In 1992, by Robert Kaplan and David Norton’s account, Mobil’s U.S. marketing and refining business had a negative cash flow of $500 million. Its profitability, they wrote, put it in last place in its industry. Its product was gasoline, which looks the same from every pump.
Mobil’s marketing research found that the drivers were not the same. Some cared only about price. For them, Mobil would have to match or beat the discount stations.
Others, Kaplan wrote, would pay a premium of up to 10 or 12 cents a gallon for a better buying experience: quick service, paying by card at the pump, clean restrooms, friendly employees, a great convenience store.
Mobil could not be the best for both. Bigger stations, more self-pay pumps, better-trained and better-paid staff and full convenience stores cost money that only higher prices could cover, and higher prices would drive the price shoppers away. “Strategy is about choice,” Kaplan wrote. The harder question came after the choice. How does a clean restroom in one town turn into profit for a $20 billion division?
Arrows on a page
Kaplan and Norton’s tool for that question was a drawing. The balanced scorecard already looked at a company from four perspectives: financial, customer, internal processes, and learning and growth. “It soon became natural,” Kaplan wrote, “to describe the causal relationships” between them. His simplest example runs from bottom to top: employees trained in quality tools cut defects and cycle times; better processes mean shorter lead times and on-time delivery for customers; satisfied customers stay and spend more; revenue and margins rise.
Draw the chain from today’s work to tomorrow’s results, then check each link. That diagram is a strategy map. Kaplan and Norton set it out in the Harvard Business Review in 2000, with Mobil as the worked example.
A map can also leave things off. Mobil’s main suppliers sold crude oil or built refineries and pipelines. They were essential, Kaplan wrote, but did nothing to set Mobil’s strategy apart. He used them to show that suppliers like these need not appear on a scorecard.
The map was meant to be read by more than the executives. Norton described an annual employee survey at Mobil’s division that asked whether people understood the strategy. At the start, in 1993, he said, 20 percent did. Five years later, 80 percent did.
The arrow that ran backward
By the authors’ account, it worked. Mobil’s division moved from last place to first in industry profitability by 1995 and stayed there for five consecutive years, through 1999. Its cash flow went from minus $500 million in 1992 to plus $900 million in 1998.
That is their telling, about one division, and the map is not the only thing that changed in those years. Kaplan was frank about the diagram’s limits. In 2010 he wrote that one might even describe the generic strategy map as a “dumbed-down” representation of causal links, and that a fuller model would estimate the size of each effect and how long it takes to arrive.
Arrows can fail. At Analog Devices, the company whose scorecard helped inspire the balanced scorecard, every quality measure on the scorecard improved while the stock fell nearly 70 percent in three years, by Kaplan’s account. The scorecard’s creator later found delivery and stock price moving in opposite directions over seven years. Kaplan and Norton wrote in 2001 that some companies, Sears among them, tested their maps with statistical correlations between the measures, to check whether, say, empowering employees really raised customer satisfaction.
The bakery’s arrows
Imagine you run a neighborhood bakery and decide to compete on freshness, not price. Your map: bake a second batch at 2 p.m., so afternoon customers find warm bread, so they come back more often, so weekly sales rise.
Each arrow can be counted. Did the second batch go out on time? Did afternoon sales rise? Did the same faces come back, which a loyalty card would show? If the second batch sells but regulars don’t return any more often, the map was wrong about the third arrow, and you have learned it in a month instead of a year.
Mapping an AI rollout
The same drawing helps before you put AI to work. A common map for a support desk: an AI assistant helps agents, so agents handle more chats, so customers are happier, so costs fall and customers stay.
Erik Brynjolfsson, Danielle Li and Lindsey Raymond measured several of those arrows at one Fortune 500 software company that gave an AI assistant to its customer support agents, 5,172 of them in their data. Chats handled per hour rose about 15 percent. Resolution rates barely moved. Customer satisfaction, measured by post-chat surveys, showed no economically significant change on average. Customers’ messages did become more positive, fewer asked to speak to a manager, and fewer agents quit, mainly among newer workers. The most experienced, highest-skilled agents saw small gains in speed and small declines in quality.
The authors say these are medium-run effects inside one firm. Even so, a map that pinned cost savings on surveyed satisfaction would have had one arrow with no clear change found on average and another that ran differently for the best agents. Draw your map before you roll out an agent, pick one number per arrow, and check each arrow separately for new and experienced staff.
Standing at the pump
Mobil’s research had found drivers who would pay for a clean restroom. The map was the promise that the restroom would be clean on the day they stopped.
Five years after the first survey, by Norton’s count, four in five people in the division said they understood the strategy.