Menu

Retail Forecasting GPT — Data Upload Specifications

Below is a clean, complete, standards-aligned specification for the data requirements that users must follow when uploading files to Retail Forecasting GPT.
Everything is grounded in the rules from the Master Specification (Document 1–3) and summarized into an upload-ready guide.

✅ Retail Forecasting GPT — Data Upload Specifications (What Data Is Required & Accepted)

Version 1.0 — For internal & user-facing documentation

This specification defines the minimum, recommended, and optional data structures that users should provide for Retail Forecasting GPT to perform forecasting, diagnostics, scenario modeling, buying, and allocation recommendations.

1. ✔️ Accepted File Formats

Retail Forecasting GPT accepts:

  • CSV (.csv)

  • Excel (.xlsx, .xls)

  • Google Sheets (via pasted table)

  • Inline pasted tables

2. ✔️ Required Minimum Data Fields

(These are the minimum viable data fields the GPT needs to produce a forecast.)

Every dataset must include:

Required Fields

Column Description
Date Daily, weekly, or monthly timestamps. Must be consistent.
Units Sold Actual units sold per time period.
SKU / Product (optional if total business) Needed for SKU/category-level forecasting.

Minimum history length requirements

  • ≥ 12 months: usable trend + some seasonality

  • ≥ 24 months: reliable seasonality

  • < 6 months: forecast will be directional only and flagged as very low confidence

3. ✔️ Strongly Recommended Fields (Improves accuracy & diagnostics)

Product / Category Structure

Column Purpose
Category / Subcategory Enables category-level forecasting + mix analysis.
Channel (Store, Ecommerce, BOPIS) Allows omnichannel forecasts + channel-shift analysis.

Price & Promotion

Column Purpose
Price / ASP Enables elasticity & promo impact detection.
Promo Flag Identifies uplift, pull-forward, baseline correction.
Promo Type (e.g., % off, BOGO, sitewide promo).

Inventory & Supply Chain

Column Purpose
On-Hand Inventory Identify stockouts, overstock risk, WOS, inventory constraints.
On-Order Determines whether buys cover forecasted demand.
In-Transit Helps identify supply delays.
Lead Time (if provided) Supports buy plan calculations.

Store / Location

Column Purpose
Store ID Allocation logic.
Region Seasonal / climate adjustments; regional demand differences.

4. ✔️ Optional (but highly valuable) Data Fields

Increases diagnostic accuracy, forecasting confidence, and retail insight quality.

Traffic & Conversion (Store/Ecom)

Column Purpose
Traffic / Sessions Diagnoses whether demand is traffic-driven or conversion-driven.
Conversion Rate Crucial for promo interpretation; channel behavior analysis.

Returns

Column Purpose
Units Returned Allows net demand forecasting and basket/fit diagnostics.

Marketing

Column Purpose
Marketing Spend Enables efficiency analysis; explains volume spikes.
Campaign Type Connects demand to controllable marketing drivers.

Fulfillment / Operations

Column Purpose
Ship Time / SLA Ecommerce performance interpretation.

5. ✔️ Required Column Formats & Standards

5.1 Date Format

Acceptable formats:

  • YYYY-MM-DD

  • MM/DD/YYYY

  • Week numbers (with stated week-ending convention)

  • Month-Year (e.g., Jan-2024)

Must be consistent throughout.

5.2 Frequency

Document 1 requires Retail Forecasting GPT to detect:

  • Daily

  • Weekly

  • Monthly

GPT will automatically summarize:

  • Missing periods

  • Outliers

  • Trend direction

  • Seasonality patterns

6. ✔️ Sample Ideal Dataset Structure

(Can be used in user documentation or templates)

Date SKU Category Channel Units Sold Price Promo Flag On Hand On Order Region
2023-01-01 A123 Tops Ecommerce 82 29.99 0 120 50 East
2023-01-08 A123 Tops Ecommerce 96 24.99 1 75 50 East
… … … … … … … … … …

7. ✔️ Optional Data Templates to Provide (for your onboarding materials)

Ask users to choose the data type that matches their business:

Template A — Total Business Forecast

Fields: Date, Units Sold
Used for: Finance, executive planning.

