Part Master Data Cleanup — Proposal

Prepared by Webisoft for The Griff Network · August 2026

In response to your Project Scope Brief (July 2026). All baseline figures below are your own, measured 2026-07-29 against your verified replica.

1 · The problem

The problem

Your brief has measured it precisely. In Division 1: 56.8% of the catalogue shares a description verbatim with at least one other part (one description covers 606 parts); structured dimensions are populated on 48 of 14,759 parts; 54 of 71 purchasing unit codes carry quantities embedded within the code itself; 48 parts marked "DO NOT USE" remain active; 632 parts have no inventory class, 1,606 no product group, and the attribute system — the structured home for material identity — is in use on zero parts. A third of the active catalogue has not transacted since before 2023. Division 2 is younger and lighter: its principal issue is 2,065 active parts that have never transacted at all.

The day-to-day consequences:

56.8%

of the catalogue shares a description verbatim with at least one other part

48 of 14,759

structured dimensions are populated

54 of 71

purchasing unit codes carry quantities embedded within the code itself

2,065

active parts that have never transacted at all

Search is a matter of tribal knowledge. With hundreds of parts sharing an identical description, every quotation, order entry, and purchase depends on institutional memory to select the correct part — and a wrong selection in converting carries the cost of scrap, rerun, and freight.

MRP, scheduling, and costing are effectively unavailable. MRP suggestions, min/max planning, and cost roll-ups are only as sound as the part master feeding them; with dimensions held in part numbers and quantities embedded in unit codes, the system cannot plan or convert.

Purchasing cannot compare. Units denoting "a roll of a specific footage" do not convert to pounds or feet, so price-per-pound comparison across suppliers is impossible by construction.

Reports and counts are inflated by thousands of dormant and duplicate records, and each new joiner inherits the archaeology.

2 · Analysis

Analysis: the root cause is a missing data model

None of these defects is a data-entry problem. They are a single structural failure expressed through different fields: the part number has been doing the database's work. A 293-column part record accepts whatever state it is given — nothing objected when 22x3 was stored within an identifier whilst the width field remained empty, and nothing prevented 8,376 parts from sharing a description.

Sound systems keep three layers distinct:

┌─────────────────────────────────────────────────────────────┐
│  BUSINESS REALITY                                           │
│  Materials, sizes, customers, suppliers, units               │
│  "70–79# bleached S2S liner, cut 12×12, sold to Customer B" │
└──────────────────────────┬──────────────────────────────────┘
                           │  described by
┌──────────────────────────▼──────────────────────────────────┐
│  DATA MODEL  ← the missing layer                            │
│  MATERIAL ─< ITEM (material × form × size, unique)          │
│  ITEM ─< CROSS-REFERENCE (customer / supplier / legacy)     │
│  UOM catalogue with true conversions · generated descriptions│
└──────────────────────────┬──────────────────────────────────┘
                           │  implemented in
┌──────────────────────────▼──────────────────────────────────┐
│  SYSTEM (Epicor Kinetic)                                    │
│  Part master fields · attributes · part cross-references ·  │
│  UOM classes · DMT loads · BAQs / reports · MRP             │
└─────────────────────────────────────────────────────────────┘

At present the middle layer does not exist, so business facts have been carried directly into system fields never intended to hold them — customer names within part numbers, sizes within identifiers and comments, lifecycle status within description text. Our methodology is to construct that middle layer first: an explicit model of your materials, items, and relationships — signed off by you before any transformation runs — with every record then required to satisfy it. Cleansing thereby becomes provable: a record either conforms to the model or appears in the exception log. Nothing is resolved silently.

For your largest family, the ~600-entry fan-out resolves to one material, ~270 items, and customer relationships as cross-references — and a duplicate size becomes a constraint violation the model rejects, rather than something a reviewer must happen to notice.

3 · Phase 1

Phase 1 — Cleanup (fixed fee: $10,000 · 10 business days)

Precisely the engagement your brief requests — standalone, item master only, both divisions treated as separate lots, no system changes — delivered through the model above.

