Scenario 2: Misaligned Governance | ShaDey Inc. Centralized Business Glossary

1 Introduction

In this scenario, ShaDey Inc. has a centralized business glossary, but governance has broken down in practice. The CDM was created by a consulting firm and never maintained. Teams built their own LDMs independently. The PDM was inherited from a legacy SAP migration and ad-hoc analytics projects. The result: glossary terms exist on paper, but they do not match what is actually implemented.

1. The glossary exists, but teams disagree on definitions (e.g., "Active Customer" means different things to Marketing, Finance, and Engineering).
2. CDM uses SAP-influenced names like "Account" and "Material" instead of business-friendly "Customer" and "Product."
3. LDM attributes are abbreviated, undocumented, and vary by team.
4. PDM tables are named after source systems (kna1_extract, mara_data) or ad-hoc projects (cust_data_v2_FINAL).
5. AI/ML teams cannot trust any feature because lineage is broken.

Maturity: DAMA DMBOK Level 1–2 (Initial/Repeatable) | Glossary Status: Exists but unenforced and outdated | Naming Convention: None; each team uses its own | Lineage: Broken; cannot trace glossary to PDM | Data Contracts: Informal or absent

Compare with Scenario 1: Aligned Governance to see what proper alignment looks like.

2 Centralized Business Glossary

The same 13 enterprise terms exist in the glossary, but definitions are contested, ownership is unclear, and the mapping to data models is inconsistent or broken. Expand each card to see the misalignment.

Customer

Customer Management High

A person or organization that has purchased from or registered with ShaDey Inc. through any channel (in-store, e-commerce, wholesale).

Full Metadata: Misalignment Details

SynonymsAccount, Buyer, Patron, Shopper

Related TermsActive Customer, Loyalty Tier, Customer Type

Data OwnerUnclear (claimed by both Sales Ops and Marketing)

Data StewardVacant since Q2 2025

Example Usage"How many new customers acquired in Q4?" (answer differs depending on which table you query)

Governing PolicyGDPR policy exists but PII masking is inconsistent across tables

CDM EntityAccount, SAP-influenced name, not matching glossary term "Customer"

LDM Entitycust, abbreviated, 5 attributes, no business rules documented

PDM Tablekna1_extract (SAP replica), cust_data_v2 (analytics), customer_export_final_FINAL (ad-hoc)

ConsistencyBroken. 3 different PDM tables, none matching glossary naming.

GlossaryCustomer

CDMAccount

LDMcust

PDMkna1_extract

Active Customer

Customer Management High

A customer who has completed at least one purchase transaction within a defined trailing period from the measurement date. Warning: the "defined period" differs across three teams.

Full Metadata: Conflicting Definitions

SynonymsCurrent Customer, Engaged Customer

Related TermsCustomer, Lapsed Customer, Loyalty Tier

Data OwnerDisputed (Marketing, Finance, and Engineering each claim authority)

Data StewardNone assigned

Governing PolicyNo enterprise-wide standard. Each team applies its own window.

CDMNot explicitly modeled; assumed to be a flag on Account

LDMcust.active_yn (CHAR(1)), no documented derivation rule

PDMkna1_extract.AKTIV (SAP field), cust_data_v2.is_active (Marketing logic), rpt_active_cust (Finance snapshot)

Consistency3 conflicting definitions across teams. Reports never agree.

Conflicting Definitions by Team

MarketingPurchased OR browsed within 6 months (includes website visits)

FinanceCompleted purchase within trailing 12 months (revenue-based only)

EngineeringLast login within 90 days (app engagement, no purchase required)

GlossaryActive Customer

CDMNot modeled

LDMactive_yn

PDM3 tables

Product

Merchandising High

A sellable item in ShaDey Inc.'s portfolio, identified by a unique SKU, belonging to a Category within a Department.

