--- # === IDENTITY === id: consulting/recipes/retail-data-infrastructure-audit/2026 canonical_question: "How do you audit retail data infrastructure: signal sources, POS latency, knowledge graphs?" aliases: - "Retail data infrastructure maturity assessment" - "POS-to-analytics pipeline latency audit" - "How to assess retail AI data readiness and knowledge graph maturity" entity_type: execution_recipe domain: consulting > recipes > Retail Data Infrastructure Audit region: global jurisdiction: global temporal_scope: 2026-2027 # === VERIFICATION === last_verified: 2026-03-30 confidence: 0.85 version: 1.0 first_published: 2026-03-30 # === TEMPORAL VALIDITY === temporal_validity: status: evolving last_breaking_change: "Initial release — Retail Data Infrastructure Audit methodology v1.0" next_review: 2026-09-26 change_sensitivity: high # === CONSTRAINTS === constraints: - "POS system read-only access required — never write to production POS databases" - "Customer PII must be anonymized before any analysis — hash customer IDs, strip names/addresses" - "Latency measurements require minimum 7-day observation window to capture weekly seasonality" - "Knowledge graph maturity scoring requires access to product catalog API or structured product feed" - "GEO audit must use current AI models (ChatGPT, Perplexity, Google AI Overviews) — results are model-version-dependent" # === SKIP CONDITIONS === skip_this_unit_if: - condition: "User needs the full retail AI diagnostic, not just data infrastructure" use_instead: "consulting/recipes/retail-ai-diagnostic-engagement-playbook/2026" - condition: "User needs a general data quality assessment, not retail-specific" use_instead: "business/product-tech/data-strategy-assessment/2026" - condition: "User needs supply chain data audit only, not full retail data infrastructure" use_instead: "Search knowledgelib.io for supply chain data audits — no dedicated unit yet" # === AGENT HINTS === inputs_needed: - key: pos_system question: "What POS system does the retailer use?" type: choice options: ["Shopify POS", "Square", "Lightspeed", "Oracle MICROS", "NCR Voyix", "Toshiba/Global", "custom/proprietary", "multiple/mixed"] - key: data_access_level question: "What level of data access has been granted?" type: choice options: ["full API access", "read-only API access", "CSV/flat file export only", "dashboard access only", "not yet negotiated"] - key: ecommerce_platform question: "What e-commerce platform is in use?" type: choice options: ["Shopify", "Magento/Adobe Commerce", "Salesforce Commerce Cloud", "SAP Commerce", "custom-built", "no e-commerce", "not sure"] - key: product_catalog_size question: "How many SKUs are in the product catalog?" type: choice options: ["< 1,000", "1,000-10,000", "10,000-100,000", "100,000+"] # === EXECUTION METADATA === execution: required_inputs: - name: "POS system access (API credentials or data export)" source: "client/IT-department" format: "API credentials or CSV export" - name: "Product catalog feed or API" source: "client/merchandising or IT" format: "API endpoint or structured data file (JSON/XML/CSV)" - name: "Technology stack inventory" source: "client/IT-department" format: "spreadsheet" - name: "Supply chain system list" source: "client/operations or supply chain team" format: "spreadsheet" outputs: - name: "Data Infrastructure Scorecard" format: "PDF + structured JSON" description: "7-sub-dimension scorecard: demand signal inventory, POS latency, supply chain integration, knowledge graph maturity, GEO readiness, real-time ratio, overall maturity level (1-5)" - name: "Pipeline Latency Report" format: "spreadsheet + visualization" description: "Measured latency for each data pipeline from source to analytics/decision layer with bottleneck identification" - name: "Knowledge Graph Maturity Assessment" format: "document" description: "Product taxonomy depth, attribute coverage, semantic embedding readiness, and gap inventory" tools_required: - name: "Data profiling tool" purpose: "Analyze data quality, completeness, and schema across POS and supply chain sources" tier: "free" cost: "$0" alternatives: ["Great Expectations", "pandas-profiling", "Ataccama", "Talend"] - name: "POS analytics connector" purpose: "Extract transaction data and measure pipeline latency" tier: "varies" cost: "$0-$500 (depends on POS vendor API)" alternatives: ["Direct API integration", "Fivetran connector", "custom Python