Product portfolios tend to grow. Companies add new products and variants all the time to meet specific customer needs, pursue new market opportunities, or exploit technological advances. They can also be reluctant to kill off older items, for fear of upsetting long-standing customers.
Finding the sweet spot in a portfolio is an ongoing challenge. Offer too little choice and customers will choose to play elsewhere, costing the company revenue. Offer too much and the cost and operational complexity of managing the assortment becomes a drag on profitability. Worse, companies can fall into both traps at once: offering too many of the wrong products, resulting in a toxic combination of high costs and lost sales.
Get the balance right, however, and the rewards can be significant. Top performers in product portfolio management enjoy three sources of competitive advantage. First, they improve overall cost efficiency by limiting the costs of maintaining unprofitable and obsolete products. Second, they focus more attention and resources on innovation, developing products and services that propel growth and profitability. Finally, they can improve resilience by simplifying component inventories and reducing supplier-related risks. Now, generative AI gives the capability to get it right without needing to dedicate huge amounts of resources.
Product portfolio optimization is difficult because it requires companies to balance so many variables. They must understand how different products and combinations of products meet different customer needs, how those products compare to competitive offerings, and how much each product and variant costs to manufacture, deliver, and support. They also need to integrate the varied, and sometimes conflicting, viewpoints of different stakeholders across the business. Analyzing these variables is further complicated by recent trends, including increased product customization, the interplay of hardware and software in modern product designs, and the need to manage complex supply chains that are subject to geopolitical disruptions.
Then there are network effects. Multiple products often reuse the same underlying components, manufacturing assets, or code. Removing a single product from the portfolio typically does little to reduce the underlying complexity of the company’s operations. Instead, portfolio optimization requires removing a carefully selected cluster of products so that the cost savings to the company’s operations outweigh the potential loss of revenue (Exhibit 1). Identifying the right clusters is a daunting task: in a portfolio of several thousand reasonably complex SKUs, the number of potential options quickly exceeds the number of particles in the universe.
Organizations now have new ways to combine, analyze, and interpret complex, diverse, and potentially ambiguous data. Recent breakthroughs in generative AI (gen AI) technologies have the potential to transform the ways they optimize their portfolios. Gen AI tools are designed to excel at exactly the types of tasks that portfolio optimization requires, such as combining large data sets, performing complex analyses, and responding to user requests (Exhibit 2). These tasks include the following:
Our interactive graphic (Exhibit 3) shows the results of a related sales analysis conducted by an industrial equipment manufacturer as part of a broader portfolio optimization exercise. The manufacturer used a combination of traditional and gen AI analytics tools. This type of analysis is designed to help companies identify products for pruning, streamlining, or repricing. This analysis addresses a concern that many companies have: that simplifying the portfolio would result in unwanted lost sales.
Each circle represents a product in the portfolio. Revenue and gross margin are represented by the size and color of the circle, respectively. Related sales are visualized by the proximity of the circles and the connecting lines. The closer the circles are to each other, the higher the percentage overlap of the customer orders in which they occur. The connecting lines show where the order overlap exceeds 50 percent.
Scroll through the graphic to see more details about the characteristics of different clusters within the portfolio and the associated optimization opportunities.
Gen AI–based product portfolio optimization is at the cutting edge of technological innovation. By the end of 2023, when we first performed the analysis above, we estimated that gen AI tools would automate about 10 percent of the full set of analysis required to optimize a company’s portfolio (Exhibit 4). By summer 2024, the availability of more advanced large language models (LLMs) that support the gen AI revolution had increased the share of automation to 30 percent. Forthcoming gen AI models will be able to evaluate portfolios in the external context, using search engines to retrieve information on evolving customer preferences and competitor announcements. We now expect the fraction of a portfolio analysis that can be automated using these tools to exceed 50 percent in 2025. Even at current levels of automation, gen AI is dramatically reducing the overall time required to complete portfolio analysis exercises. One company planned an eight-week project to analyze its portfolio of about 20,000 SKUs. With the help of gen AI tools, it completed the job in half the time.
This is important because automation reduces the time required to conduct a portfolio analysis by more than 90 percent. And the faster that companies can complete their analyses, the broader and deeper their portfolio reviews can become. Without gen AI, even a basic portfolio analysis takes months of work, and most companies focus their efforts at either end of their SKU distribution. Product management teams spend most of their time and resources optimizing the high-runners—the best-performing 20 percent of products that generate 80 percent of their gross margin. At the other end of the distribution, they may periodically prune the worst performers from the long tail of low-margin products but without properly considering the network effects mentioned earlier, resulting in suboptimal decisions.
As gen AI technologies make it possible to automate more tasks, companies will be able to perform detailed analyses across a much larger portion of their portfolios, uncovering the hidden sources of cost and complexity that come from underperforming products in the complex middle and long tail of their portfolios.
This could be a game changer. Our analysis of complex portfolios, such as those of machinery manufacturers, consumer durable producers, and automotive suppliers, suggests that 50 to 70 percent of the total optimization impact can come from products in these parts of the portfolio, with benefits accruing through a combination of replacement, repricing, and product phase-out. In profit-and-loss terms, this translates into a six- to ten-percentage-point margin improvement, or a 2 to 5 percent increase in revenue for companies that optimize their entire product portfolio.
Applying gen AI at scale to support high-impact, commercially sensitive decisions requires a robust, enterprise-grade environment. Organizations will naturally want to use the latest AI models, but they need to surround those models with an appropriate infrastructure that ensures data security and confidentiality while allowing for the upload of diverse structured and unstructured data. Building code execution into the gen AI environment allows for rapid validation, result computation, and visualization (Exhibit 5).
An enabling environment for gen AI starts with granular product data. Data storage costs have plummeted in recent years, and gen AI is dramatically reducing the cost of analyzing large volumes of complex data. Together, those trends are changing the cost–benefit balance in favor of storing more detailed product data over longer timescales. As a starting point, we suggest that companies collect at least three years of sales, procurement, and production data in a single location, with clear mapping between data from different parts of the organization.
Companies then need to decide what they want to achieve through portfolio optimization. This requires cross-functional alignment on the desired overall impact of the effort and on the parts of the portfolio that will be targeted for complexity reduction. These conversations, especially the latter, can be difficult because different stakeholders across the organization will have their own concerns, preferences, and “sacred cows.” A first-principles approach, with up-front agreement on the assumptions used to estimate complexity costs, can help facilitate a fact- and data-driven discussion of contentious issues.
Once the analysis is complete, similarly disciplined cross-functional collaboration is required to define and then execute portfolio simplification initiatives. For these reasons, best-in-class companies prioritize eliminating entire platforms rather than individual SKUs to ensure that potential savings are maximized, considering the network effects across the portfolio. Ensuring that initiatives deliver these savings requires careful planning to define the necessary actions across the organization and rigorous tracking to ensure that these actions are completed. Senior leadership decisions may be required to resolve trade-offs and maintain the momentum of change. This approach would then enable the most successful companies to continuously build a muscle to prevent complexity from creeping back in and review portfolio health on an annual basis.
