Syntha Lab RU

Fashion business IT architecture

A map of systems (POS, WMS, ERP, OMS, CRM/CDP, PLM, MDM, integration bus) that answers three questions: where each piece of data lives, who owns it and what breaks on a peak day. It is used to choose systems, compute total cost of ownership and meet personal data requirements.

Brand, customer, organisation Brand and retail Verified: 2026-11-14

Open interactively →

Why this document

The IT director and the owner read this when they pick a system, change a vendor or prepare for peak load. It shows which system owns which data, so you can see where stock, price or customer is recorded twice and drifts apart. Without a map every integration is built ad hoc, ownership cost is counted from licenses alone, and the weakest link fails on the biggest sales day. The map is also how you learn where customers' personal data sits and who answers for it.

At a glance

What is this document and why is it needed?
A map of systems (POS, WMS, ERP, OMS, CRM/CDP, PLM, MDM, integration bus) that answers three questions: where each piece of data lives, who owns it and what breaks on a peak day. It is used to choose systems, compute total cost of ownership and meet personal data requirements.
Who owns the document and how often is it rebuilt?
Maintained by the CTO together with data owners: the commercial, operations and finance directors. Reviewed once a year and before every major rollout; cost of ownership and risks are shown to the owner at budget time.
What mistakes are the most common?
Choosing by license price -> TCO is 2–3x higher because of customization and integrations -> compute 3-year TCO and exit cost. Several sources of truth for product -> prices and stock differ between channels -> one source system per entity and daily reconciliation. Load tested only on a normal day -> the web shop falls on a sale -> load test with 2–3x headroom a month before the peak and a change freeze before it.

For whom

Brand
A brand needs the center of gravity in product: PLM (tech packs, bills of materials, versions), ERP with production and material buying, style-color-size tracking, WMS or a logistics contract, and a B2B portal for wholesale clients. The key risk is master data drifting apart between PLM, ERP and sales channels.
Retail
A retailer needs the center of gravity in sales and stock: POS and fiscal receipts, store-level accounting, replenishment, OMS for omnichannel, CRM/CDP with a loyalty program, marketplace integration. PLM is usually unnecessary until there is a private label. The key risk is stock mismatch between store, warehouse and web shop.

What it consists of

1. System register

  • For each system: purpose, data owner, vendor, version, contract term, user count, criticality (Tier 1: an outage over an hour stops revenue; Tier 2: tolerates a day; Tier 3: tolerates a week).
  • POS and tills: receipt, discounts, returns, fiscal data; ERP: buying, finance, costing; WMS: bin storage, receiving, picking; OMS: orders of all channels and statuses; CRM/CDP: customer, consents, segments; PLM: product dossier; MDM: single reference for products, stores, counterparties.

2. Data and owner map

  • Entities: product (style, color, size, SKU, barcode), stock, order, receipt, customer, supplier, price. For each: the source system (golden record), consumer systems, exchange frequency, owner.
  • Matrix tracking style-color-size: the unit of record is the SKU (style x color x size); style and color serve planning and reporting, size serves stock and orders. The key must be the same in every system.

3. Integrations and data bus

  • Flow chart: direction, format, frequency (event, hourly batch, nightly), retry queue on failure, monitoring and alerts. Point-to-point is acceptable up to 4–5 systems; beyond that a bus or integration platform is needed.
  • Retailer: Reconciliation: stock in WMS equals the sum of stock in OMS and ERP; till revenue equals ledger revenue; a mismatch above tolerance creates a task for the owner.

4. Total cost of ownership (TCO)

  • One-off: licenses, implementation, data migration, integrations, training. Recurring: subscription or support, infrastructure, administrators, customization (usually a share of one-off cost per year), upgrades.
  • Hidden: downtime during cutover, data loss, single-contractor dependence, exit cost (data export).

5. Resilience and security

  • RTO (time to restore) and RPO (data you can afford to lose) by Tier; backups, restore tests, backup network for tills, offline till mode.
  • Retailer: Peak-day load profile and capacity headroom; role-based access, action log, separate environments, incident response plan.