script"] - name: "Knowledge graph analysis tool" purpose: "Assess product taxonomy structure, attribute coverage, and embedding readiness" tier: "free" cost: "$0" alternatives: ["Neo4j", "NetworkX", "custom SPARQL queries", "spreadsheet analysis"] - name: "AI search tools" purpose: "GEO audit — test product discoverability in AI-generated answers" tier: "free" cost: "$0-$20" alternatives: ["ChatGPT", "Perplexity", "Google AI Overviews", "Bing Copilot"] credentials_needed: - service: "POS system" type: "Read-only API key or export access" where_to_get: "Client IT department" free_tier_limits: "N/A — requires client authorization" - service: "Product catalog API" type: "API key or feed URL" where_to_get: "Client merchandising/IT team" free_tier_limits: "N/A — internal system" estimated_duration: "3-5 days" estimated_cost: "$3K-$6K (as standalone) or included in $20K diagnostic engagement" # === DISTRIBUTION === canonical_source: "https://knowledgelib.io/consulting/recipes/retail-data-infrastructure-audit/2026" suggested_citation: "Source: knowledgelib.io — AI Knowledge Library (verified 2026-03-30)" # === RELATED UNITS === related_kos: depends_on: [] feeds_into: - id: "consulting/recipes/retail-ai-diagnostic-engagement-playbook/2026" label: "Master diagnostic engagement — this audit produces Dimension 1" related_to: - id: "business/retail-transformation/retail-digital-maturity-assessment/2026" label: "Broader retail digital maturity benchmark" # === SOURCES === sources: - id: src1 title: "DAMA-DMBOK: Data Management Body of Knowledge" author: DAMA International url: https://www.dama.org/cpages/body-of-knowledge type: industry_report published: 2017-07-01 reliability: authoritative - id: src2 title: "The AI-First Retailer: Competing in the Age of Intelligence" author: McKinsey & Company url: https://www.mckinsey.com/industries/retail/our-insights/the-next-frontier-of-retail type: industry_report published: 2025-06-15 reliability: authoritative - id: src3 title: "Knowledge Graphs in Retail: Product Discovery and Recommendation" author: Noy, N. & Gao, Y. (Google Research) url: https://research.google/pubs/pub48160/ type: academic_paper published: 2019-05-01 reliability: high - id: src4 title: "Real-Time Analytics in Retail: From Batch to Stream" author: Confluent url: https://www.confluent.io/resources/report/data-streaming-retail/ type: industry_report published: 2025-02-10 reliability: high - id: src5 title: "Generative Engine Optimization: Search in the Age of AI" author: Aggarwal, P. et al. url: https://arxiv.org/abs/2311.09735 type: academic_paper published: 2024-03-15 reliability: high --- # Retail Data Infrastructure Audit ## Purpose This recipe executes a structured audit of retail data infrastructure across 7 sub-dimensions: demand signal inventory, POS-to-analytics pipeline latency, supply chain data integration, product knowledge graph maturity, AI retrieval readiness (GEO), real-time vs batch decision-making ratio, and overall maturity benchmarking. It produces a scored data infrastructure report that feeds Dimension 1 of the Retail AI Diagnostic Engagement. [src1, src2] ## Prerequisites - [ ] **POS system access** — read-only API credentials or trailing 90-day transaction data export - [ ] **Product catalog feed** — structured product data (API endpoint, XML feed, or CSV export with attributes) - [ ] **Technology stack inventory** — list of all data systems: POS, ERP, WMS, CRM, e-commerce, analytics, CDP - [ ] **Supply chain system list** — EDI partners, supplier portals, logistics platforms, warehouse systems - [ ] **Data processing agreement signed** — covers POS transaction data and product catalog access - [ ] **IT point of contact** identified for system access questions during audit ## Constraints - POS system access must be read-only — never write to production POS databases under any circumstances. [src1] - Customer PII must be anonymized before any analysis — hash customer IDs, strip names, addresses, and payment details. Aggregate to transaction level. - Latency measurements require a minimum 7-day observation window to capture weekly seasonality patterns (weekday vs weekend, restock cycles). Single-day measurements are unreliable. [src4] - Knowledge graph maturity scoring requires access to the product catalog API or a structured product feed — cannot be scored from the public-facing website alone. [src3] - GEO audit results are