AI & Automation5 min readNetray Engineering TeamUpdated March 4, 2025

Building an ERP Data Quality Framework for Manufacturing

An ERP data quality framework is a repeatable system for measuring, monitoring, and fixing the accuracy of the master and transactional data your plant runs on. It combines profiling, agreed quality dimensions, automated rules that run on every load, a scorecard with named owners, and a remediation workflow. Manufacturers need it because bad data costs cash directly: a wrong lead time inflates safety stock, a stale standard cost distorts margin by product family, and a duplicate item number splits demand history so MRP orders the same part twice.

Profile Before You Prescribe

Start with a two-week profiling pass rather than a policy document. Run column-level statistics across the master tables that drive planning: item, itemwhse, bom or job routing, customer, vendor, and the price and cost tables. You are looking for population rates, distinct value counts, format anomalies, and out-of-range values. Typical findings in a twenty-year-old SyteLine or LN instance include 15 to 30 percent of items with a lead time of zero or a default carried over from a decade-old conversion, unit-of-measure mismatches between purchase and stocking UOM, obsolete items still flagged as planned, and customer records with three spellings of the same parent company. Quantify each finding in dollars or hours before proposing a fix; that is what secures the budget.

The Six Dimensions Applied to Manufacturing Data

Use the DAMA dimensions, but express them in shop-floor terms so business owners can argue with them. Completeness asks whether every active item has a lead time, planner code, commodity code, and ABC class. Accuracy compares a value to reality, such as whether the on-hand quantity matches a cycle count. Consistency checks the same fact across systems, like item UOM in ERP versus MES. Timeliness measures lag between a physical event and its ERP posting. Validity enforces format and domain rules, such as ECCN or Schedule B codes matching a controlled list. Uniqueness catches duplicate item, customer, and supplier records that fragment history.

Completeness: percentage of active items with lead time, planner, commodity code, and ABC class populated
Validity: ECCN, HTS, and UNSPSC codes checked against current published reference lists
Uniqueness: fuzzy match on manufacturer part number and description to find duplicate item masters
Timeliness: median hours between physical receipt or labor event and the corresponding ERP transaction

Automated Rules and Where to Run Them

Rules belong in two places. Preventive rules run at entry: a required-field policy on the item master form, a validation extension that rejects a purchase UOM inconsistent with the stocking UOM, or a workflow approval for any standard cost change above a threshold. Detective rules run in the pipeline as tests on every warehouse load, expressed as code so they are versioned and reviewed. A practical starting set is 40 to 60 rules covering the master data that MRP and costing depend on. Fail the load loudly for critical rules such as an orphaned job transaction with no matching job, and record a warning with a trend for softer rules such as a missing commodity code.

Scorecards, Ownership, and Remediation Workflow

A quality score nobody owns changes nothing. Publish a scorecard per data domain with a named business owner: item master to engineering or planning, customer master to sales operations, supplier master to procurement. Show the current pass rate per rule, the trend over 13 weeks, and the count of open exceptions with age. Route each failed record into a work queue with an assignee and an SLA rather than emailing a spreadsheet. Track the backlog burn-down the same way you track a maintenance backlog, with a weekly opened-versus-closed chart so leadership can see whether the team is gaining ground. Review the scorecard in an existing operations meeting rather than a separate data governance forum; attendance and accountability are far higher when quality sits next to the other numbers a plant manager is already answering for.

One named business owner per domain, published on the scorecard with their pass-rate target
Exception queues with an SLA: 5 business days for planning-critical fields, 20 for descriptive fields
A 13-week trend per rule so improvement is visible and regressions are caught early
A quarterly review that retires stale rules and adds new ones from recent production incidents

How Netray Automates ERP Data Quality

Netray deploys AI agents that profile your SyteLine, LN, or M3 master data, propose a rule set grounded in what your planning and costing logic actually consumes, and generate those rules as tests in your pipeline. Agents cluster duplicate item and supplier records using description, manufacturer part number, and specification text, then produce merge candidates with confidence scores for a human to approve rather than auto-merging. Clients commonly move item master completeness from the low 70s to above 95 percent within two quarters, and cut duplicate item creation sharply by validating at entry. For controlled environments, the same agents run entirely on-prem.

Frequently Asked Questions

What is a good data quality score for ERP master data?

There is no universal number because rules differ, but a useful target is 95 percent or better on planning-critical fields such as lead time, planner code, and unit of measure, and 85 percent on descriptive attributes. What matters more than the absolute score is the trend and the exception backlog age. A stable 92 percent with a shrinking backlog is healthier than a 97 percent snapshot nobody maintains.

How do I find duplicate item numbers in ERP?

Exact matching finds almost nothing because duplicates differ in formatting. Use fuzzy matching on normalized manufacturer part number, then blocking on commodity code plus description similarity, and compare specification attributes. Score each candidate pair and route high-confidence pairs for human review. Never auto-merge item masters: merges affect BOMs, open orders, and historical costing, so they need engineering approval and a documented sequence.

Who should own ERP data quality, IT or the business?

The business owns the data; IT owns the plumbing that measures it. Assign a named steward per domain, typically planning or engineering for item master, sales operations for customers, and procurement for suppliers. IT provides profiling, automated rules in the pipeline, exception queues, and the scorecard. Programs that put IT in charge of the actual values consistently stall because IT cannot decide what a correct lead time is.

Key Takeaways

1Profile Before You Prescribe: Start with a two-week profiling pass rather than a policy document. Run column-level statistics across the master tables that drive planning: item, itemwhse, bom or job routing, customer, vendor, and the price and cost tables.
2The Six Dimensions Applied to Manufacturing Data: Use the DAMA dimensions, but express them in shop-floor terms so business owners can argue with them. Completeness asks whether every active item has a lead time, planner code, commodity code, and ABC class.
3Automated Rules and Where to Run Them: Rules belong in two places. Preventive rules run at entry: a required-field policy on the item master form, a validation extension that rejects a purchase UOM inconsistent with the stocking UOM, or a workflow approval for any standard cost change above a threshold.