Full Metadata: Misalignment Details

SynonymsItem, SKU, Article, Material (SAP legacy)

Related TermsCategory, Department, Brand, Variant, Supplier

Data OwnerVP, Merchandising (owns glossary term but not data models)

Data StewardAssigned but has no authority over PDM naming

Governing PolicySKU format should be [DEPT]-[CAT]-[SEQ] but SAP uses MATNR (18-digit) and e-commerce uses its own product_id

CDMMaterial, SAP-influenced name. Business calls it "Product"; CDM calls it "Material".

LDMmat, abbreviated, 4 attributes (matnr, maktx, mtart, matkl). No business rule documentation.

PDMmara_data (SAP replica), product_catalog_export (e-commerce), item_master (warehouse)

ConsistencyBroken. "Product" in glossary → "Material" in CDM → "mat" in LDM → "mara_data" in PDM.

GlossaryProduct

CDMMaterial

LDMmat

PDMmara_data

Sale

Sales & Revenue High

A completed purchase transaction in which a customer acquires one or more products through a channel at an agreed price.

Full Metadata: Misalignment Details

SynonymsTransaction, Purchase, Sales Transaction, Billing Document

Related TermsNet Sales, Order, Return, Promotion, Channel

Data OwnerHead of Sales Operations

CDMBillingDocument, SAP VBRK-influenced name instead of "SalesTransaction"

LDMbilling_doc, mixes orders and sales in the same entity, no clear grain

PDMvbrk_vbrp_joined (SAP replica) and pos_transactions (POS system). Not reconciled.

ConsistencyBroken. "Sale" → "BillingDocument" → "billing_doc" → "vbrk_vbrp_joined". Sales mixed with orders.

GlossarySale

CDMBillingDocument

LDMbilling_doc

PDMvbrk_vbrp_joined

Net Sales

Sales & Revenue High

Gross sales revenue minus returns, discounts, allowances, and sales taxes. Warning: three different calculation formulas exist in production.

Full Metadata: Formula Conflicts

SynonymsRevenue, Net Revenue, Top Line

Related TermsSale, Return, Promotion, Gross Sales

Data OwnerCFO owns the definition, but Marketing and Ops use their own formulas

Data StewardFinance analyst (no cross-team authority)

Governing PolicyNo single enterprise formula. Each dashboard implements its own calculation.

CDMNot explicitly modeled; assumed to be derivable from BillingDocument

LDMNo calculated field defined. Each team computes in their own SQL.

PDMrpt_revenue_daily (Finance), mktg_net_sales_cube (Marketing), ops_sales_summary (Operations)

ConsistencyBroken. 3 formulas, 3 tables, none matching. Leadership sees 3 different "Net Sales" numbers.

Conflicting Formulas by Team

FinanceSUM(gross_amount) - SUM(returns) - SUM(discounts) - SUM(tax)

MarketingSUM(gross_amount) - SUM(returns) (excludes discounts and tax)

OperationsSUM(line_total) (pre-aggregated, includes tax, excludes returns)

GlossaryNet Sales

CDMNot modeled

LDMNot defined

PDM3 tables

Inventory On Hand

Supply Chain High

The quantity of a product physically available for sale at a specific location, measured in sellable units, excluding reserved, damaged, or in-transit stock.

Full Metadata: Misalignment Details

SynonymsStock on Hand, Available Inventory, SOH, Unrestricted Stock

Related TermsInventory, Safety Stock, Reorder Point, Days of Supply

Data OwnerVP, Supply Chain

Governing PolicyGlossary says "excludes reserved" but SAP table MARD includes all stock types in LABST field. Warehouse system tracks sellable only.

CDMStockPosition, different name from glossary's "Inventory"

LDMstk_pos, abbreviated, mixes stock types (unrestricted + reserved + blocked)

PDMmard_extract (SAP), wms_inventory_live (WMS). Numbers differ by 8–12% due to definition mismatch.

ConsistencyBroken. "On Hand" means different things in SAP vs. WMS. Permanent 8–12% variance.

GlossaryInventory On Hand

CDMStockPosition

LDMstk_pos

PDMmard_extract

Supplier

Procurement Medium

An external organization that manufactures, distributes, or imports products sold by ShaDey Inc., bound by a procurement contract.

