[DSC Europe 22] Using AI to solve complex business problems: optimizing markdowns amidst record inventory surplus - Pietro Quagli
Download as PPTX, PDF0 likes7 views
The document discusses the challenges of optimizing markdowns in retail, emphasizing the importance of effective markdown strategies to manage excess inventory, maximize sales, and respect customer expectations while adhering to business rules. It introduces a prescriptive approach utilizing AI and historical data to determine optimal markdown levels and forecasts, resulting in enhanced sales and margins. The findings from a summer 2022 markdown campaign for an Italian fashion company demonstrate the significant impact of well-planned markdowns on revenue growth.
[DSC Europe 22] Using AI to solve complex business problems: optimizing markdowns amidst record inventory surplus - Pietro Quagli
- 1. CONFIDENTIAL AND PROPRIETARY. evo trademarks registered in the UK, the US and the EU. Patent-pending, all rights reserved. Any use without specific permission of an authorised Evo legal rightsholder is strictly prohibited. November 14 2022 Using AI to solve complex business problems: optimizing markdowns amidst record inventory surplus
- 3. | 3 ― Context and problem ― Our solution ― Conclusions Agenda
- 4. | 4 Did you know? Matching supply and demand is hard Excess inventory Unsold product $1.43 inventory per $1 of sales $2.04 trillion inventory in the US, ~11% GDP Availability gaps 75% of managers see availability as strategic Discount pressure 50.8m tonnes electrical & 1.3BN tonnes food wasted every year SOURCES: Wall Street Journal, Financial Times, Forbes
- 5. | 5 Realise the volume and sales $ opportunity Grow margin Minimise residual stock Give away sales and margin without increasing volume Leave high residuals, waste and disposal costs Confuse customers Erode brand and image Poor Markdowns Good Markdowns Balancing priorities to create value The challenge: how to set good markdowns that fulfill customer expectations and respect business rules The markdown challenge
- 6. | 6 The challenge: how to set good markdowns that fulfill customer expectations and respect business rules The markdown challenge Respecting rules and conventions Customer Expectations Business Rules Clear and logical prices Well-defined markdown periods Company markdown protocols i.e. price points, discount levels, etc. Acceptable frequency and number of price changes Residual stock levels Brand concerns
- 7. | 8 Why prescriptive approaches for high complexity processes Low complexity High complexity Standard markdown models: Rely on large numbers of rules Suffer from forecast errors Need long lead times to achieve optimal results Require constant manual input Struggle to achieve impact Prescriptive, autonomous systems: Make decisions based on profit outcomes Learn from feedback for faster results Require less manual work Allow for simulations and new questions Create largest possible value from markdowns
- 8. | 10 Calculating markdowns – algorithm Using historical sales data and recent full-price sales to predict future sales at full price Using historical discount to infer price demand elasticity to simulate the effect of different discount intensities Describe the discount level decision as a linear optimization problem maximizing parameteric objective function Select admissable discount level for each item according to custom business rules defined via Excel interface Historical data Optimization Forecast Elasticity New discounted pricelist Business rules
- 9. | 11 Impact summer season Results of the markdown campaign (June 20 to August 22): +15.3% sales, +10.2% revenue,+ 2.9% margin In the second phase of sales (from wk31) sales accelerate (peak +30% in wk32), with a lower margin Test: stores with Evo recommended price-list Control: stores with a traditional price-list -15% -10% -5% 0% 5% 10% 15% 20% 25% 30% 35% 26 27 28 29 30 31 32 33 % change: test vscontrol group Margin Revenues Units sold Wave 1 Wave 2 Results measured for the summer 2022 discount season of an Italian fashion company
- 10. | 12 Example: leveraging elasticity to increase revenues . Relatively high performer. Inventory Rotation: 7 weeks Sensitive to price. Price elasticity: -2.80 Test 1.499 Control 2.015 +26% Revenues, k EUR 30% 50% 65% 14% 25% Test 20% Control 30% +10pp Level of discount Women’s shirt – light blue A lower price, even for a high-rotation product, can achieve greater revenues & sales Results measured for the summer 2022 discount season of an Italian fashion company
- 11. | 13 Forecast challenge Forecast needed to decide the discounts The discount affects the forecast
- 12. | 14 Removing markdown effects Fetch historical demand price and the percentage markdown from past data: q(t), p,m(t) Use the elasticity to rescale demand: q'(t)=q(t)*(1+ε*(1-m(t)) Predict future demand using rescaled historical data: q (t+Δt)=F(q'(t))
- 13. | 15 Discount level selection Define possible markdown level for future: mi (t +Δt) Rescale demand forecast with elasticity: qi(t +Δt)=q(t)*(1+ε*mi (t +Δt) Select markdown level such that the revenues are maximal: argmax(p(1- mi (t +Δt))* qi (t +Δt))
- 14. | 16 Refining the optimization problem 𝑎𝑔𝑚𝑎𝑥 𝑈(𝑚𝑖) − 𝑃𝑗(𝑚𝑖) The optimization problem can be generalized by: • including different objective functions • adding constraints to implement the business rules
- 15. | 18 Conclusions The automated combination of elasticity rescaling and forecast enables adaptable and responsive markdowns The prescriptive approach unlocks the potential of complex markdown strategy Markdowns become an opportunity rather than a solution to a problem Modularity grants clarity and consistency of recommendations
- 16. Thanks for you attention linkedin.com/in/pietro-quaglio-434456b7
- 17. | 20 Demand elasticity The elasticity of demand relates to price variation and quantity sold: Its estimator can be empirically derived by considering historical price changes:
- 18. | 21 Refining the optimization problem 𝑎𝑔𝑚𝑎𝑥 𝑈(𝑚𝑖) − 𝑃𝑗(𝑚𝑖) The optimization problem can be generalized by: • including different objective functions • adding constraints to implement the business rules Hard constraints: • Unique discount level for each item • Increasing markdown during the season • Min/max discount by week/category Soft constraints: • Number of inventory pieces per discount level • Number of sold pieces per discount level • Number of changes in the whole season
- 19. | 22 In a perfect world, no retailer would have to apply markdowns. The forecast would be 100% accurate, empowering you to supply each channel with the exact products that could be sold. Unfortunately, that’s still science fiction. Markdowns are therefore an essential retail strategy. To fully optimize revenues and margins while minimizing unsold inventory, markdowns must be carefully planned using data to inform decisions. Markdown management: why it is a priority Markdown management: how do you find the right markdown profile to maximise sales and margin and minimise residual inventory, while respecting customer and business rules?
- 20. | 23 Markdown management: why it is a priority Why are markdowns essential in retail? - To dispose of seasonal items - To adjust assortments - To increase sell-through - To maximise revenue and profit - To appeal to the cluster of customers that only buy in sales periods 0.0% 2.0% 4.0% 6.0% 8.0% 10.0% 12.0% Fixed Cost Volume Variable Cost Price Retail Levers: 1% improvement creates x% Operating Profit improvement* No lever a retailer can pull is more profitable than improving pricing. In 2018 alone, poorly executed markdowns cost retailers in the U.S. $300B—even before Covid-19 disruptions further diminished the impact of markdowns.