 2–3 unseen exceptions
• Expected output: Evidence brief, exception classification, age, owner, permitted next action, and source links
• Judge: One purchasing or branch operations lead
• Baseline: Time to recognize, touches to resolve, aged-exception count, service outcome, and inventory days for the cohort
• Pass: All true priority exceptions surfaced; no false instruction to move or substitute product; at least 30% lower median recognition-plus-triage time; judge accepts at least 8 of 10 briefs without reconstructing evidence
• Unacceptable errors: Wrong product identity, stale availability, hidden customer constraint, unauthorized substitution, double allocation, or automatic transaction
• Human boundary: Every buy, transfer, allocation, substitution, customer promise, and write-off remains human-approved

Production controls

• Source-of-truth declaration for each field
• Freshness timestamp and stale-data rejection
• Deterministic exception rule versioning
• Customer and product constraint retrieval with citations
• Read-only first release; dual control for later write actions
• Full action and override log
• Service-level guardrail beside inventory target
• Separate reporting for cash, margin, capacity, and customer outcomes

Kill criteria

Stop if native ERP alerts already meet the test, data freshness cannot support the decision, the exception definition changes by branch faster than it can be governed, false positives exceed the team's review capacity, or inventory-day improvement reduces fill rate or moves stock elsewhere rather than releasing cash.

Confidence

Opportunity 3: Acquisition master-data mapping

Thesis

[HYPOTHESIS] A deterministic-first mapping workbench can reduce the time required to reconcile one acquired item, supplier, or customer record family while keeping record ownership and cutover approval with Core & Main.

Financial driver

• Integration capacity
• Earlier shared visibility or cross-branch service if mapping is on the critical path
• Avoided control and pricing risk

No EBITDA value is assigned without proof that mapping delays a measured business outcome or removes cash cost.

Public evidence

[FACT] Core & Main reported 43 acquisitions since 2017 in its Q2 2026 presentation. Fiscal 2025 included two acquisitions, five locations, and $76 million of aggregate transaction value.S1S3 [FACT] Seven greenfield locations opened in the first half of fiscal 2026.S3

[FACT] Nearly all branches operate on an integrated technology platform.S1 A current Dynamics 365 architecture role works across business-process and technology owners.S11

Current-state hypothesis

[HYPOTHESIS] Each acquisition or greenfield requires some combination of record extraction, matching, approval, loading, and reconciliation.

[UNKNOWN] Whether master-data mapping is centralized, the source systems, record counts, match rates, ownership, defect rate, cutover timeline, and the role of current Dynamics 365 work.

Workflow contract

• Trigger: Acquisition close or greenfield setup reaches a named record-migration stage.
• Inputs: Source and target item, supplier, customer, location, pricing, and control records plus mapping policy.
• Required output: Proposed exact matches, candidates, conflicts, unmapped records, rationale, and source evidence.
• Accountable owner: Named data-domain owner and integration lead.
• Human decision: Canonical record, merge, new record, price or term, access, accounting treatment, and cutover.

Intervention alternatives

1. Standardize acquisition data requests and templates.
2. Clean and deduplicate at source.
3. Use native Dynamics 365 data-management and matching functions.
4. Use deterministic exact, normalized, and reference-key matching.
5. Use AI only for low-confidence text normalization and candidate explanation.
6. Do not allow autonomous merge, delete, price, customer, supplier, or financial mapping.

Recommended reference architecture

Versioned source extract + approved target master
→ field normalization
→ exact and deterministic reference matching
→ candidate retrieval for unresolved records
→ confidence and conflict classification
→ owner approval queue
→ approved export
→ reconciliation against source counts and control totals

The workbench should be disposable if native ERP tooling performs as well. It must export approved mappings and an audit log in a standard format.

Required data and access

• One historical acquired record family
• Source and target extracts at the original mapping date
• Data dictionary and mapping policy
• Approved final mappings and subsequent corrections
• Record owner decisions
• Read-only proof environment with masked sensitive fields where possible

Value equation and sensitivity

Integration capacity value per record family
= records requiring review
× minutes removed per record
÷ 60
× loaded hourly cost

All inputs are [ASSUMPTION]. This is capacity, not savings. It excludes software operation, validation, remediation, and adoption. It cannot be multiplied by 43 historical deals; the technology and integration process have changed across that period.

Proof design

• Examples: One record family with 5–10 representative exception classes, not merely 5–10 rows
• Hidden cases: A held-out record sample with known approved mappings
• Expected output: Exact matches, ranked candidates, unmapped records, conflicts, evidence, and confidence
• Judge: Data-domain owner plus integration lead
• Baseline: Match accuracy, manual minutes per record, unresolved count, and post-load correction rate
• Pass: 100% precision on auto-classified exact matches; at least 95% top-three candidate recall on review cases; zero unauthorized merges; at least 30% lower review time; complete reconciliation to source counts
• Unacceptable errors: Customer or supplier merge across distinct legal entities; wrong unit, product, price, tax, credit, or accounting mapping; lost source record; unlogged override
• Human boundary: Every non-exact mapping and every production load requires owner approval

Production controls

• Field-level access and masking
• Versioned source and target extracts
• Deterministic match tier before probabilistic candidate search
• Immutable approval and override log
• Segregation of duties for sensitive domains
• Reconciliation counts, totals, and duplicate checks
• Rollback and export of approved mappings
• Drift monitoring after target-master changes

Kill criteria

Stop if native Dynamics 365 tooling meets the acceptance test, source data lacks stable identifiers or ownership, mapping is not on the integration critical path, record volume is too low, post-load correction remains high, or the capacity value does not exceed build and operating cost.