Template B — SKU-Level Forecast

Fields: Date, SKU, Units Sold, On-Hand, On-Order, Promo Flag
Used for: Buyers, planners, allocation.

Template C — Category + Channel Forecasting

Fields: Date, Category, Channel, Units Sold, Price, Promo Flag
Used for: Merchandising, omnichannel.

Template D — Full Diagnostic Dataset

Includes all recommended fields.

8. ✔️ Data Quality Rules (GPT Requires These for Accurate Interpretation)

8.1 No missing dates within the frequency

If missing, GPT:

  • flags uncertainty

  • interpolates only when safe

8.2 Outliers must not be removed by the user

GPT detects:

  • promo spikes

  • stockouts

  • one-time events

  • supply disruptions

8.3 Consistent units

Units must mean the same thing across all rows:

  • Units shipped?

  • Units sold?

  • Gross demand?

If unclear → GPT will ask.

8.4 No aggregated totals mixed with disaggregated rows

(e.g., don’t mix daily + weekly + monthly in same column)

9. ✔️ What Users Must Tell GPT If Not in Data

To avoid incorrect assumptions, users must supply:

  • Lead time

  • Planned promotions

  • Service level target (90%, 95%, etc.)

  • OTB constraints

  • Whether minimizing stockouts or overstock is more important

  • Category role (traffic driver, margin driver, seasonal hero, basic, etc.)

These influence forecasting, buy plans, allocation, and inventory recommendations.

10. ✔️ Final Summary (User-Facing)

To get the best forecasts and inventory recommendations from Retail Forecasting GPT, upload a file with:

Required Minimum

  • Date

  • Units Sold

  • SKU (if SKU-level)

Strongly Recommended

  • Price

  • Promo indicator

  • Inventory (On-hand, On-order)

  • Channel

  • Category

Optional but Highly Valuable

  • Traffic & conversion

  • Returns

  • Marketing spend

The cleaner the dataset, the higher the forecast confidence and the more precise the buying & allocation recommendations.

✅ 1. One-Page Data Upload Specification (PDF-ready text)

Retail Forecasting GPT — Data Requirements v1.0

Purpose

To ensure accurate forecasting, inventory planning, and allocation decisioning, Retail Forecasting GPT requires structured retail data inputs. This one-page specification defines the minimum, recommended, and optional fields that users should include when uploading datasets.

Minimum Required Fields (for forecasting to run)

The system can generate a forecast as long as the following fields are present:

Field Description
Date Daily, weekly, or monthly timestamps (consistent frequency).
Units Sold Units sold per period.
SKU / Product (optional for total-business) Required for SKU or category-level forecasting.

History requirements

  • ≥ 12 months: usable trend & basic seasonality

  • ≥ 24 months: reliable seasonal detection

  • < 6 months: directional forecasts only; low confidence

Strongly Recommended Fields (improves accuracy & interpretation)

Product / Structure

  • Category

  • Channel (Store / Ecommerce / BOPIS / Marketplace)

Pricing & Promotion

  • Price / ASP

  • Promo flag (Y/N)

  • Promo type (% off, BOGO, sitewide, etc.)

Inventory & Supply Chain

  • On-Hand Inventory

  • On-Order

  • In-Transit

  • Lead Time (if not in data, user must supply manually)

Store / Location

  • Store ID

  • Region

Optional but Valuable Fields (enhanced retail diagnostics)

Traffic & Conversion

  • Traffic (store) or sessions (ecommerce)

  • Conversion rate

Returns

  • Units returned

Marketing

  • Marketing spend

  • Campaign type

Operations

  • SLA / ship time

  • Warehouse / FC identifier

Formatting Rules

  • Consistent date format

  • Consistent frequency (do not mix daily + weekly)

  • No subtotal/summary rows

  • Units Sold must always represent sell-through, not inventory movement

  • Missing periods should not be filled manually (GPT will detect and handle)

User Inputs Required if Not in File

  • Lead time

  • Planned promotions

  • Service level target (e.g., 95%)

  • OTB constraints

  • Category roles

  • Priority between stockout avoidance vs. overstock mitigation

Example Ideal Data Structure

Date SKU Category Channel Units Sold Price Promo Flag On Hand On Order Region
2023-01-01 A123 Tops Ecommerce 82 29.99 0 120 50 East

 

✅ 2. CSV / Excel Templates (Copy-Paste Ready)

Below are the three templates you asked for:
Simple SKU-Level, Category/Channel, and Full Diagnostic.

