Five pixels vanished from a LinkedIn ad banner. Then something larger vanished with them: roughly $1 million in revenue over a month. The change had looked harmless, the sort of visual adjustment that might glide through a review with a nod and a tasteful comment about breathing room. At LinkedIn's scale, however, the smallest piece of interface carried a balance sheet in its pocket.
Makram Mansour tells that story as a case for testing before release. It also serves as a compact map of his career. He began with the tiny things inside machines - current, delay, oscillators, circuit behavior - and moved toward the tiny choices inside products. In both worlds, intuition is useful, models are better, and evidence gets the final vote.
Today Mansour leads Marketplace at ID.me, the digital identity company, and advises experimentation platform Kameleoon. Before that came Intuit's central experimentation platform, IXP, and its company-wide Experimentation Everywhere initiative. Before Intuit came LinkedIn's internal platform, TREX. Long before either, there were semiconductors, simulations and an electrical engineering doctorate from the University of Illinois Urbana-Champaign.
An engineer learns to hide the difficult parts
Mansour's early published work lives in the precise dialect of electrical engineering: deep-submicron CMOS devices, interconnect models, power, delay. His 2004 doctoral dissertation studied low-phase-noise oscillator design. The titles are formidable. The instinct beneath them is plain enough: describe a complicated system accurately enough that someone can make a better decision.
That instinct followed him through engineering roles at Intel, Fujitsu, Berkeley Design Automation and National Semiconductor. At Texas Instruments, it became more visibly human. Semiconductor design could feel like black magic to a new user. Mansour worked on online WEBENCH tools that let people enter what they needed while software handled the equations, component selection and simulation behind the screen.
Hobbyists were moving into robotics and LED projects. They did not necessarily have an applications engineer sitting beside them, nor should they have needed one. A simple interface opened a door into difficult technology. This is the first recurring move in Mansour's work: do not remove the sophistication; place it somewhere the user does not have to wrestle with it.
“Don't push tech on people, push solutions to problems.”Makram Mansour
The phrase sounds obvious. Most expensive mistakes do, once phrased correctly. Product organizations are especially talented at falling in love with their machinery. A new model, platform or architecture arrives wearing the fragrance of destiny. Mansour keeps dragging the conversation back to the problem statement: Who has this problem? What evidence says it matters? Which assumption, if false, makes the rest of the plan decorative?
The dignity of a feature that disappears
At LinkedIn, experimentation became Mansour's full-time medium. He managed TREX, the platform behind tests across the professional network. The job was partly technical and partly constitutional. A good experimentation platform determines how a company settles arguments. It defines which metrics count, which guardrails cannot be ignored and how much evidence is enough.
Consider LinkedIn Stories. The format had worked brilliantly in a social setting elsewhere, and someone joining from Instagram helped bring the idea into the room. LinkedIn did not need a symposium on whether professionals yearned to post ephemeral updates. The team built a version, launched it and watched. It did not spread as hoped. Professional networking had different use cases. Stories went away.
The team also tried a business-card scanner for Japan, where exchanging cards has deep professional significance. Again, plausibility was not adoption. The scanner failed to catch on. These are easy stories to flatten into failure theater: brave team tries, data speaks, everyone applauds. The harder truth is organizational. People have to be rewarded for exposing a weak idea early, even after they have invested taste and reputation in it.
Mansour says mature companies understand that 80 to 90 percent of ideas will not work. The figure is less depressing than liberating. If the portfolio is expected to contain misses, a failed test need not become a trial of character. The aim shifts from always being right to becoming right at a tolerable price.
The program is the experiment
At Intuit, Mansour's canvas widened. He led work across experimentation, personalization and content-management systems, then moved into emerging products. Intuit contains businesses with different rhythms and customers - QuickBooks, TurboTax and Mailchimp among them. Running tests is one challenge. Making learning travel between those businesses is another.
His preferred metaphor is an operating system. People, process and technology have to cooperate. A splendid testing tool cannot rescue incentives that punish failed launches. A center of excellence cannot create institutional memory if teams bury their findings in private decks. Training matters because controlled experiments are powerful equipment; handing someone the keys without teaching them what the gauges mean is an unusually efficient way to produce confidence without competence.
“If you don't spread the learning, it's like you didn't learn anything,” Mansour has said. This may be his most useful line. An experiment that changes one decision has local value. A finding that prevents five teams from repeating the same mistake begins to compound. The real platform is not merely the software that assigns users to variants. It is the route by which a surprise in one corner becomes judgment everywhere else.
Is the problem coherent and the assumption explicit?
Will a real customer choose the proposed solution?
Can the idea produce durable economics?
For emerging businesses, the evidence cannot always come from a clean A/B test. There may be no traffic, no finished product and no sensible control group. Mansour advocates finding the leap-of-faith assumptions and shrinking them into minimum viable tests: rough prototypes, customer observation, a deliberately small demonstration. His “three fits” sequence asks whether the idea works on paper, in the market and in the bank. Each gate earns the expense of the next.
AI makes the steering wheel more important
AI has lowered the effort required to draft interfaces, write variants and build prototypes. Mansour is enthusiastic about that acceleration; he advises Kameleoon on AI integration and has praised prompt-based experimentation for removing setup work. Yet cheaper production does not make every idea wise. It moves the bottleneck.
When teams can generate more things to test, they still have limited traffic and attention. They must choose which question deserves an audience, decide what harm a guardrail should catch, and interpret results without confusing motion for progress. Five pixels remain capable of expensive mischief. Now they can be generated before lunch.
This explains why Mansour can argue both for high-velocity experimentation and for fewer, better tests. Volume is useful when it increases the search for good opportunities. It becomes theater when no decision changes. A team running one hundred thousand experiments has built an impressive scoreboard; whether it has built judgment is a separate inquiry.
“I've always believed that experimentation is supposed to help teams test ideas fast, not become a massive operational burden.”Makram Mansour
At ID.me, his official focus is Marketplace: commerce and loyalty built around verified communities. It is a new domain for the same operating habit. A marketplace has two sides, competing incentives and countless tempting optimizations. Trust is also unforgiving. Growth cannot be separated from the experience people have while proving eligibility and using a benefit. Success metrics need guardrails because a number can rise while the relationship underneath it thins.
Big-Mac in the Batcave
The severe version of an experimentation leader is a walking spreadsheet who refers to lunch as an uncontrolled trial. Mansour's public portrait is more playful. Colleagues call him Mac or Big-Mac. He loves Formula 1, Batman and comic books. He turned his garage into a Batcave for his car, an act of environmental storytelling that Bruce Wayne's facilities team would surely approve.
The hobbies fit. Formula 1 is a world where minute changes meet ruthless measurement, yet data never abolishes the driver's judgment. Batman is, among other things, a relentless prototyper with a memorable design system. Mansour's motto - “where there's a will, there's a way” - supplies the comic-book caption.
Still, his career makes a quieter argument than any motto. Expertise is not the ability to predict every outcome. It is the ability to design a useful encounter with uncertainty. Sometimes the answer is an online tool that hides the semiconductor math. Sometimes it is a test that removes a beloved feature. Sometimes it is a shared archive that saves another team from learning the same lesson at full price.
Five pixels can cost $1 million. They can also buy an education, provided the organization is paying attention.