Full Metadata: Misalignment Details

SynonymsVendor, Manufacturer, Distributor

Related TermsProduct, Lead Time, Quality Rating, Contract

Data OwnerVP, Procurement

CDMVendor, SAP-influenced name. Glossary says "Supplier"; CDM says "Vendor".

LDMvndr, abbreviated, 4 attributes (lifnr, name1, land1, ktokk). No quality rating.

PDMlfa1_extract (SAP replica), uses SAP field names (LIFNR, NAME1, LAND1)

ConsistencyBroken. "Supplier" → "Vendor" → "vndr" → "lfa1_extract". Names diverge at every layer.

GlossarySupplier

CDMVendor

LDMvndr

PDMlfa1_extract

Promotion

Merchandising Medium

A time-bound pricing incentive (discount, BOGO, coupon, markdown) applied to one or more products or categories to drive sales volume.

Full Metadata: Misalignment Details

SynonymsDeal, Campaign, Offer, Markdown, Condition (SAP)

Data OwnerVP, Merchandising

CDMPricingCondition, SAP condition record concept, not "Promotion" from the glossary

LDMcond_rec, mixes permanent pricing and promotions in one entity

PDMkonv_extract (SAP KONV replica), no distinction between promo and standard pricing

ConsistencyPartially broken. Promotions and pricing are conflated in all layers below the glossary.

GlossaryPromotion

CDMPricingCondition

LDMcond_rec

PDMkonv_extract

Order

Sales & Revenue High

A customer's expressed intent to purchase one or more products. An order may be pending, confirmed, fulfilled, or cancelled. Distinct from a completed sale.

Full Metadata: Misalignment Details

SynonymsSales Order, SO, Sales Document

Data OwnerHead of Sales Operations

CDMSalesDocument, SAP term, mixes orders, quotes, and contracts in one entity type

LDMsd_doc, generic, no distinction between order subtypes. Status field is a free-text VARCHAR.

PDMvbak_vbap_joined (SAP), ecom_orders (e-commerce). Not reconciled; same order can appear in both.

ConsistencyBroken. Orders mixed with quotes and contracts. Duplicates across 2 PDM tables.

GlossaryOrder

CDMSalesDocument

LDMsd_doc

PDMvbak_vbap_joined

Shipment

Logistics Medium

A physical movement of goods from a warehouse or distribution center to a customer delivery address or store location.

Full Metadata: Misalignment Details

SynonymsDelivery, Dispatch, Consignment, Outbound Delivery

Data OwnerVP, Logistics & Distribution

CDMDelivery, SAP calls it "Delivery" (LIKP); glossary says "Shipment"

LDMdeliv, abbreviated, 5 attributes. Ship and delivery dates in different formats (DATE vs. VARCHAR).

PDMlikp_lips_joined (SAP replica), shipment_tracking (3PL system). No FK between them.

ConsistencyPartially broken. Two systems track the same physical shipment with no link.

GlossaryShipment

CDMDelivery

LDMdeliv

PDMlikp_lips_joined

Return

Sales & Revenue Medium

A customer-initiated reversal of a completed sale, resulting in the return of goods and issuance of a refund or credit to the customer.

Full Metadata: Misalignment Details

SynonymsRefund, Credit, Reverse Sale, Credit Memo

Data OwnerHead of Sales Operations

CDMNot separately modeled; returns are a subtype of SalesDocument (doc type = RE)

LDMMixed into sd_doc with a type flag. No dedicated return entity.

PDMEmbedded in vbak_vbap_joined WHERE doc_type = 'RE'. Requires filtering, easy to miss and count returns as sales.

ConsistencyBroken. Returns are not modeled separately; they are hidden inside sales tables, causing double-counting.

GlossaryReturn

CDMSalesDocument

LDMsd_doc (mixed)

PDMvbak_vbap (filtered)

Feature Store Record

Data Science High

A pre-computed, versioned, point-in-time correct feature vector derived from governed data products, stored in the centralized ML feature store.

Full Metadata: Misalignment Details

SynonymsFeature Vector, ML Feature, Feature Row

Data OwnerHead of Data Science

Governing PolicyNo feature governance. Each data scientist creates features ad-hoc from whatever tables they can find.