Template A — SKU-Level Forecasting (Minimal)

Date,SKU,Units Sold
2023-01-01,A123,82
2023-01-08,A123,96

Template B — Category + Channel Forecasting (Recommended)

Date,Category,Channel,Units Sold,Price,Promo Flag
2023-01-01,Tops,Ecommerce,82,29.99,0
2023-01-08,Tops,Ecommerce,96,24.99,1

Template C — Full Diagnostic Dataset (Ideal)

Date,SKU,Product Name,Category,Subcategory,Channel,Region,Units Sold,Price,Promo Flag,Promo Type,On Hand,On Order,In Transit,Traffic,Sessions,Conversion Rate,Units Returned,Marketing Spend,Campaign Type
2023-01-01,A123,Classic Tee,Tops,Basics,Ecommerce,East,82,29.99,0,,120,50,0,,4500,0.018,4,1200,Email
2023-01-08,A123,Classic Tee,Tops,Basics,Ecommerce,East,96,24.99,1,30OFF,75,50,0,,5100,0.021,6,1800,Promo

✅ 3. Data Validation Checklist (Copy-Paste For Onboarding / UI)

✓ Data Frequency

  • All rows use the same frequency (daily / weekly / monthly)

  • No mixed granularities

  • No missing date periods, or acceptable for GPT to infer

✓ Required Fields Present

  • Date

  • Units Sold

  • SKU (if SKU-level or category-level analysis desired)

✓ Recommended Fields Present

  • Category

  • Channel

  • Price

  • Promo flag

✓ Inventory Fields

  • On-hand inventory

  • On-order inventory

  • Lead time provided (in file or text)

✓ Data Integrity

  • No totals/summary rows

  • No negative units (unless returns dataset)

  • No duplicate rows for same date/SKU unless intentional

  • No aggregated weekly + daily data mixed

✓ Optional High-Value Enhancers

  • Traffic / sessions

  • Conversion

  • Returns

  • Marketing spend / campaign type

✅ 4. Data Dictionary (Friendly + Developer Version)

4.1 Business-Friendly Version

Field Definition Why It Matters
Date Period of sale Establishes trend & seasonality
SKU Unique product identifier SKU forecasting & allocation
Category Product grouping Category-level seasonality & trends
Channel Where sale occurred Channel-shift modeling
Units Sold Units sold in that period Core forecasting driver
Price Actual selling price Elasticity & promo analysis
Promo Flag Indicates promotion Corrects baseline demand
Promo Type Discount type Determines promo uplift ranges
On Hand Inventory currently available Stockout diagnostics
On Order Incoming supply Buy plan calculations
In Transit Inventory on the way WOS & availability
Traffic / Sessions Shoppers or site visits Diagnoses traffic vs conversion problems
Conversion Sales ÷ traffic Identifies conversion issues
Units Returned Units returned Net demand & fit/quality diagnostics
Marketing Spend Paid media investment Interprets spikes & channel behavior
Region Geographic grouping Regional seasonality & weather effects

4.2 Developer-Friendly Version (Field Types + Constraints)

Field Type Allowed Values Notes
Date date ISO8601 or consistent string No blank rows
SKU string any Required for SKU/category models
Category string any Strongly recommended
Channel string store/ecom/BOPIS/marketplace Case-insensitive
Units Sold integer ≥ 0 Core forecast driver
Price float ≥ 0 Optional but ideal
Promo Flag boolean/int 0/1, TRUE/FALSE Detects uplift
Promo Type string %off, BOGO, etc Optional
On Hand integer ≥ 0 Required for inventory planning
On Order integer ≥ 0 Optional
In Transit integer ≥ 0 Optional
Traffic / Sessions integer ≥ 0 Optional
Conversion Rate float 0–1 Optional
Units Returned integer ≥ 0 Optional
Marketing Spend float ≥ 0 Optional
Region string any Optional
Views: 2