Jon Lorenzini Wants Marketing’s Prettiest Graphs to Earn Their Keep

Jon Lorenzini stepped in front of more than 300 executives, analysts and clients at a Google data summit and promised them graphs. It was just before the break, a dangerous hour for any speaker, and he knew exactly how romantic the proposition sounded. The charm came from acknowledging the problem. Marketing measurement is awash in lovely curves, handsome dashboards and reports with the polished serenity of objects that may never be touched again.

Lorenzini’s standard is less serene. If a measurement company cannot improve performance by more than it costs, he has said, the client should fire it. If the client learns something and never acts, the arrangement has failed in a different way. The point is not to admire a dashboard. The point is to make a demonstrable change.

This is the governing idea of his career: evidence ought to end with a verb. Increase. Cut. Reallocate. Test. At LiftLab, where he is Chief Marketing Science Officer, the verbs often concern large media budgets. Yet the instinct was formed over a long apprenticeship in the less glamorous machinery of advertising, close to the campaigns, clients and compromises that produce the data in the first place.

He learned the system by changing seats

In 2008, Lorenzini was a social media manager at Virgin Mobile in New Jersey. He then crossed to the agency side: print activation at MediaVest, search analytics at MediaCom, digital analytics at ZenithOptimedia. The client names attached to those jobs - Bristol Myers Squibb, Dell, General Mills - hint at the range of media problems passing across his desk.

Then came the platforms. At Google, beginning in 2013, he led analytics for food, beverage and restaurant advertisers before moving into brand insights and reach-based measurement. A global consumer-packaged-goods measurement role followed at Facebook. At Snapchat, from 2020 to 2022, he led a marketing science group covering retail, restaurants, travel and energy.

The sequence matters because advertising looks different from each chair. A brand knows its margins and anxieties. An agency knows the awkwardness of execution. A platform sees immense behavioral signals and also the boundary of its own garden. Lorenzini once offered a neat example from Snap: a teenager sees an ad, then a parent buys. Identity-level attribution can be precise about the person it recognizes and still miss the household where the purchase occurs.

“The real value of measurement only shows up when it drives change.”Jon Lorenzini, on his Google summit talk

The argument before the job

Lorenzini’s account of joining LiftLab begins at an ex-Googler gathering in New York. He met John Wallace, now his CEO, and the two argued for a while. This was not social failure. Lorenzini likes the Socratic method: defend an opinion, worry at its language, discover how the other person thinks. “Horns locking,” as he called it, can be informative.

Wallace later approached him with an idea for creative A/B testing software. Lorenzini thought the proposal was terrible. Matching enough identities through Snap to create a useful sample would be messy. The disagreement did not end the relationship. It clarified the problem worth solving. Impact, education and compensation are Lorenzini’s stated order when weighing opportunities, and LiftLab offered the first two in abundance.

It is a revealing origin story for a measurement executive. Trust did not begin with instant agreement. It grew from the ability to expose assumptions without making the disagreement personal. That habit now appears in Lorenzini’s descriptions of good marketing science. Modelers need the marketer’s business context. Marketers need enough transparency to understand when a signal is strong or weak. Neither side gets to retreat behind vocabulary.

The next dollar tells on the average

Lorenzini’s preferred north star is marginal return. Average return asks how a channel has performed across all the money already spent. Marginal return asks what the latest dollar produced, or what the next one is likely to do. The difference is not academic. A channel can boast a fine historical average while its newest spending has wandered onto the flat, expensive end of a diminishing-returns curve.

Imagine two channels. One has received plenty of money and now strains to find fresh customers. Another has a lower average return but room to scale. Moving budget from the saturated channel to the underfed one can increase revenue without increasing total spend. In Lorenzini’s summit example, the back-of-envelope gain was $500,000 on the same budget. The curve became useful only when somebody moved the money.

This is why he wants the CMO to enter a meeting with the CFO carrying a bottom-up forecast rather than a mist of marketing confidence. Finance may set an attainment target. Marketing should be able to answer what media can contribute, what another increment of budget is likely to return, and where the model is uncertain. The useful output is not “our iROAS is two.” It is “at $1 million, should we spend $1.1 million or $900,000?”

Let the experiment argue back

Models are necessary because marketers want context across television, search, social, affiliate, print and out-of-home. Tests answer a narrower causal question. Lorenzini’s position is that the two methods should not behave like rival camps. A model can identify where evidence is weak. An experiment can create better evidence at different spending levels. Those results can recalibrate the response curve, which improves the next forecast and helps choose the next test.

“Experiments shouldn’t just validate your model,” he wrote in 2026. “They should fix it.” The distinction is sharp. Validation can become ceremony, a test performed to place a gold star on an existing answer. Repair admits the model is provisional. Lorenzini is fond of the statistician George Box’s idea that all models are wrong and some are useful. Usefulness grows when weak signals are labeled, challenged and strengthened.

His move from identity-level tests to geo-testing at LiftLab reflects the same pragmatism. Individual-level experiments offer rigorous control but can struggle when scaled audiences change or when platforms rely on different identity graphs and match rates. Geo tests exchange some neatness for a view of whole markets. They create their own headaches, from media prices to regional differences, but they also observe effects that do not fit tidily inside a cookie or login.

Serious methods, useful mischief

For someone occupied by rigor, Lorenzini has little reverence for the costume of seriousness. His LinkedIn name includes “JLo.” His personal homepage announces a preference for iterative improvement over delayed perfection, then concedes that he does many things, “some of them well.” It links to WhimWe, a community project involving margaritas and Mexican food. The résumé shares space with the side quest.

The jokes often smuggle in a methodological point. In one deadpan video from Japan, he promised advanced decomposition, “mastication techniques” and an analysis of extraterrestrial paths to purchase. In an April Fools’ announcement, he unveiled a fictional last-click product that crowned branded search the supreme advertising channel. Anyone who has watched a conversion report give final credit to the easiest click could recognize the target.

His more direct skepticism is just as pointed. New advertising channels should be treated as experiments, he argues, and their self-reported analytics deserve caution. Artificial intelligence needs a job, not merely a launch. Writing about LiftLab’s AI tool Miles, he emphasized its ability to translate client questions into structured database queries. “Without a clear problem statement, AI tools largely go unused.” Novelty is not a use case.

“I look for the ones who make me a little jealous.”Jon Lorenzini, on finding practitioners to learn from

A practitioner among abstractions

Lorenzini’s advice for learning is to find people who make a skill look effortless, ask them questions and then do the work. Textbooks can provide a foundation. Mastery arrives through repeated practice and direct conversation with people who understand the context around the technique. It is an appealing philosophy for marketing science, a discipline that can become drunk on abstraction while somebody else is negotiating inventory, cleaning data or defending a budget.

He applies the principle to hiring, too. Calling hiring the most important part of his job, Lorenzini has written about looking for people who can enter the weeds and still summarize the wider picture. Autonomy requires trust; trust requires confidence in how someone thinks. The old argumentative meeting with Wallace appears again in miniature. The working relationship is partly a shared ability to expose reasoning.

The ambition behind all this is not to abolish uncertainty. Lorenzini’s public work is more interesting because it keeps uncertainty visible. Data can be sparse. Spend patterns can be flat. Tests can become expensive. Models can lack business context. A transparent system shows where the signal weakens and creates a plan to improve it.

Marketing has never lacked numbers. Its more persistent shortage is numbers that survive contact with finance, media operations and human doubt. Lorenzini’s contribution is a demanding, faintly mischievous test: make the graph explain the next move. If it cannot, enjoy its colors. Then get back to work.