model-version-dependent — document which AI model versions were used and the test date. Results may change with model updates. [src5] ## Tool Selection Decision ``` Which path? ├── Retailer has modern cloud POS with API (Shopify POS, Square, Lightspeed) │ └── PATH A: API-Direct — real-time latency measurement, automated profiling ├── Retailer has legacy POS (Oracle MICROS, NCR Voyix, Toshiba) │ └── PATH B: Export-Based — CSV/flat file analysis, manual latency estimation ├── Retailer has multiple POS systems (acquisitions, regional variance) │ └── PATH C: Multi-Source — audit each system, reconciliation analysis └── Retailer has only dashboard access (no raw data) └── PATH D: Interview-Based — structured questionnaire, proxy metrics ``` | Path | Tools | Cost | Speed | Output Quality | |------|-------|------|-------|---------------| | A: API-Direct | POS API, Python profiling, automated latency tools | $0-$200 | 3 days | Excellent — measured, not estimated | | B: Export-Based | CSV analysis, pandas-profiling, manual timing | $0 | 4-5 days | Good — thorough but estimated latency | | C: Multi-Source | Multiple connectors, reconciliation scripts | $0-$500 | 5 days | Good — captures system fragmentation | | D: Interview-Based | Structured questionnaire, proxy metrics | $0 | 2-3 days | Adequate — estimated, not measured | ## Execution Flow ### Step 1: Demand Signal Source Inventory **Duration**: 0.5-1 day **Tool**: Spreadsheet + IT stakeholder interview Inventory every data source that produces demand signals — information the retailer uses (or could use) to predict, respond to, or influence customer demand. Catalog each source with: - **Source name** (e.g., "Shopify POS transactions", "Google Analytics 4", "weather API") - **Signal type** (transactional, behavioral, environmental, social, competitive) - **Refresh rate** (real-time, hourly, daily, weekly, manual) - **Data format** (API, CSV, database table, dashboard) - **Current usage** (active in decisions, collected but unused, not collected) - **Integration status** (integrated into analytics stack, standalone, siloed) Target: identify 15-30 signal sources for a mid-size retailer. Common sources often missed: weather data, local event calendars, social media sentiment, competitor pricing feeds, Google Trends, customer service ticket topics. [src2] **Verify**: Inventory spreadsheet complete with 15+ sources identified. Each source has refresh rate and integration status documented. **If failed**: If IT cannot provide a comprehensive list, supplement with stakeholder interviews and public documentation review. Flag undiscovered sources as a gap. ### Step 2: POS-to-Analytics Pipeline Latency Measurement **Duration**: 1-2 days (plus 7-day observation if measuring live) **Tool**: POS API or export + timestamp analysis Measure the end-to-end latency from transaction occurrence at the point of sale to data availability in the analytics/decision layer. This is the single most important metric for AI readiness — AI models that operate on stale data produce stale decisions. **For API-accessible POS (Path A)**: 1. Record a test transaction timestamp at the POS terminal 2. Query the POS API repeatedly until the transaction appears 3. Query the analytics/BI platform until the same transaction appears 4. Calculate: POS to API latency, API to analytics latency, total end-to-end latency 5. Repeat across 7 days at different times (peak hours, overnight, weekend) **For export-based POS (Path B)**: 1. Document the export schedule (how often, when triggered, format) 2. Review export file timestamps vs transaction timestamps 3. Document ETL pipeline processing time 4. Calculate estimated end-to-end latency from transaction to analytics availability **Benchmark against maturity levels**: [src4] - Level 1: > 24 hours (batch daily) - Level 2: 4-24 hours (multiple daily batches) - Level 3: 1-4 hours (near-real-time) - Level 4: 1-60 minutes (streaming with micro-batches) - Level 5: < 1 minute (true real-time streaming) **Verify**: Latency measured or estimated for the complete POS-to-analytics pipeline. Maturity level assigned with evidence. **If failed**: If live measurement is not possible, use export schedules and ETL logs to estimate. Document as "estimated" vs "measured" in the report. ### Step 3: Supply Chain Data Integration Assessment **Duration**: 0.5-1 day **Tool**: System inventory + integration mapping Assess how completely supply chain data flows into the retailer's analytics and decision-making systems. Map data flows across 6 supply chain domains: 1. **Inventory positions** — across all locations (stores, DCs, in-transit). Real-time or batch? Unified view or fragmented? 