6. Brand and retailer: what exactly is built

  • Brand: PLM as the source of truth for the product: bill of materials version, fabric, trims, fit comments; the approved version goes from PLM to ERP, not a file. Control: every style in production has exactly one approved specification version.
  • Retailer: Store-level tracking: stock by store and warehouse, receipts, transfers, sales, returns; replenishment runs on store stock by SKU. Control: the sum of store and DC stock equals ledger stock on the date.
  • Brand: B2B portal and EDI with wholesale clients: order, confirmation, ASN, shipment status, reconciliation statements; pre-pack and size proportion must sit in the dictionary, otherwise the order has to be rebuilt by hand.
  • Retailer: Omnichannel: one stock view for the web shop and stores, click-and-collect, ship-from-store, returns in any channel; the stock refresh latency must be shorter than the time in which a second customer can buy the same item.
  • Brand: Own retail and brand D2C: two roles in one company, so wholesale-retail integration goes through an internal settlement: a shipment to an own store is priced at the transfer price, otherwise the management result is distorted.
  • Retailer: Integration with marketplaces and brand suppliers: catalogs, prices, stock, orders; an incoming brand catalog must be mapped to your dictionary (SKU and size mapping), otherwise duplicated cards and mix-ups.
  • Brand: Size proportions and MOQ in ERP: a production order is assembled from channel proportions and a reorder reserve; the system must recompute the order when fabric or factory MOQ changes.
  • Retailer: Loyalty program and customer base: points tracked as a liability, mailing consents, segments; when exporting data to a contractor a processing agreement and field minimization are required.

How to read the result and which decisions follow

  • TCO over three years is RUB 133.1 m (0.82% of annual revenue), the most expensive system is “ERP (accounting, buying, finance)”. Decision: the budget is counted on the line of recurring payments, not the implementation price: over 36 months they form the main part for most systems.
  • Brand: the center of gravity is in PLM and ERP: an error in the product master data (fabric, hardware, version) runs into buying and sales channels. Decision: appoint a master data owner and a change approval procedure before choosing systems.
  • Retailer: the center of gravity is in POS, WMS and OMS: a sales peak of x6 gives 4.1 requests per second in the peak hour and 12.3 with headroom. Decision: load testing before sales, not after the first failure; tills must work offline.
  • Brand: the B2B portal and style-color-size tracking are needed by wholesale clients; without them orders come in spreadsheets and diverge from stock. Decision: count the portal integration with ERP as a mandatory part of the project.
  • Retailer: CRM/CDP and loyalty points are liabilities and personal data; take the choice of system together with the lawyer and with a check of data storage on the territory of Russia.

Control checks

  • Sum check: the total of RUB 133.1 m equals 38.40 one-off plus 94.7 recurring over 36 months (code).
  • Brand: cost of ownership must include customization and downtime, not only licenses; without them the decision understates real costs.
  • Retailer: process client data under the localization and consent requirements; check legal details against the current edition (unverified).
  • Brand: the peak coefficient and calls per order are assumptions; measure them on your own system logs.
  • Retailer: check backup and the recovery plan with a drill shutdown, not only a document.

Where the data comes from

Data sourceFields
Software vendor contracts and invoicesLicenses, subscription, terms, exit penalties, vendor SLA
Incident logs and monitoringNumber and length of outages by system, response time, cause
Peak-day statistics (tills, web shop)Receipts and orders by hour, peak coefficient, response time
Product and counterparty referencesShare of duplicates, empty attributes, code mismatches between systems

How to build it step by step

  1. Step 1. List every system, including spreadsheets that actually hold data; assign an owner and a Tier.
  2. Step 2. Draw the entity map: for product, stock, customer and price define a single source system.
  3. Step 3. Fix the SKU key (style-color-size) and coding rules; check that barcode and SKU code are unique.
  4. Step 4. Draw the flows between systems and mark failure points; set monitoring and a retry queue for each flow.
  5. Step 5. Write selection requirements: functions, integrations, load, rollout time, local support availability, exit cost.
  6. Step 6. Compare 2–3 options by 3-year TCO and by risk; do not choose by license price.
  7. Retailer: Step 7. Compute peak-day load and the required headroom; run a load test before the season.
  8. Step 8. Close personal data requirements: data list, purposes, consents, storage locations, access rights, retention.
  9. Step 9. Describe RTO/RPO, the recovery plan and till fallback mode; rehearse twice a year.
  10. Step 10. Approve a 12–18 month roadmap with budget and metrics (mismatch rate, downtime, share of automated flows).
  11. Brand: Step 11. Brand: link PLM, ERP and the B2B portal with one style-color-size dictionary and check that one approved specification version reaches production.
  12. Retailer: Step 12. Retailer: test the end-to-end path of a customer order (web shop, OMS, WMS or store, delivery, return) and measure the stock refresh time at each point.

Formulas

MeasureFormulasNote
TCO over a periodTCO = sum of one-off + sum of monthly recurring x months + customization cost + downtime costDowntime is revenue per hour x outage hours x (1 - share recovered by offline mode).
Peak loadOrders per peak hour = daily orders x peak coefficient x peak share / peak hours; requests per second = orders per hour x requests per order / 3600Design with 2–3x headroom: campaigns and retry storms push load above plan.
Stock mismatch rateMismatch = sum of |stock in system A - stock in system B| / sum of stockComputed by SKU and warehouse daily; a rising rate signals a broken flow.
Retailer: Cost per storeIT per store per month = TCO / months / number of storesCompare to store revenue rather than total revenue: this shows the load on small stores.

