Data Analytics for Tire Finished Goods Inventory & Production Planning

The setup

We addressed these objectives through three stock cases.

• 01Safety stockProtect against order fluctuations in the OEM channel.
• 02Prebuild stockBuild inventory before high-order months for high-volume products.
• 03Cycle stockReduce production setups for strategic-mix products.

Where our work fits

Output: a prioritized list of SKUs that fall into each of the three stock cases, together with their target weekly-cover values.

The list goes to planners for review before it reaches aggregate production planning, which is what makes it decision support rather than a model output.

01Safety stock

Fixed vs dynamic TWCV

Inventory is set according to Target Weekly Cover Value (TWCV), the number of weeks for which inventory should cover forecasted demand.

• Current approach: Keep inventory to cover the 2.5 weeks’ forecast.
  • That is, TWCV = 2.5 for all OEM SKUs.
• Our approach: Customized and dynamic TWCV
  • product-dependent (between 2–5 weeks)
  • time-dependent (modified over months based on forecasts)
• Production decision if Current Weekly Cover Value (CWCV) < TWCV
  • Production may not start immediately due to constraints.

Data: Capacity Utilization: Forecast / Prod. Capacity
Result: Targeted Weekly Cover Value for the t months

initialization: t = 1; weight W such that 0 ≤ W ≤ 1
while length of decision horizon ≥ t do
  if length of decision horizon − t ≥ 3 then
    C_t = W × max{CU_t, CU_{t+1}} + (1 − W) × max{CU_{t+2}, CU_{t+3}}
  else if length of decision horizon − t = 2 then
    C_t = W × max{CU_t, CU_{t+1}} + (1 − W) × CU_{t+2}
  else if length of decision horizon − t = 1 then
    C_t = W × CU_t + (1 − W) × CU_{t+1}
  else
    C_t = CU_t
  end

  if 0 ≤ C_t ≤ 0.25 then
    TWCV_t = 2
  else if 0.25 < C_t ≤ 0.50 then
    TWCV_t = 3
  else if 0.50 < C_t ≤ 0.75 then
    TWCV_t = 4
  else
    TWCV_t = 5
  end

  update t = t + 1
end

Dynamic TWCV for a sample SKU

The proposed target changes over time, while the current approach remains fixed at 2.5 weeks.

Estimated inventory impact across SKUs

Based on November 2021 inventory levels, the counterfactual estimate is ~6% lower total inventory.

02Prebuild stock

Which products should be prebuilt before peak demand?

Seasonal Demand

Three alternatives to meet peak demand

To identify which products should be prebuilt before peak demand, we defined four product dimensions.

Product dimensions

Scores were normalized to [0,1], and weights were determined through an AHP study with Brisa managers.

Prebuild Priority Score

The four normalized dimensions are combined into the weighted Prebuild Priority Score.

PPS_i = 0.161A_i + 0.060S_i + 0.485C_i + 0.294B_i

where A_i, S_i, C_i, and B_i are normalized dimension scores.

Example SKU Profiles

Different dimension scores can lead to similar Prebuild Priority Scores.

Other approaches we tested

• Hierarchical clustering: 4 dimensions, weighted Euclidean distance
• k-means: 3 weighted dimensions, with seasonality as a filter
• Fuzzy c-means: 2 dimensions, with soft cluster membership
• Decision tree: assign newly launched SKUs to existing classes

For segmentation variants, diagnostics, and classifier details, see the thesis or slides.

03Cycle stock

The trade-off

Current: relatively small batch sizes → high setup cost
Proposed: produce selected products in larger batch sizes

The challenge was finding the right balance between availability, production changeovers, and inventory cost.

Dimensions

Dimension scores were normalized to [0,1].

Cycle Stock Priority Score

The four normalized dimensions are combined into the weighted Cycle Stock Priority Score.

CSPS_i = 0.25S_i + 0.125P_i + 0.125G_i + 0.50I_i

where S_i, P_i, G_i, and I_i are the normalized dimension scores.

Discussion & Conclusions

Status in 2022: Being implemented by the Analytics department of the company.

Other questions that our study sheds light on

• For which SKUs should the company purchase more molds and increase production capacity?
• For which SKUs should preorder incentives to dealers be increased or decreased?
• Which SKUs should be promoted in sales channels at certain times of the year, based on their weekly-cover inventory levels?

Other potential product dimensions for segmentation

• Forecast accuracy
• Product substitutability
• Lifecycle stage
• Supply risk