Step	What we shall do	What we shall need from you	Result
1. Data model (days 1–2)	Draft the material/item/cross-reference model, description grammar, and UOM catalogue from your data; review it with your experts against real parts	Item-master extracts; one half-day working session; corrections to our readings	Signed target-state specification — the standard against which everything is measured
2. Pilot ETL on a subset (days 3–4)	Build the parsing rules and pipeline; run one difficult family plus a random sample end to end	Your naming-habit parsing rules; same-day review of the pilot	Validated rules, a measured auto-parse rate, and a pilot dataset reviewed against your own parts
3. First load into Epicor (day 5)	Generate DMT-compatible load files for the pilot; support your test load into a non-production company	Non-production company access; DMT templates; Kinetic build version	A proven load path ahead of scale — no surprises at cutover
4. Iterate to full coverage (days 5–9)	Run the full catalogue through the pipeline nightly; cluster whatever the rules cannot resolve and present it with evidence and a proposed resolution; each of your decisions is encoded as a new rule and applied throughout	~30–45 min/day with a named decision owner per division (convened only when required; ≈17–27 expert-hours in total for Lot 1, 4–6 for Lot 2)	Exceptions diminish daily; one answer resolves an entire cluster — a 606-part duplicate is one question, not 606
5. Verify & hand over (day 10)	Random-sample audit (ANSI Z1.4, AQL 1.0); 100% verification of crosswalk and conversions; delta re-run against a fresh extract at your load date	UAT sign-off	Your deliverables, as itemised below

Deliverables, in the formats requested: the cleaned item master (UTF-8 CSV, with an XLSX review copy) · the complete old-to-new crosswalk (CSV/XLSX) · DMT-compatible load files (CSV per template) · the exception log with reason codes (XLSX, filterable). All seven in-scope transformations in your brief are covered: consolidation to one governed record per material and size, customer identifiers moved to cross-references, regenerated descriptions, dimensions and specifications extracted into structured fields, units of measure normalised with true conversions, identifier hygiene with every rename tracked, and the crosswalk and exception log themselves.

You asked for clean data; Phase 1 also leaves you the machinery that produced it — the signed data model, the versioned rule catalogue, the ETL pipeline, and a data-quality scorecard: re-runnable queries scoring your part master against the model (duplicate rate, dimension coverage, UOM validity, classification completeness, dormancy). Your own figures of 2026-07-29 form the baseline; an empty violation report is the acceptance criterion. Any lot failing the sampling audit is reworked at our cost.

Enablement is included, not an extra. Throughout the engagement we hold open office hours for your team — working consultation, at no additional fee, on the processes and interfaces that surround the data: part-creation practice, adjudication routines, use of the scorecard and crosswalk, and how the rule catalogue is read and extended. The intent is that adoption runs as smoothly as the cleanup itself: by handover, the people who create and steward parts are already working to the new standard, rather than being introduced to it.

The timeline assumes the daily adjudication cadence above; should decisions move to a slower rhythm, the elapsed time extends accordingly, but the effort and the fee do not change.

4 · Phase 2

Phase 2 — Making it permanent (scoped from Phase 1 findings · engagement total: $25,000)

Cleansing decays unless the standard is enforced where data is created — your own brief acknowledges as much ("prevention and pruning"). Phase 2 turns Phase 1's machinery into a system your team adopts, taking one of two forms (or a blend), selected on the basis of what Phase 1 reveals about your workflows:

Clean-behind automation.

Your people work exactly as they do today; the Phase 1 pipeline runs continuously against your nightly replica, scoring and correcting new records by the rule catalogue and routing genuine exceptions to the same adjudication queue. The scorecard becomes a live dashboard.

Constraint-at-entry interface.

A lightweight translation layer implementing the data model's constraints at the point of creation — non-conforming data cannot enter, or the user is shown the record's quality score and what is missing before saving. New parts are born conforming.

In either form, the enablement office hours continue and deepen: we shall support and coach your team as required — named decision owners, a part-creation playbook, training on the chosen interface or pipeline, and full handover of the rules — so that the standard remains yours to keep, not ours to rent.

5 · Commercials

Commercials

	Scope	Fee	Timeline
Phase 1 — Lot 1 (Paper & Film)	Full cleanup: parse, consolidate, restructure	$8,000	Delivered day 10
Phase 1 — Lot 2 (Decorative Films)	Dormant-catalogue pruning and description normalisation	$2,000	Delivered end of week 1
Phase 1 combined	Both lots, plus model, rules, pipeline, scorecard	$10,000 fixed	10 business days
Phase 2	Systematisation per §4, scoped jointly from Phase 1 findings	$15,000 (indicative)	Scoped at Phase 1 close
Engagement total		$25,000

We confirm this is deliverable as a standalone, fixed-fee data engagement — item master only, no platform components — as scoped in your brief. Phase 1 stands alone: you receive everything your brief requests whether or not Phase 2 proceeds. Fees are fixed at signature, with a 30-day post-load correction warranty covering defects against the signed specification. A delta re-run against a fresh extract is included within 2 business days of your chosen load date.