Formulas in notation

Total cost of ownership
TCO = Σ one-off + Σ monthly × M + customisation + downtime
M is the number of months in the horizon (36); customization and downtime are added apart from licenses.
Orders per peak hour
O_h = O_d × k × s / h
O_d is orders on a normal day, k the peak coefficient, s the share of the day in peak hours, h peak duration in hours.
Requests per second
RPS = O_h × q / 3600
q is the number of calls to systems per order.
Design capacity
RPS_design = RPS × H
H is headroom for the unexpected (3 in the example).
Share of revenue
s = (TCO / years) / Revenue × 100
Shows the IT load on the business; compare with your own history, not with other norms.

Worked example

Exercise: 3-year TCO for a 60-store chain, RUB m

SystemOne-offPer monthOver 36 months
POS (tills, 120 tills)1.800.3012.6
WMS (warehouse)6.000.4522.2
ERP (accounting, buying, finance)14.000.7039.2
OMS and web shop4.500.2513.5
CRM / CDP3.000.2813.1
PLM (for private label)2.400.157.8
MDM (master data)2.000.126.3
Integration bus3.500.2010.7
Security and backup1.200.187.7
Total38.402.63133.1

Notional: revenue RUB 5,400m a year. Three-year TCO is RUB 133.1m, on average RUB 44.4m a year, i.e. 0.82% of revenue and RUB 62k per store per month. Peak: 700 orders a day normally, x6 on a sale = 4,200; the busiest 4 hours carry 368 orders an hour, or 4.1 requests per second at 40 requests per order; with x3 headroom the system must sustain 12.3 requests per second. All numbers are notional.

Figures

TCO over 36 months by system
Bar chart: POS (tills, 120 tills) 1.8, WMS (warehouse) 6, ERP (accounting, buying, finance) 14, OMS and web shop 4.5, CRM / CDP 3, PLM (for private label) 2.4010203040POS (tills, 120 tills)POS (tills, 120tills)One-off: 1.8Recurring over 36 months: 10.812.6WMS (warehouse)WMS (warehouse)One-off: 6Recurring over 36 months: 16.222.2ERP (accounting, buying, finance)ERP (accounting,buying, finance)One-off: 14Recurring over 36 months: 25.239.2OMS and web shopOMS and web shopOne-off: 4.5Recurring over 36 months: 913.5CRM / CDPCRM / CDPOne-off: 3Recurring over 36 months: 10.113.1PLM (for private label)PLM (for privatelabel)One-off: 2.4Recurring over 36 months: 5.47.8MDM (master data)MDM (master data)One-off: 2Recurring over 36 months: 4.326.32Integration busIntegration busOne-off: 3.5Recurring over 36 months: 7.210.7Security and backupSecurity and backupOne-off: 1.2Recurring over 36 months: 6.487.68RUB m
  • One-off
  • Recurring over 36 months
Data as a table
MeasureOne-offRecurring over 36 months
POS (tills, 120 tills)1.810.8
WMS (warehouse)616.2
ERP (accounting, buying, finance)1425.2
OMS and web shop4.59
CRM / CDP310.1
PLM (for private label)2.45.4
MDM (master data)24.32
Integration bus3.57.2
Security and backup1.26.48

TCO over 36 months by system. One-off implementation cost and recurring payments over three years in RUB m: ERP and WMS give more than half.

Peak-day load on the systems
Line chart: Load: 0.7 → 5.4; Design capacity with headroom: 2 → 16.30510152012345678our peak x6Load, 1: 0.7Load, 2: 1.4Load, 3: 2Load, 4: 2.7Load, 5: 3.4Load, 6: 4.1Load, 7: 4.8Load, 8: 5.4Design capacity with headroom, 1: 2Design capacity with headroom, 2: 4.1Design capacity with headroom, 3: 6.1Design capacity with headroom, 4: 8.2Design capacity with headroom, 5: 10.2Design capacity with headroom, 6: 12.3Design capacity with headroom, 7: 14.3Design capacity with headroom, 8: 16.3requests/s
  • Load
  • Design capacity with headroom
Data as a table
MeasureLoadDesign capacity with headroom
10.72
21.44.1
326.1
42.78.2
53.410.2
64.112.3
74.814.3
85.416.3

Peak-day load on the systems. Requests per second in the peak hour as the peak coefficient grows, and the design capacity with a ×3 headroom. Horizontal axis: peak coefficient to a normal day.

Data map: who owns what in omnichannel. Master data and the integration bus connect PLM, ERP, warehouse, web shop, tills and the client base.