2. **Supplier data** — lead times, fill rates, cost changes. How integrated? EDI or manual? 3. **Logistics/shipping** — carrier tracking, delivery estimates. API-integrated or manual lookup? 4. **Warehouse operations** — WMS data (pick rates, dock-to-stock time). Integrated into planning? 5. **Returns/reverse logistics** — return data fed back into demand planning? 6. **Demand planning signals** — are supply chain constraints visible to demand planners? For each domain, score: - **Completeness** (1-5): How much of the data is captured digitally? - **Integration** (1-5): How well does it flow into analytics without manual steps? - **Timeliness** (1-5): How current is the data when it reaches decision-makers? **Verify**: All 6 supply chain domains mapped with completeness, integration, and timeliness scores. **If failed**: If supply chain team is unavailable, use IT system inventory and stakeholder interviews to estimate. Flag gaps. ### Step 4: Product Knowledge Graph Maturity Assessment **Duration**: 0.5-1 day **Tool**: Product catalog analysis + knowledge graph tools Assess the retailer's product data structure and readiness for AI-powered discovery and recommendation. [src3] Evaluate 5 knowledge graph dimensions: 1. **Taxonomy depth** — How many levels in the product category hierarchy? (1-2 = flat, 3-4 = standard, 5+ = rich) 2. **Attribute coverage** — What percentage of SKUs have complete, structured attributes (size, color, material, use case, compatibility)? 3. **Relationship mapping** — Are product relationships captured (complementary, substitute, accessory, bundle)? 4. **Semantic richness** — Are product descriptions structured enough for embedding? (keyword-only vs natural language vs structured + unstructured) 5. **Entity resolution** — Can the same product be identified across channels (in-store SKU = e-commerce product = supplier item)? **Benchmark**: - Level 1: Flat catalog, minimal attributes, no relationships - Level 2: 2-3 level taxonomy, basic attributes, no relationships - Level 3: Multi-level taxonomy, 70%+ attribute coverage, some relationships - Level 4: Rich taxonomy, 90%+ attributes, relationship graph, embedding-ready - Level 5: Full knowledge graph with semantic embeddings, cross-channel entity resolution **Verify**: Knowledge graph maturity level assigned (1-5) with specific evidence for each dimension. Attribute coverage calculated as a percentage. **If failed**: If product catalog API is unavailable, audit the public-facing website for taxonomy depth and attribute coverage. Note reduced accuracy. ### Step 5: AI Retrieval Readiness (GEO Audit) **Duration**: 0.5-1 day **Tool**: ChatGPT, Perplexity, Google AI Overviews, Bing Copilot Evaluate how well the retailer's product and brand content surfaces in AI-generated answers — the emerging discovery channel that will reshape retail search. [src5] Execute 20-30 test queries across 3 categories: 1. **Product discovery queries** (10): "best [category] for [use case]", "what [product type] should I buy for [scenario]" 2. **Brand/store queries** (10): "[retailer name] [product category]", "does [retailer] carry [specific product]" 3. **Comparison queries** (10): "[retailer] vs [competitor] for [category]", "where to buy [product] cheapest" For each query, record: - Which AI model was tested (GPT-4o, Perplexity, Gemini, etc.) - Whether the retailer was mentioned in the response - Position in the response (first mention, middle, end, not present) - Whether product data was accurate (correct prices, availability, attributes) - Source attribution (was the retailer's website cited?) **Score**: (mentions / total queries) x 100 = GEO visibility score. Benchmark: < 20% = critical gap, 20-40% = below average, 40-60% = average, 60-80% = good, > 80% = excellent. **Verify**: GEO audit spreadsheet complete with 20+ queries tested, visibility score calculated, and specific gap examples documented. **If failed**: If AI tools are unavailable or blocked, use Google Search "AI Overview" results as the primary source. Document which tools were used. ### Step 6: Real-Time vs Batch Decision-Making Ratio **Duration**: 0.5 day **Tool**: Decision inventory + system