Overlooked Software Performance Flaws That James Pulley Says Are Destroying Revenue During Peak Seasons

Every year, as peak retail seasons and high-profile events roll around, the same headlines appear: websites crash, transactions fail, and customers take their business elsewhere. James Pulley, a veteran performance engineer with decades of experience solving complex software failures, has seen it happen too often. "There is so much that can be done that is low-cost and high-value to prepare organizations," he says. "The problem is, they rarely act soon enough."

Pulley's work spans global product launches, high-scale digital rollouts, and retail platforms that handle millions of simultaneous users. Across industries, he has found three recurring and preventable issues that can cripple performance during traffic surges.

The first is allocating resources too early. Many organizations hand out system resources to users the moment they land on a site, whether they are about to buy or just browsing. "If those resources are not on the revenue path, you are distorting performance for the people who are," Pulley says. He compares it to starting a game of Tetris and immediately filling the screen with blocks. "There's a finite number of resource 'blocks.' If you give them out too early, they are just sitting there, unavailable to someone ready to check out." The fix is simple: delay allocation until the user actually needs the resource. This frees capacity for high-intent customers without adding cost.

The second problem is holding resources for too long. The opposite of early allocation, this issue arises when inactive sessions or shopping carts are left open far beyond their useful life. Marketing teams sometimes push for 30-minute or longer timeouts, hoping customers will return and complete a purchase, but data shows most returns happen within minutes or after many hours, making the extended hold unnecessary. Pulley recalls an anonymized case from a major global brand whose shopping cart system slowed every page load. "When they fixed the cart allocation issue, 97% of requests to the cart system disappeared overnight," he says. "The painful part is that marketing resisted the change for two years, that's around $240 million in lost sales opportunity."

Perpetual carts, items sitting for months or even years, create a different risk. Not only do they consume unnecessary disk space, but if overloaded, they can fragment memory and even be exploited for denial-of-service-style disruptions. Pulley advises setting triggers based on cart size, age, or value to automatically clear or archive stale data.

The third issue is a mix of poor cache management and excessive monitoring. Content Delivery Networks are meant to ease the load on origin servers by storing and serving cached content from locations closer to the user.

Yet Pulley finds many organizations fail to cache enough content, or they overload pages with marketing analytics scripts that slow everything down. "If you have 15 people in marketing, you'll probably have 15 beacons loading before your actual content," he says. "Every extra script adds delay, and there's a tipping point where users think the site's broken." He notes that if a site is unresponsive twice in a session, most visitors will leave for good. His solution is to cache as much as possible and consolidate analytics through a proxy service that sends one data stream to multiple tools. "Otherwise you end up in what I call 'quantum monitoring', where the act of measuring performance is what's degrading it."

While these problems cause losses year-round, their impact multiplies during predictable high-traffic periods, Black Friday, product launches, sports merchandise releases, and ticket sales for major tours. Pulley recalls a sports retailer whose site failed both on the announcement of a high-profile player signing and again when jerseys went on sale. "They didn't manage the cache on the news page, and they handed out a default cart to every visitor. It was a perfect storm."

Pulley emphasizes that none of these fixes require massive infrastructure overhauls. "These are configuration and process changes. You can implement them quickly and cheaply, but they require cross-team cooperation," he says. That often means overcoming resistance from marketing or other stakeholders who prioritize visibility over speed. From his perspective, the conversation needs to shift. "Performance is not a technical metric; it's a revenue metric. Faster systems convert better, retain more customers, and protect long-term brand value. The question is not whether you can afford to make these changes. It's whether you can afford not to."

In October of this year, Pulley will be in Nashville delivering a presentation on the intersection of artificial intelligence and software performance. "AI can help in every stage, from gathering requirements to analyzing results, but ultimately, you have to get the fundamentals right," he says. "You can't optimize what's broken."

Whether it's allocating smarter, releasing resources faster, or reducing self-inflicted monitoring slowdowns, Pulley's message is clear: peak-season failures are not inevitable. With foresight, discipline, and a willingness to challenge old assumptions, companies can avoid the headlines and the lost revenue that come with them.