  1. PLM: product, bill of materials → MDM: product, price, SKU
  2. MDM: product, price, SKU → ERP: buying, finance; OMS and web shop: orders
  3. ERP: buying, finance → WMS: warehouse, stock
  4. WMS: warehouse, stock → OMS and web shop: orders
  5. OMS and web shop: orders → POS: receipt, return; CRM/CDP: client, points
  6. POS: receipt, return → CRM/CDP: client, points
  7. CRM/CDP: client, points

Specifics by category

Apparel
The inventory key is SKU = style × color × size: 400 styles × 4 colors × 7 sizes give 11,200 SKUs a season (an assumption). This requires bulk card upload, one size dictionary in MDM, a barcode at SKU level and archiving of cards by season. The typical defect is size mix-up in the warehouse. Measure the stock mismatch rate by SKU, not by style; a 0.5-1% tolerance is a working hint, tune it to your count accuracy.
Footwear
The unit is a pair, but there are several size systems: EU (2/3 cm step, Paris point), UK/US and Mondopoint (foot length in mm under ISO 9407). Keep foot length in mm in MDM and derive EU/UK/US through a conversion table, otherwise the web shop and the warehouse show different numbers. A 36-41 run with half sizes gives 11 sizes per style-color, so SKUs grow twice as fast as in apparel. A shoebox is a separate WMS unit with its own barcode.
Accessories
There is no size; SKU = style × color or material; many SKUs have 1-3 pcs in stock at a high unit price. Material and hardware attributes are needed, plus a serial number for expensive bags. Items made of precious metals need a separate tracking and marking circuit through the state precious-metals information system (GIIS DMDK) with fineness and weight fields; costume jewelry without precious metals is usually outside this circuit (unverified, confirm with a lawyer). Count stock more often: losing one unit costs more.
Kids
Size follows height: a 6 cm run (56, 62, ... 176); age is only a storefront hint, not the key. MDM needs an age-group flag under TR CU 007/2011 (under 1 year: height up to 74 cm; children up to 14; teens 14-18): it switches conformity documents and material requirements. The 10% VAT rate is stored on the card by the OKPD2 code from the decree no. 908 list, not by a “kids” flag. Children’s data: guardian consent, minimum fields, an age group instead of a birth date.
Men
The basics range lives for years: the SKU repeats from season to season, so a stable code and auto-replenishment matter more than card launch speed. The size grid is shorter and steadier (for example S-XXL or 46-60), so WMS and OMS flows are simpler. The repeat-purchase share is higher, so CRM with the customer’s size has a bigger effect than trend recommendations. Check that a change of pattern does not create a new SKU where the customer expects the same one.
Women
The range is wider and changes faster: more styles, colors, capsules and drops; a fashion card lives 6-10 weeks (an assumption) against several years for basics. You need a fast launch (PLM → ERP → web shop in days), several size systems (letters, RU, EU, US) in one dictionary and vanity-sizing awareness: the same number gives a different measure across brands. The online and fit-return shares are higher, so OMS must receive the return reason early. SKU count for the same revenue may be 1.5-2 times higher than in menswear (an assumption).

Common mistakes

  • Choosing by license price -> TCO is 2–3x higher because of customization and integrations -> compute 3-year TCO and exit cost.
  • Several sources of truth for product -> prices and stock differ between channels -> one source system per entity and daily reconciliation.
  • Load tested only on a normal day -> the web shop falls on a sale -> load test with 2–3x headroom a month before the peak and a change freeze before it.
  • No till offline mode and no restore test -> a network outage stops sales and the backup cannot be restored -> offline mode, drills and measured RTO/RPO.
  • Customer data is exported to contractors without a contract and consents -> fines and leak risk -> a processing register, processing agreement, field minimization.

Russian specifics

  • Localization: recording, systematization, accumulation, storage, updating and retrieval of personal data of Russian citizens on collection are performed using databases located in Russia (Art. 18 part 5 of 152-FZ); separate requirements apply to cross-border transfer. The exact wording and exceptions are not checked against the law text (unverified): check with a lawyer. The operator must have consents, define purposes and protection measures (152-FZ). This covers CRM, CDP, loyalty programs and the web shop. Rule: Law 152-FZ of 27.07.2006: Federal Law of 27.07.2006 No. 152-FZ on personal data (rule check not completed).
  • Retailer: Tills must follow the cash register rules (54-FZ): data is sent to a fiscal data operator; when connectivity is lost, receipts must be stored and sent after restoration. Rule: Law 54-FZ of 22.05.2003: Federal Law of 22.05.2003 No. 54-FZ on the use of cash registers.
  • Retention periods and protection requirements depend on the data class and business size; confirm with the data protection officer and a lawyer. Domestic-software requirements in procurement were not verified.
  • Brand: when producing or importing light-industry items the ledger needs marking codes for each unit, tied to the SKU; integration with the marking-system operator is part of the architecture (format and dates unverified, check against the current rules).
  • Retailer: till software and stock must accept marking codes at sale and return; a scanning error blocks the sale, so test on a pilot till before the season (the withdrawal-from-circulation procedure is unverified).

Open interactively →