Syntha Lab RU

Analyst (BI)

Turns data into verifiable conclusions: defines metrics, builds reports, measures the effect of decisions and guards data quality.

Brand and retail

Open interactively →

About the role

The analyst, including the BI specialist, turns data into verifiable conclusions: defines metrics, builds reports, measures the effect of decisions and watches data quality. The main principle is that a number without a definition and base means nothing. So the first task is not a chart but a metric dictionary and agreement on what counts as an order, sale, stock and return.

The analyst works at the junction of all functions: from purchasing taking lead times and OTIF, from sales pace and tickets, from marketing acquisition and retention, from finance margin and cash. The analyst answers for forecast accuracy and can compare it with a naive baseline to see whether the model creates value. They distinguish baseline from effect, correlation from cause, and for assessing a promotion or innovation propose a control group.

The occupational hazard of the role is beautiful dashboards on bad data. So the analyst is the first to notice reference problems and becomes an ally of data owners.

A working day

The day begins with a check that yesterday’s data loaded and reconciles with the ledger: discrepancies in sales, stock, prices. Then regular reports and requests from functions. In the afternoon the analyst investigates variances and prepares explanations, not just numbers: what changed, by how much and why. In the evening refines models and report specifications and records definition changes.

Documents: the metric dictionary, report specifications, a log of rule changes, reports and dashboards, exports from accounting, tills and platforms, a data-quality log, reports on the effect of experiments and promotions. The analyst also ensures any figure can be explained: from which system, over which period, by which definition.

Typical decisions

  • Which ten metrics to treat as primary and how to fix their definitions.
  • Whether a forecast is good enough to use, compared with a naive baseline.
  • How to assess the effect of a promotion or change: baseline, control group, observation window.
  • Which report to build first and which data fields to close to make it work.
  • Which data-quality dimension to fix first and who is responsible.
  • Which tolerance (traffic light) to set for a metric and what to do in the red zone.

Study path

  1. The metric dictionary: definitions, bases, sources and owners. KPI dictionary and common definitions
  2. Forecast accuracy by week: window, bias and value added against a naive forecast. Forecast accuracy by week: window, bias and forecast value added
  3. S&OP and forecast bias: how to catch a systematic skew. S&OP: forecast bias, error and the balance of demand and supply
  4. The report data model: facts, dimensions, row unit, price and category history. Data model of standard reports
  5. Data readiness for standard reports: close the field with the biggest shortfall. Data readiness for standard reports
  6. Data quality: four dimensions, a weighted score and stop conditions. Data quality and data owners
  7. Master data: what lives at which level and where keys come from. Master data: SKU, SKU, barcode
  8. The effect of a change by difference-in-differences: comparison with a control. Control group: promotion effect (difference-in-differences)

What to understand

Metric dictionary
Every metric has one definition: numerator and denominator, with or without VAT, units or money, which date. Differences between departments are almost always definitions.
Gross and net sales
Gross is sales before returns and discounts, net after. Always label which sales a report shows and bring them to one base before comparing.
Baseline and effect
The effect of a promotion or change is measured against a control group or baseline, not against last week. Difference-in-differences and holdout protect against seasonality.
Forecast accuracy
Forecast error (WAPE) and bias show whether the plan can be trusted. One-sided bias is worse than random error: it leads to systematic over-buying or shortage.
Reference books and data quality
Unified reference books for sizes, attributes and categories make reports comparable. Check duplicate SKUs, empty attributes and mismatched units.
Tolerances and traffic lights
A deviation matters if it exceeds a tolerance set in advance. Without a tolerance a traffic light paints random noise.
Scraping and external data
Competitor prices and market estimates must be brought to one base and date. Scraping gives shelf prices, not sales: do not confuse them.
CRM metrics and coverage
Active, repeat, new and returned customers, cohort retention and CRM coverage: how many sales you see and can reach. Without coverage, cohort metrics mislead.
Store: pace, receipts, the sales equation
A single scheme to explain sales change over four factors, period pace and receipt bands. They explain a deviation before the final number.
Season results and the report dictionary
The season ladder, sell-through by delivery, style classes, year on year. Every report (rep-*) has a spec: fields, filters, period.
Sensitivity and scenarios
Tornado and plan scenarios show which assumptions cost most. That sets the priority for data collection.
Master data: style, color-style, SKU, barcode
Each field lives at its level: style, color-style, SKU, pack. Define the table grain (what one row means) before loading: it decides what can be added and compared at all.
Data quality: completeness, uniqueness, validity, timeliness
Four checks give an overall quality score, and the data owner is accountable for fixing. Score the fields reports need, not all fields.
Common measures and report readiness
The measures dictionary fixes the definition, source, owner and frequency. Check data sufficiency before a report: otherwise a figure looks precise but cannot be reproduced.
Data exchange templates
The product, stock, sales, order, supplier and customer templates set columns, types and validation rules. They are defaults; take required marking and EDI fields from the operator.