Confidence

7. Portfolio architecture

No shared AI platform is justified before one workflow passes. If quote readiness works, only the following components have evidence-based reuse:

• Identity and permissions: Reuse company identity; apply project, branch, supplier, customer, and data-domain access.
• Source connectors: Read approved document, product, quote, ERP, and order sources. Do not create a second master.
• Canonical evidence envelope: Store source, record ID, revision, page or field, retrieval time, and permission context with every output.
• Deterministic rules: Required fields, version checks, exact matches, freshness, and prohibited actions run before language-model output.
• Permissioned retrieval: Return only approved product, customer, supplier, and policy evidence.
• Evaluation harness: Keep representative and hidden cases, expected outputs, unacceptable errors, and owner scores.
• Approval queue: Route exceptions to the person who owns the decision.
• Audit and event log: Record input versions, output, citation, approval, override, and downstream action.
• Monitoring and exception operations: Watch freshness, extraction failure, drift, reviewer disagreement, and unresolved queues.
• Transition and export: Export evidence, mappings, decisions, and logs in ordinary formats.

The shared capability is traceable evidence around a human decision. It is not an autonomous purchasing or quoting agent.

8. First 100 days

Days 0–30: Verify

Days 31–60: Prove

Days 61–100: Operate selectively

Only a passing proof advances.

9. Red team: why the thesis may be wrong

9.1 Existing capability may be sufficient

Nearly all branches already operate on an integrated technology platform. Online Advantage exposes common customer records and transaction evidence. Core & Main is recruiting for Dynamics 365 architecture work. Product and supplier tools may already contain document-generation, version, and exception functions not visible publicly.S1S8S11

If current configuration passes the acceptance test, the correct decision is to configure it, not build another application.

9.2 Public evidence may overstate repetition

A public bid page and two job postings do not establish enterprise volume. Treatment-plant quotes may be specialized, infrequent, or already centralized. Purchasing work at one location does not prove the same process or system at another.S7S9S10

The first diligence step is a volume and variance count, not an architecture discussion.

9.3 The company is already performing

Sales, gross profit, earnings, and cash generation are healthy. Q2 operating margin improved. Management reports market-share gains, municipal strength, greenfield expansion, and a long acquisition record.S1S3S4

A new workflow can destroy value if it distracts specialists, delays quotes, creates duplicate records, or standardizes away local expertise.

9.4 Value pools overlap

Quote capacity, conversion improvement, rework avoidance, and customer retention can describe the same economic event. Inventory release, lower expedite cost, and better fill rate can also trade against each other. The report keeps them separate; a live business case must choose the primary measure before testing.

9.5 Data and access may defeat the proof

Plans, addenda, supplier files, product documents, prices, and availability may lack stable identifiers or version history. Product rights and customer contracts may restrict use. If the proof cannot establish an authoritative source and revision, a polished packet is unsafe.

9.6 Adoption may dominate technical feasibility

A model can extract a specification and still fail operationally. Specialists may need different package shapes, branches may follow different rules, suppliers may change evidence formats, and a review queue may become one more screen. The proof must replace work, not add an audit ritual.

9.7 Attribution may fail

Win rate depends on price, availability, relationships, scope, competitors, and timing. Inventory depends on seasonality, demand, supplier terms, and service policy. Acquisition timing depends on people, policy, training, and cutover decisions. Technical success alone does not establish financial value.

9.8 Consequential decisions carry safety and contract risk

Water, wastewater, fire protection, and treatment-plant products can be governed by engineering specifications, approved materials, codes, and customer contracts. Silent product substitution or a fabricated compliance claim is unacceptable. Human approval and source evidence are operating requirements, not optional review steps.

9.9 The opportunity may be too small

The quote sensitivity produces only about $47,100 of gross profit in the base case before operating cost. The base capacity case is $90,000 and cannot be added if that same capacity causes the conversion gain. A team with low volume or little rework should use a checklist or native feature and stop.

10. Diligence questions

Business and economics

1. How many treatment-plant or similarly document-heavy packages does the proposed team process each year?
2. What is annual quoted value, won value, and gross margin for that workflow?
3. What share of lost or delayed work is attributable to completeness, addenda, submittals, or response time?
4. How much specialist and manager time does a package require by stage?
5. Which measured outcome would justify proceeding: throughput, conversion, rework, cycle time, working capital, or risk?
6. What annual operating cost and adoption burden would make the opportunity uneconomic?

Workflow

7. What event starts the quote package, and what event marks it complete?
8. Which package sections are mandatory by project type?
9. Who owns requirement interpretation, product choice, approved equals, price, and customer submission?
10. How are addenda received, versioned, reconciled, and acknowledged today?
11. What are the five most common exceptions, and which are consequential?
12. How many package variants exist across teams or branches?

Systems and data

13. Which systems hold the authoritative quote, product, supplier, price, availability, and document records?
14. What current system can already extract, compare, retrieve, or assemble the required evidence?
15. What is the deployed and planned scope of Dynamics 365?
16. Are product documents versioned and linked to stable item identifiers?
17. Can historical price and availability be reconstructed at the original decision date?
18. What data may leave Core & Main systems during a proof?

Acceptance and risk

19. What exact error must never pass silently?
20. What readiness rubric can the judge apply before seeing any model output?
21. How many hidden cases can be reserved?
22. What confidence threshold requires abstention?
23. Which customer, supplier, engineering, privacy, security, or regulatory terms restrict the workflow?
24. What audit evidence must survive a dispute or project review?

Expansion

25. If quote readiness passes, which second team uses materially the same inputs and rubric?
26. Which components are reusable without creating a second product or customer master?
27. What production metric would trigger rollback?
28. Who owns monitoring, correction, and rule changes after launch?

11. Source ledger

12. Assumption and unknown ledger

13. Corrections and change log