Product Portfolio Management at HP A Case Study in Information Management ISM 158: Business Information Strategy April 13, 2010 Outline • The benefits and challenges of product variety • Analytics for variety management • Implementation and impact at HP 2 Product variety at HP today Over 2,000 laser printers 3 Over 20,000 enterprise server & storage SKUs Over 8,000,000 possible desktop & notebook PC configurations Why offer product variety? • Expand market reach – offer something for everyone –Many geographies –Many customer types (consumer, small-to-medium business, enterprise) –Many industries (healthcare, technology, energy, government…) • Be a “one stop shop” – offer comprehensive solutions • Increase brand visibility • Win marketshare 4 Challenges of product variety Company Suppliers Customers 5 • Product design costs • Sales & marketing costs costs • Administrative costs • Sales Productivity costs • Inventory-driven costs • • • • Availability / stockouts Delivery time predictability Order cycle time Confusion • Forecast inaccuracies • Inventory-driven • Obsolescence costs Challenges of variety: illustration of inventory driven costs Two similar laptop models: • • • • The two laptop models have independent, identically distributed random demand D1, D2 in each week. Variance of D1, D2 is 2. “Safety stock” inventory of each product is typically k where k is a constant related to the desired service level. Total safety stock: 2k Pool into a single laptop model: • • • 6 Assume no loss in demand (total random demand for single product is D=D1+D2.) Variance of D=D1+D2 is 22. If we apply the same service level objective, then required safety stock for the pooled product is (2)k. By pooling demand from two independent products with equal volumes, the required safety stock and associated inventory-driven costs is reduced by (22)/2 = 29%. Variance of D1+D2 Var(D1+D2) = E[(D1+D2 – E(D1+D2))2] = E[(D1+D2 – 2)2] where = E[D1] =E[D2] = E[((D1 – ) + (D2 – ))2] = E[(D1 – ) 2] + E[(D2 – )2] + 2E[(D1 – )(D2 – )] = Var[D1] + Var[D2] + 2Cov[D1,D2] Since D1,D2 are independent, then: Var(D1+D2) = Var[D1] + Var[D2] = 22 7 The organizational divide Marketing Marketing 8 Supply Chain More platforms More skus More features Better forecasting Precise buffer stocks Less inventory More market share More choices Lower cost Shorter order cycle Reliable deliveries Product Variety Management Lifecycle Pre-launch Variety Management • Before bringing a product to market, estimate its Return On Investment (ROI) • Explicitly consider the costs of variety in this ROI analysis 9 Post-launch Variety Management • After products have been launched, use sales data to maximize value from the existing portfolio Outline • The benefits and challenges of product variety • Analytics for variety management • Implementation and impact at HP 10 Post-launch variety management • Use order history to understand products’ relative importance – Evaluate unimportant products for discontinuance – Improve operational focus on key products. For example: • Divert limited resources toward forecasting & managing key products • Allocate inventory budget toward key products to improve availability • How to evaluate products relative importance from order history? – Rank by revenue – Rank by units shipped – …. Limitations of simple product rankings • Ignores interdependencies among products Order coverage • A customer order is covered by a product portfolio if all of its products are included in the portfolio covered order non-covered order A product portfolio • Order, revenue or margin coverage of a portfolio is the number, revenue or margin of historical orders that can be completely fulfilled from the portfolio 13 Designing a product portfolio to maximize coverage • Problem statement: Given a portfolio size n, find the portfolio of n products that maximizes revenue coverage relative to a given set of recent orders A diversion: a brief introduction to linear programming Decision variables x = x1 x2 ct x Maximize Subject to: Ax b x0 xn Linear objective function c t x c t = (c1, c2, …, cn) is an n-vector of objective coefficients Linear constraints A x b, x 0 A = a11 a12 … a1n a21 a22 … a2n … Solution technique: the Simplex Method (George Dantzig, 1947) am1 am2 … amn b= b1 b2 bm is an mvector of resources is an m x n matrix of constraint coefficients A diversion: integer linear programming Maximize ct x Subject to: Ax b Integer-valued Decision variables x = x1 x2 xn Linear objective function c t x c t = (c1, c2, …, cn) is an n-vector of objective coefficients xi 0,1,2,…., i =1,…,n Linear constraints A x b A = Solution technique: Branch-and-Bound and variations a11 a12 … a1n a21 a22 … a2n … am1 am2 … amn b= b1 b2 bm is an mvector of resources is an m x n matrix of constraint coefficients Designing a product portfolio to maximize coverage • Problem statement: Given a portfolio size n, find the portfolio of n products that maximizes revenue coverage relative to a given set of recent orders • An integer programming formulation: Maximize o Ro yo Objective function Subject to: yo xp for each (o,p) where product p is in order o p xp n xp ,yo 0,1 Decision variables Constraints Notation xp=1 if product p is included yo=1 if order o is covered Ro revenue of order o Revenue Coverage Optimization Tool (RCO) • Rank products according to their importance to revenue coverage • RCO ranking corresponds to efficient frontier of revenue coverage and portfolio size • Use RCO ranking to identify: – Core Portfolio – Extended Portfolio – Possible candidates for discontinuance % of revenue covered 100 80 60 40 20 0 0 18 300 600 900 # of products 1200 Evolution of RCO formulation Integer Program IP(n) IP(n): Find product set of size n that maximizes total revenue of orders covered. Notation: Maximize o Ro yo xp=1 if product p is included Subject to: yo xp if product p is in order o p xp n xp ,yo 0,1 yo=1 if order o is covered Ro revenue of order o Evolution of RCO formulation Integer Program IP(n) Lagrangian Relaxation LR() IP(n): Find product set of size n that maximizes total revenue of orders covered. LR(): Maximize revenue of covered orders minus lambda times portfolio size. Maximize o Ro yo Maximize o Ro yo - (p xp) Subject to: yo xp if product p is in order o p xp n xp ,yo 0,1 Subject to: yo xp if p is in order o 0 x p, y o 1 “Selection problem” Evolution of RCO formulation Integer Program IP(n) Lagrangian Relaxation LR() LR(): Maximize revenue of covered orders minus lambda times portfolio size. Min s-t cut is an optimal solution to selection problem (Balinsky 1970) • Maximize o Ro yo - (p xp) Subject to: yo xp if p is in order o 0 x p, y o 1 “Selection problem” Parametric Bipartite Max Flow Problem orders product s min cut R1 s . . . t . . . Max flow min cut (FordFulkerson) Rn Performance evolution Integer Program IP(n) CPLEX days + memory limitations Prior algorithm for HPLabs Lagrangian bipartite SPMF arc Relaxation parametric balancing LR() max flow CPLEX hours + memory limitations C++ 20 minutes for many values HPLabs SPMF vertex balancing C++ C++ 2 minutes for all 10 seconds for all Computation times on Personal Systems Group’s typical worldwide 3 month order data 22 Comparison to traditional ranking • RCO • Revenue impact • Maximum order revenue • Units shipped • Revenue generated 23 Outline • The benefits and challenges of product variety • Analytics for variety management • Implementation and impact at HP 24 Product discontinuance decisions • Take aim at products in the tail of the ranking • These products don’t generate much revenue of their own, nor do they enable sales of other high-revenue products • This analysis enabled fact-based discussions between marketing and sales organizations • It led to discontinuance of over 3000 products since 2004 % of revenue covered 100 80 60 40 20 0 0 300 600 # of products 25 900 1200 The Recommended Offering program • Define Recommended Offering: the top ranked products covering 80% of revenue • Shift inventory investment to Recommended Offering products • Offer customers quick delivery time on orders that are completely within the Recommended Offering • Significantly improved order cycle time & competitiveness % of revenue covered 100 80 60 40 20 0 0 26 300 600 # of products 900 1200 Summary of business impact • Over $500M in savings and $180M in ongoing annual savings • Significant order fulfillment improvements • Thousands of SKUs eliminated Marketing Analytics Supply Chain Fact-based discussions Data-driven decisions Power of analytics Our customers are the real OR winners! 27 Takeaways • The benefits and challenges of product variety • Perspectives of different organizations within a firm on product variety • Metrics to understand product importance from order history • How effective use of analytics can bridge the organizational divide and bring about operational efficiencies and competitive advantage 28 Thank you