CDMNot modeled; CDM predates AI/ML initiatives by 3 years

LDMNot modeled

PDMds_features_alice, ml_features_v3_bob, feature_cache_tmp, personal schemas, no catalog

ConsistencyBroken. No CDM/LDM mapping. Features built from untraceable ad-hoc SQL. No versioning, no lineage.

GlossaryFeature Store Record

CDMNot modeled

LDMNot modeled

PDMad-hoc tables

Model Training Dataset

Data Science High

A point-in-time correct, labeled dataset assembled for training or retraining an ML model. Should be immutable with full lineage.

Full Metadata: Misalignment Details

SynonymsTraining Set, ML Dataset, Training Snapshot

Data OwnerHead of Data Science

Governing PolicyNo policy. Datasets created from personal feature tables, not reproducible, may contain PII.

CDMNot modeled

LDMNot modeled

PDMtraining_data_churn_v2_FINAL, alice_train_set_backup, mutable, no versioning, stored in personal schemas

ConsistencyBroken. No lineage, not reproducible, PII leakage risk, no immutability guarantee.

GlossaryTraining Dataset

CDMNot modeled

LDMNot modeled

PDMad-hoc files

3 Enterprise Data Models

The CDM, LDM, and PDM exist but were built independently. Entity names, attribute names, and physical table names do not align with the business glossary, or with each other.

Conceptual Data Model (SAP-Influenced)

11 entities, but names are derived from SAP technical terminology, not the business glossary. A consultant built this from SAP transaction codes, not from business workshops.

Account

Glossary: "Customer". SAP-derived name. Mismatch

Material

Glossary: "Product". SAP MARA-derived. Mismatch

Plant

Glossary: "Store / Channel". SAP WERKS concept. Mismatch

Vendor

Glossary: "Supplier". SAP LFA1-derived. Mismatch

StockPosition

Glossary: "Inventory". Partially aligned; includes reserved stock.

SalesDocument

Glossary: "Order" + "Sale" + "Return" mixed. Major mismatch

BillingDocument

Glossary: "Sale". SAP VBRK-derived. Mismatch

PricingCondition

Glossary: "Promotion". Mixes promotions with permanent pricing.

Delivery

Glossary: "Shipment". SAP LIKP name. Close but different.

CreditMemo

Return subtype, financial view only. Physical return not modeled.

(ML not in CDM)

AI/ML FeatureSet and Training Dataset are not represented. CDM predates ML.

The CDM was built by reverse-engineering SAP tables, not by modeling business concepts. Entity names like "Account", "Material", "Plant", and "Vendor" are SAP terminology that business users do not recognize. The glossary-to-CDM link is broken at the first step.

Logical Data Model (Inconsistent Attributes)

Customer (as "cust"): 5 abbreviated attributes

Attribute	Type	Constraint	Glossary Link
`kunnr`	VARCHAR(10)	PK	SAP field name, not "customer_id"
`name1`	VARCHAR(40)	NOT NULL	SAP field name, should be "full_name"
`email`	VARCHAR(100)		Length too short (max 100 vs. RFC 5321 max 254)
`active_yn`	CHAR(1)		Y/N flag, no derivation rule documented
`cust_type`	VARCHAR(20)		Free-text, not ENUM. Values: "retail", "Retail", "RETAIL", "b2b", etc.

Compared to Scenario 1, this LDM is missing: loyalty_tier, registration_date, preferred_channel, ltv_segment, and primary_store_id. No SCD tracking. No business rules documented.

Supplier (as "vndr"): 4 abbreviated attributes

Attribute	Type	Constraint	Glossary Link
`lifnr`	VARCHAR(10)	PK	SAP field name, not "supplier_id"
`name1`	VARCHAR(40)	NOT NULL	SAP name, should be "supplier_name"
`land1`	CHAR(3)		SAP country field, not "country_code" (ISO 3166)
`ktokk`	VARCHAR(4)		SAP account group. No business meaning documented.

