A dashboard can be technically correct and still describe the market incorrectly if the product master underneath it contains weak categories or unsafe SKU links.
Audit before the error becomes routine
Catalog issues are often discovered late: a segment changes unexpectedly, a known product cannot be found in its category, one brand is fragmented across spellings, or a “confirmed” match contains a different size or processing method. An audit exposes these risks before they become recurring manual corrections or reporting disputes.
Step 1. Define what the catalog will support
Different business uses require different critical attributes. A useful audit begins by establishing the decision that the catalog is expected to support.
| Use case | Critical fields |
|---|---|
| Category reporting | Domain, category, brand, product type |
| Pricing and promotion analysis | Weight or volume, pack, format, unit count |
| RGM analysis | Segment, price tier, size, variant, comparable products |
| PIM onboarding | Complete controlled attributes and standards |
| Product matching | Differences that must prevent false identity |
Step 2. Assemble the minimum input file
An initial review normally requires an Excel or CSV extract with product ID, original title, current category or domain where available, existing attribute fields and any links to another master or retailer catalog. Domain definitions or attribute dictionaries improve the diagnostic, but the title and present assignments already reveal substantial quality issues.
Step 3. Inspect source-title quality
Before classification, verify whether the input names carry sufficient signal for automation. Look for missing titles, technical placeholders, duplicates differentiated only by punctuation, unstable measurement formats, unexplained abbreviations and mixed-language or mixed-alphabet tokens.
A frequency table of unparsed tokens and recurring patterns shows where a controlled dictionary or rule would create value, and where rare or ambiguous wording should remain in review.
Step 4. Parse names into reviewable attributes
Even when a production parser does not yet exist, a preliminary attribute view makes audit decisions explicit.
| Title | Category | Brand | Size / Pack | Status |
|---|---|---|---|---|
| Cheese Belebeevskiy creamy 45% sliced 190g | Cheese | Belebeevskiy | 190 g / slices | Type requires confirmation |
| Milk Vkusnoteevo pasteur. 3.2% jug 930g | Milk | Vkusnoteevo | 930 g / jug | Processing explicitly supported |
The audit should never claim more detail than the title and standards support. Separating confident fields from uncertain ones is essential.
Step 5. Review domain boundaries
A usable domain reference includes a clear definition, positive examples, product anchors, exclusions and guidance for boundary cases. It should prevent an overly broad token from driving the classification. Terms such as “dairy”, “children”, “sensitive” or “clean” may be useful signals, but only within category-specific context.
Look for overlapping definitions, products that fit multiple domains, categories with many unanchored assignments, rules based on vague single words and frequent errors that lack an exclusion rule.
Step 6. Separate false assignments from missed recognition
False assignments
A wrong domain, brand or type creates a filled field that users may trust. These issues are the priority: they contaminate analysis while appearing complete.
Missed recognition
A visible weight, packaging type or claim that is not extracted reduces coverage and suggests a safe dictionary or pattern enhancement. These issues are important, but usually follow false-positive control.
Step 7. Verify SKU links between sources
When multiple retailer files or a retailer-to-master mapping are involved, confirm that every high-confidence link agrees on critical identity fields.
| Field | Reason for validation |
|---|---|
| Category and product type | Distinct product types should not be joined. |
| Brand and line | Similar spelling does not always identify the same product. |
| Weight or volume | Pack size affects SKU identity and pricing analysis. |
| Packaging and variant | Pack or flavour can represent a separate SKU. |
| Technology or formulation | Pasteurized versus UHT, powder versus gel and similar distinctions matter. |
Classify links into confirmed, suspicious and rejected groups, with concise reason codes for review.
Step 8. Turn findings into a correction plan
| Problem class | Typical correction |
|---|---|
| Wrong domain driven by a general token | Refine anchors and add exclusions. |
| Fragmented brand spelling | Add controlled aliases to a standard dictionary. |
| Missed size or volume | Apply tested extraction and unit normalization rules. |
| Unsafe matched pairs | Tighten critical-attribute checks for matching. |
| Large uncertain group | Create a review queue by confidence and issue pattern. |
Step 9. Fix metrics for repeated control
Track the share of confidently categorized rows, coverage of critical attributes, number of suspicious assignments, confirmed and rejected mappings, recurring unparsed patterns and regressions after rule updates. “More fields filled” should not be treated as quality unless correctness is controlled.
Final pre-analysis checklist
- The business purpose and critical attributes are defined.
- Original SKU titles and identifiers are preserved.
- Size, volume, packaging and key abbreviation formats have been reviewed.
- Titles have been parsed into reviewable candidate attributes.
- Domains include anchors, exclusions and boundary-case guidance.
- False assignments and missed recognitions are separated.
- Existing cross-source matches have been validated.
- Issues are prioritized by analytical risk.
- Correction rules and quality metrics are documented for the next iteration.
Check the quality of your SKU data.
Send an Excel or CSV sample with product titles and available classifications. Indicuma will prepare a written snapshot of visible issues, ambiguity patterns and priority recommendations.
Get a SKU Data Quality Snapshot →