Data preprocessing and analysis

[ARTC-Doris]

Item List Tracking

Data Transform pre-process data for modelling - demand (sales)

• [ ] basic filtering & cleaning
• [ ] SKU master ID mapping (different scenarios: promo/discontinued/case config/Pack Type)
• [ ] resampling to Weekly/Monthly time series per SKU/product family
• [ ] Issue: https://github.com/artc-dsc/TCCC_Data_Processing_Analysis/issues/1 to be completed by June

  Data Transform pre-process data for modelling - external factor to be completed by July

• [ ] weather - aggregate to weekly monthly (required COKE geo-location mapping plus taking 10%min, 90%max, mean, median etc. )
• [ ] promotion - depends on discrete event or periodic series to handle differently (required data ready by June)
• [ ] other macro trends - only analysis and if we have spare time

Analysis target by August

1. conduct thorough analysis of sales patterns to be completed by July
2. segment demand characteristics for differentiated forecasting strategy to be completed by August
   • [ ] SKU demand profiling, e.g. min/max/std/forecast-abilities/seasonal parameters etc. issue https://github.com/artc-dsc/TCCC_Data_Processing_Analysis/issues/2
   • [ ] correlation analysis between SKU-external factor and SKU-SKU
   • [ ] SKU master data segmentation, e.g. ABC, product family
   • [ ] segmentation based on characteristics
   • [ ] other ad-hoc analysis

Notes

Background: North India, 110SKU * 9 facilities, 3 years sales Requirement: rolling forecast to next 12 month, focus on 13-week and 1 month accuracy

[ZengyuCao-ARTC]

TODO List in plan

Processing - demand (sales) completed by June

• [ ] SKU master ID mapping (different scenarios: promo/discontinued/case config/Pack Type)
• [ ] resampling to Daily/Weekly/Monthly time series per SKU/product family

  Processing - external factor completed by July

• [ ] weather - aggregate to weekly monthly (required COKE geo-location mapping plus taking 10%min, 90%max, mean, median etc. )
• [ ] promotion - depends on discrete event or periodic series to handle differently (required data ready by June)
• [ ] other macro trends - only analysis and if we have spare time

  Analysis target by August

• [ ] SKU master data segmentation, e.g. ABC, product family
• [ ] SKU demand profiling, e.g. min/max/std/forecast-abilities/seasonal parameters etc.
• [ ] correlation analysis between SKU-external factor and SKU-SKU
• [ ] segmentation based on characteristics
• [ ] other ad-hoc analysis

Notes

Background: North India, 110SKU * 9 facilities, 3 years sales Requirement: monthly rolling forecast to next 12 month

@ARTC-Doris Please review and comment.

[ARTC-Doris]

@ZengyuCao-ARTC timeline need to be changed a bit to align with milestone review. Please see if any difficulty to meet the proposed timeline as below.

1. Processing - demand (sales) completed by Jun
2. Processing - external factor completed by Jul (provided to received TCCC promotion data by end of June)
3. Analysis completed by Aug (Milestone 1)

Segmentation relies on demand data, is it possible to conduct initial study after step 1, and finetune in Step 3?

[ZengyuCao-ARTC]

@ZengyuCao-ARTC timeline need to be changed a bit to align with milestone review. Please see if any difficulty to meet the proposed timeline as below.

1. Processing - demand (sales) completed by Jun
2. Processing - external factor completed by Jul (provided to received TCCC promotion data by end of June)
3. Analysis completed by Aug (Milestone 1)

Segmentation relies on demand data, is it possible to conduct initial study after step 1, and finetune in Step 3?

1&2 okay, for analysis I will prioritize those required by model dev. Analytic insight will be delivered first, and API coding & integration might need more time. Agree that segmentation can be conducted stage-by-stage. What are step 1 & 3 referring to?

[ARTC-Doris]

• Right, milestone 1 is for analytical insights review, API coding and integration can be completed by Sep and Nov sequentially.
• Please ignore the "step1&3", basically I was trying to say that "segmentation can be conducted stage-by-stage"