Key measures

MeasureFormulaHow to read
Forecast error (WAPE)Sum of absolute errors ÷ sum of actual salesCount by category and horizon.
Forecast biasSum (forecast − actual) ÷ sum of actualsA positive value means systematic over-forecast.
Promotion upliftPromotion actual − baselineBaseline must be estimated before the promotion.
Effect versus control(Test after − before) − (control after − before)Requires a comparable control group.
Share of correct product cardsCards without category or attribute errors ÷ checkedNo threshold: decide by impact on reports.
Indicator traffic lightDeviation from plan versus toleranceThe tolerance is set by the metric owner.
CRM coverageSales with an identified customer ÷ all salesLow coverage limits every cohort conclusion.
Cohort retention at a periodReturned cohort customers ÷ cohort customersCompare cohorts at one horizon.
Contribution by season ladder stepStep amount ÷ gross salesShows which step converts turnover into contribution worst.
Plan sensitivityChange in result when an assumption moves by a set stepTake assumptions with the highest sensitivity first.
Data quality scoreAverage of completeness, uniqueness, validity and timeliness of the needed fieldsSet weights and tolerance yourself; read each reference book separately.
Report readinessReports with all needed fields ÷ all standard reportsShows which field to fix first.
Data exchange lagTime from an event in the source to its appearance in the reportFor stock and prices the price of lag is higher than for reference books.
Forecast accuracy with valueForecast error versus a naive forecast (what the method added)If the method is no better than a naive one, the complexity is not justified.

Questions the role asks

Route through Lab

  1. Forecast and budget
  2. Sales forecast: method, horizon, error
  3. S&OP: forecast bias, error and the balance of demand and supply
  4. Promotion result: incremental profit
  5. Control group: promotion effect (difference-in-differences)
  6. Channel incrementality: test against control
  7. How to read market figures: bringing estimates to one basis
  8. Product classifier check
  9. Owner dashboard: 10 indicators and a traffic light
  10. Variance tolerance and traffic light
  11. Baseline (sales without a promotion)
  12. WACC: cost of capital, ROIC against WACC and economic profit
  13. Consolidation of two entities: eliminating turnover and unrealized profit
  14. Management balance sheet from the P&L and turnover figures
  15. Forecast value added by stage and the price of error in roubles: the season assortment
  16. Driver decomposition of change (LMDI)
  17. CRM and the customer base: retention, segments, win-back
  18. Customer base: active, repeat, new and returning
  19. Cohort retention: curve, churn and cohort LTV
  20. CRM coverage: how many sales and customers you can see and reach
  21. Store sales and receipts: traffic, conversion, ticket
  22. Sales equation: traffic, conversion, units per receipt, price per unit
  23. Season results: brand, category, SKU
  24. Season result: from gross sales to contribution
  25. Brand history by season: growth, discounts, margin and sell-through
  26. Scenarios and plan stress test
  27. Plan sensitivity and breaking point
  28. The language of commerce: order, stock, reports
  29. Master data: SKU, SKU, barcode
  30. SKUs, barcodes and the labor of product cards
  31. Data quality and data owners
  32. Completeness and quality of product master data
  33. Systems landscape and data exchange
  34. Stock sync latency: the price of stale data
  35. KPI dictionary and common definitions
  36. Forecast accuracy by week: window, bias and forecast value added
  37. Merchandise plan traffic light
  38. Data model of standard reports
  39. Data readiness for standard reports
  40. From the product card to ready reports
  41. Common process measures

First 30 days

  1. Read the rep-* terms and compare with your reports. Season results report
  2. Compute CRM coverage. CRM coverage: how many sales and customers you can see and reach
  3. Build cohort retention. Cohort retention: curve, churn and cohort LTV
  4. Decompose store sales by the equation. Sales equation: traffic, conversion, units per receipt, price per unit
  5. Build the season ladder. Season result: from gross sales to contribution
  6. Build the plan sensitivity tornado. Plan sensitivity and breaking point
  7. Describe the master data grain and levels. Master data: SKU, SKU, barcode
  8. Score the reference books and name owners. Data quality and data owners
  9. Build the common measures dictionary. KPI dictionary and common definitions
  10. Check data readiness for standard reports. Data readiness for standard reports
  11. Walk the path from product card to ready reports. From the product card to ready reports

Common mistakes

  • Changing a metric definition between reports without a note.
  • Comparing a promotion with last week instead of a control group.
  • Measuring forecast accuracy only overall, not by category.
  • Treating scraped prices as selling prices.
  • Not recording whether a metric is with or without VAT.
  • Building cohorts on a base where a small share of sales is visible.
  • Releasing a report without a spec: fields, filters, period and source.
  • Loading data without defining the table grain.
  • Scoring every field instead of those the reports need.
  • Keeping one measure with different definitions in different reports.

Reports to keep

  • Metric dictionary: Definitions, formulas and sources of all metrics.
  • Forecast accuracy report: WAPE and bias by category and period.
  • Promotion post-mortem: Plan, actual, baseline, incremental profit.
  • Data quality report: Empty and wrong attributes, duplicates, mismatched units.
  • Customer base report: Coverage, active, repeat, cohorts, churn.
  • Brand history report: Brand seasons with growth, discounts, margin.
  • Season results report: The season ladder with cuts.
  • Report data readiness: Which standard reports can be built and which fields are missing.

Data to keep

  • Measures dictionary, reference books and the data quality score. ERP and accounting system; PLM and production records
  • Sales, stock and orders as common tables with a known grain. POS: receipts; Warehouse and stock system; ERP and accounting system
  • Warehouse and reports (BI): access and refresh. Plan and budget model

Open interactively →