Missing: lead_time_days, quality_rating, contract_start_date, payment_terms, is_certified, primary_category_id. The LDM has 4 of the 10 attributes needed.

Physical Data Model (Ad-Hoc Tables)

No naming convention. Tables are named after source systems, project names, or developer initials. Multiple tables for the same concept with no reconciliation.

kna1_extract (SAP customer replica)

CREATE TABLE kna1_extract ( KUNNR VARCHAR(10) PRIMARY KEY, -- SAP customer number, not UUID NAME1 VARCHAR(40), -- SAP field name, not "full_name" NAME2 VARCHAR(40), -- no glossary mapping ORT01 VARCHAR(40), -- city in SAP German abbrev LAND1 CHAR(3), -- country code AKTIV CHAR(1), -- active flag, no derivation rule ERDAT VARCHAR(8) -- date as YYYYMMDD string, not DATE type );

cust_data_v2 (analytics team ad-hoc)

CREATE TABLE cust_data_v2 ( id INT IDENTITY PRIMARY KEY, -- auto-increment, no business key cust_num VARCHAR(20), -- sometimes KUNNR, sometimes email cust_name VARCHAR(100), -- concatenated, inconsistent format is_active BOOLEAN, -- Marketing's 6-month definition segment VARCHAR(50), -- free-text: "high value", "HIGH", etc. last_updated DATETIME -- no timezone, no SCD tracking );

mara_data, vbrk_vbrp_joined, additional ad-hoc tables

mara_data (SAP material replica)

CREATE TABLE mara_data ( MATNR VARCHAR(18) PRIMARY KEY, -- SAP material number, not UUID/SKU MAKTX VARCHAR(40), -- description in SAP, not "product_name" MTART VARCHAR(4), -- material type code, no ENUM MATKL VARCHAR(9) -- material group, maps to "category" loosely );

vbrk_vbrp_joined (SAP billing replica)

CREATE TABLE vbrk_vbrp_joined ( VBELN VARCHAR(10), -- billing document number POSNR VARCHAR(6), -- line item KUNAG VARCHAR(10), -- sold-to customer (SAP) MATNR VARCHAR(18), -- material, not "product_id" FKIMG DECIMAL(13,3), -- quantity NETWR DECIMAL(15,2), -- net value per SAP (includes some deductions) WAERK CHAR(5) -- currency );

Also in production: customer_export_final_FINAL, product_catalog_export, item_master, pos_transactions, ecom_orders, rpt_revenue_daily, mktg_net_sales_cube, ops_sales_summary, ds_features_alice, ml_features_v3_bob, training_data_churn_v2_FINAL, alice_train_set_backup. Total: 20+ tables for concepts that Scenario 1 covers with 8 governed tables.

4 Glossary-to-Model Alignment Flow

In Scenario 2, the flow from glossary to PDM is broken at every step. Names diverge, definitions conflict, and physical tables cannot be traced back to glossary terms.

Broken Flow: "Customer"

Business Glossary

Customer
"A person or org that has purchased from ShaDey Inc."

name change

CDM

Account
SAP-derived name, different from glossary

abbreviated

LDM

cust
5 attrs, SAP field names, no rules

fragmented

PDM

kna1_extract
cust_data_v2
customer_export_final_FINAL

Broken Flow: "Net Sales"

Business Glossary

Net Sales
"Gross sales − returns − discounts − tax"

not modeled

CDM

Not modeled
No derived metric concept

no formula

LDM

Not defined
Each team writes its own SQL

3 versions

PDM

rpt_revenue_daily
mktg_net_sales_cube
ops_sales_summary

When naming diverges at CDM and definitions fragment at LDM, the PDM becomes a collection of ad-hoc tables that nobody fully understands. Impact analysis is impossible; changing one definition could break any of 20+ tables, and nobody knows which ones. Leadership meetings devolve into "which number is right?" arguments.

5 Term Deep Dives: Broken Traceability

Tracing 8 key terms through every layer. Red indicates misalignment; amber indicates partial alignment. Compare with Scenario 1 where every row is green.