analysis Calculate the ratio of decisions made with real-time data vs batch data. This reveals how much of the retailer's decision-making is constrained by data staleness. Inventory 10 key retail decisions and classify each: 1. Pricing changes (real-time / daily batch / weekly manual) 2. Inventory replenishment triggers (real-time / daily / weekly) 3. Markdown decisions (real-time / weekly / seasonal) 4. Labor scheduling adjustments (real-time / weekly) 5. Promotional targeting (real-time / campaign-based) 6. Fraud detection (real-time / batch review) 7. Customer service routing (real-time / manual) 8. Assortment changes (real-time / quarterly) 9. Supply chain rerouting (real-time / manual) 10. Customer personalization (real-time / segment-based) **Score**: Count of real-time decisions / total decisions = real-time ratio. - Level 1: 0-10% real-time (almost entirely batch/manual) - Level 2: 10-25% real-time - Level 3: 25-50% real-time - Level 4: 50-75% real-time - Level 5: 75%+ real-time [src4] **Verify**: All 10 decisions classified with evidence. Real-time ratio calculated. **If failed**: If decision process documentation is unavailable, classify based on stakeholder interviews and system capabilities. ### Step 7: Maturity Level Benchmarking and Report Generation **Duration**: 0.5-1 day **Tool**: Scorecard template + analysis synthesis Synthesize all 6 preceding steps into a composite Data Infrastructure Maturity Score: | Sub-Dimension | Weight | Score (1-5) | Evidence Summary | |---------------|--------|-------------|-----------------| | Demand signal inventory (completeness) | 15% | {score} | {sources found} / {expected} | | POS-to-analytics latency | 25% | {score} | {measured latency} | | Supply chain data integration | 15% | {score} | {avg across 6 domains} | | Product knowledge graph maturity | 20% | {score} | {attribute coverage %, taxonomy depth} | | GEO readiness (AI retrieval) | 15% | {score} | {visibility score %} | | Real-time decision ratio | 10% | {score} | {ratio %} | Calculate weighted composite score. Assign overall maturity level (1-5). Produce final report: - Composite score with breakdown per sub-dimension - Critical gaps ranked by AI readiness impact - Quick win recommendations (achievable in 0-3 months) - Infrastructure investment recommendations (3-12 months) - Comparison to industry benchmarks where available [src2] **Verify**: All 7 sub-dimensions scored with evidence. Composite score calculated. Report produced with gap analysis and recommendations. **If failed**: If any sub-dimension has insufficient data, score as "incomplete" with confidence flag. Provide estimated range based on available proxy data. ## Output Schema ```json { "output_type": "retail_data_infrastructure_scorecard", "format": "PDF + JSON", "sections": [ {"name": "composite_maturity_score", "type": "number", "description": "Weighted composite data infrastructure maturity score 1-5", "required": true}, {"name": "sub_dimension_scores", "type": "array", "description": "7 sub-dimension scores with weight, evidence, and gaps", "required": true}, {"name": "pipeline_latency_report", "type": "object", "description": "Measured or estimated latency per data pipeline with bottleneck identification", "required": true}, {"name": "knowledge_graph_assessment", "type": "object", "description": "5-dimension KG maturity with attribute coverage percentage", "required": true}, {"name": "geo_audit_results", "type": "object", "description": "AI visibility score with per-query results and gap examples", "required": true}, {"name": "gap_analysis", "type": "array", "description": "Ranked infrastructure gaps by AI readiness impact", "required": true}, {"name": "recommendations", "type": "array", "description": "Quick wins (0-3mo) + infrastructure investments (3-12mo)", "required": true} ], "expected_sections": "7", "sort_order": "gap impact descending" } ``` ## Quality Benchmarks | Quality Metric | Minimum Acceptable | Good | Excellent | |---------------|-------------------|------|-----------| | Demand signal sources identified | > 10 | > 15 | > 25 | | POS latency measurement method | Estimated from logs | Sampled (3+ days) | Measured (7+ days) | | Supply chain domains assessed | 4/6 | 5/6 | 6/6 | | Product attribute coverage measured | Estimated from sample | Measured on 50%+ SKUs | Measured on 90%+ SKUs | | GEO audit queries executed | 10+ queries | 20+ queries | 30+ queries | | Real-time