Glossary Term	CDM Entity	LDM Entity / Attribute	PDM Table(s)	Status	Root Cause
Customer	`Account` (wrong name)	`cust`, abbreviated, 5 attrs	`kna1_extract`, `cust_data_v2`, `customer_export_final_FINAL`	Broken	SAP-derived CDM, no naming standard
Active Customer	Not modeled	`cust.active_yn`, no rule	3 tables, 3 definitions (6mo / 12mo / 90d)	Broken	No enterprise standard definition
Product	`Material` (SAP name)	`mat`, 4 SAP attrs	`mara_data`, `product_catalog_export`, `item_master`	Broken	SAP naming, multiple systems
Net Sales	Not modeled	Not defined; ad-hoc SQL per team	`rpt_revenue_daily`, `mktg_net_sales_cube`, `ops_sales_summary`	Broken	3 conflicting formulas in production
Inventory On Hand	`StockPosition` (close)	`stk_pos`, mixes stock types	`mard_extract`, `wms_inventory_live`	Partial	Semantic mismatch: includes reserved
Supplier	`Vendor` (SAP name)	`vndr`, 4 SAP attrs	`lfa1_extract`	Broken	SAP naming at every layer
Order / Return	`SalesDocument` (mixed)	`sd_doc`, orders + returns + quotes	`vbak_vbap_joined`, must filter by doc_type	Broken	3 glossary terms collapsed into 1 entity
Feature Store / Training	Not in CDM	Not in LDM	Personal schemas: `ds_features_alice`, `ml_features_v3_bob`	Broken	CDM predates AI/ML, no governance

In Scenario 2, no column in the PDM can be reliably traced back to a glossary term. Impact analysis requires manual detective work across 20+ tables. A schema change in kna1_extract could break reports, ML features, and downstream exports, and nobody knows the full blast radius until something fails in production.

6 Business Impact

How misaligned governance and an inconsistent glossary damage every aspect of data operations at ShaDey Inc.

Analytics & Reporting

"Net Sales" shows 3 different numbers across Finance, Marketing, and Ops dashboards
Leadership spends 40% of data review meetings debating which number is correct
Monthly reconciliation takes 2–3 analyst-days per cycle

Trust & Confidence

Business users do not trust the data; they maintain shadow spreadsheets
"Which number is right?" is the most common question in leadership meetings
New analysts spend 3–4 weeks just learning which tables to trust

AI/ML Failure Risk

Features built from ad-hoc tables contain definition drift; "active customer" means 3 things
Training data is not reproducible; retraining produces different results
PII leakage risk: no governance on personal-schema feature tables

Governance & Compliance

GDPR audit: cannot prove where customer PII resides across 20+ tables
SOX compliance: Net Sales formula is not standardized for financial reporting
Data steward position vacant; no enforcement of glossary definitions

Interoperability

Cross-domain queries fail because "Customer" is KUNNR in one table and cust_num in another
Joining SAP tables to analytics tables requires manual key mapping per query
API contracts use different entity names than the database

Operational Cost

Estimated $1.2M/year in reconciliation, rework, and duplicate pipeline maintenance
New data products take 3–6 months instead of weeks due to "data archaeology"
20+ tables to maintain vs. 8 governed tables in Scenario 1

7 Key Takeaways

ShaDey Inc. has a glossary, but it is not enforced. When CDM entities use SAP names ("Account", "Material", "Vendor") instead of glossary terms ("Customer", "Product", "Supplier"), the glossary becomes documentation that nobody reads.

"Customer" → "Account" → "cust" → "kna1_extract". At each layer, the name drifts further from the business term. By PDM, a new analyst cannot tell which table represents which glossary concept without tribal knowledge.

When "Active Customer" has 3 definitions and "Net Sales" has 3 formulas, every ML feature that depends on these concepts inherits silent inconsistency. The model does not know which definition it was trained on, and neither does the team.

The gap between Scenario 1 and Scenario 2 is not theoretical. It manifests as: 3x more tables to maintain, monthly reconciliation cycles, leadership trust erosion, compliance audit failures, and 3–6 month delays on every new data product.

See how this should work: Scenario 1: Aligned Governance