decisions classified | 6/10 | 8/10 | 10/10 | **If below minimum**: Extend audit by 1-2 days. Prioritize measured over estimated metrics. Request additional system access from IT. ## Error Handling | Error | Likely Cause | Recovery Action | |-------|-------------|----------------| | POS API returns no data or errors | Credentials insufficient or API rate-limited | Verify credential scope with IT, request elevated read-only access, fall back to CSV export | | Product catalog feed is incomplete | Feed excludes certain categories or attributes | Document coverage gaps, supplement with public website scraping for missing categories | | Latency measurement shows inconsistent results | Caching, CDN, or ETL scheduling artifacts | Extend measurement window to 14 days, sample at different hours, document variability | | Knowledge graph assessment impossible (no structured data) | Retailer has unstructured product descriptions only | Score as Level 1, document the gap, recommend taxonomy and attribute enrichment as quick win | | GEO audit blocked by corporate firewall | AI tools restricted on corporate network | Run GEO audit from external network, use personal accounts for testing | | Supply chain team unresponsive | Different reporting structure or engagement fatigue | Escalate to executive sponsor, offer 30-minute focused interview instead of full audit | ## Cost Breakdown | Component | Standalone | Within Diagnostic | At Scale (multi-region) | |-----------|-----------|-------------------|------------------------| | Demand signal inventory | $500-$800 | Included | $800-$1.2K per region | | POS latency measurement | $800-$1.5K | Included | $1.5K-$2.5K per region | | Supply chain assessment | $500-$800 | Included | $800-$1.2K per region | | Knowledge graph audit | $500-$1K | Included | $1K-$2K per region | | GEO audit | $300-$500 | Included | $500-$800 per region | | Real-time ratio + report | $400-$700 | Included | $700-$1K per region | | **Total** | **$3K-$6K** | **Included in $20K engagement** | **$5K-$9K per region** | ## Anti-Patterns ### Wrong: Measuring latency once and extrapolating Taking a single POS latency measurement on a Tuesday afternoon and using it as the definitive metric. Result: misses the 3x latency spike during Saturday peak hours and the overnight batch delay that makes Monday morning data 18 hours stale. [src4] ### Correct: Measure across a full weekly cycle Sample latency at minimum 4 time points per day across 7 days. Report the P50, P95, and P99 latency, not just the average. Peak latency determines AI decision quality during the moments that matter most. ### Wrong: Scoring knowledge graph maturity from the website alone Evaluating product data richness by browsing the public-facing e-commerce site. Result: overestimates maturity because the website shows curated data while the backend catalog has 40% of SKUs with missing attributes. [src3] ### Correct: Audit the source catalog, not the presentation layer Request the raw product feed or catalog API export. Calculate attribute coverage across all SKUs, not just featured products. The gap between website presentation and backend reality is itself a critical finding. ### Wrong: Ignoring the demand signals that are not collected Only inventorying data sources that already exist in the analytics stack. Result: misses the 5-10 high-value signal sources (weather, events, social sentiment, competitor pricing) that would unlock the next maturity level. [src2] ### Correct: Inventory what should exist, not just what does Start with a reference list of 25-30 retail demand signal types. Check each one. The uncollected signals often have the highest marginal value because competitors are not using them either. ## When This Matters Use when an agent needs to audit retail data infrastructure as a standalone engagement or as Dimension 1 of the Retail AI Diagnostic Engagement Playbook. This recipe produces a scored, evidence-based assessment of 7 data infrastructure sub-dimensions that directly determine AI deployment readiness. Requires POS system access and product catalog access as prerequisites. ## Related Units - [Retail AI Diagnostic Engagement Playbook](/consulting/recipes/retail-ai-diagnostic-engagement-playbook/2026) - [Retail Digital Maturity Assessment](/business/retail-transformation/retail-digital-maturity-assessment/2026) - [Retail AI Strategy](/consulting/strategy/retail-ai-strategy/2026)