Inside $3.5 Billion of Paid Growth: Creative Testing, Attribution Humility and Channel Diversification

Becker began as a web developer who fell in love with SEO, then moved into paid search because clients wanted measurable ROI that organic rankings could not clearly provide. What started as a freelance practice run from a walk-in closet grew into Thrive Digital, a 130-person agency managing roughly $500 million a year in ad spend for clients such as Uber, Asana and Tempur-Pedic.

The Uber relationship began in 2013 at the TED conference, when he shared a taxi with Uber founder Garrett Camp. Becker had been using paid search to sit on Uber-related branded keywords and pocket $20 referral credits per signup, earning tens of thousands of dollars. He told Camp about the loophole, expecting trouble, and Uber's team instead hired him to fix the problem, a relationship that lasted ten years.

Becker's moral is that people often call successful founders lucky when in fact you have to make your own luck. He took a small risk by speaking up, with little to lose, and says many people never make such moves because they fear the downside.

Asked whether paid growth is a habit that startups should avoid, Becker says it is dangerous only when a business relies on it entirely. He compares the marketing mix to an investment portfolio: you would not put all your savings in a single stock, and you should not expose revenue to the volatility of one channel's rising CPCs.

He says he thinks of himself as a fund manager handling other people's money, and that his first rule is not to forget offline and classic marketing. Direct mail, email and SEO all work. The crash-and-burn cases he has seen involve brands that scaled on a single platform loophole, which later closed.

Becker says the question changes with company stage. Early on, the issue is whether you have product-market fit and whether you are over-relying on paid. Later, the question is no longer whether it works but at what scale, which depends on creative resources, professional marketers on staff and technical capacity to fix tracking.

Quick payback suits direct-to-consumer and e-commerce businesses, but lead-generation businesses must model abstract concepts such as lifetime value to CAC and payback period. With lead scoring models, he says, you can predict which leads are likely to convert and bid accordingly, even when revenue arrives much later.

His rule of thumb is that if other channels, such as email, SEO or direct mail, already work for you, paid is likely to work too. The common mistakes are assuming that platform data gives complete end-to-end attribution and expecting an overnight turnaround, since every business has different metrics and results.

Becker warns against survivorship bias: every 2010s unicorn used performance marketing, but plenty of companies that did so flatlined. Companies still scaling largely through paid, such as Grammarly and Athletic Greens, understand cost per acquisition against lifetime value and know how much they can spend before ROI falls off a cliff.

He says the privacy changes from Apple and the 2022 economic downturn hit ad channels hard, but expects the storms to pass. Successful teams today combine rigorous creative testing, a clear view of attribution's strengths and weaknesses, validation work, and a focus on bottom-line profitability rather than raw growth.

On SEO versus paid, he says they are not mutually exclusive and doing both supports diversification. Paid attribution is more linear, which finance teams like, while SEO runs through a black-box algorithm with hundreds of signals. That tangibility is why he leaned into paid, though he says good SEO has indisputable payoffs.

Becker explains that creatives are the assets behind social and display campaigns, from motion graphics to user-generated video. Google and Facebook have automated most of the manual levers over the last decade, so creative is one of the few areas left for humans. A common failure is a brand design team handing over a pile of assets with no sense of how the funnel works, or repeating one message to everyone and causing banner blindness.

His testing structure is simple: one audience at the ad set level, with two nearly identical creatives that differ by a single variable such as copy or image. Because ad servers may deliver unequal impressions, he normalizes with metrics like click-through rate. Winners become the base for the next challenger, so learnings flow back to the creative team continuously.

Two examples stand out. Highly polished brand ads regularly lost to unpolished, iPhone-shot influencer content because it felt authentic. A furniture company's social performance unlocked when an art director put the office dog on the couch, which Becker says doubled or tripled ROAS. Lenny adds a parallel from Airbnb, where listing photos without people converted better.

Becker stresses that the real question is whether a company finds such insights by accident or builds a rigorous process around testing. He adds that no exotic tooling is needed, since most testing can be done inside Meta Ads and Google Ads themselves.

By Google, Becker means the whole ecosystem including YouTube, and by Facebook, Instagram and probably WhatsApp soon. He sees interesting opportunity in TikTok and Amazon. TikTok's ad platform reminds him of Facebook six or seven years ago: less sophisticated with weaker attribution, but cheaper clicks because fewer companies have figured it out. Amazon, he says, is specific to direct-to-consumer, not B2B SaaS.

On TikTok, founder-led brands filming handheld iPhone videos and companies that partner well with influencers do best. Anything direct to consumer should investigate it, he says, and someone will eventually crack B2B on TikTok. The catch is that the platform is extraordinarily creative-dependent, needing new assets several times a week, so getting to a million dollars a month takes a creative machine.

For B2B SaaS he is channel-agnostic and focuses on how to think about spending. The common mistake is optimizing for the cheapest cost per lead, assuming more leads at the top of the funnel means more revenue, when not all leads and sales are equal.

Becker argues that focusing first on cost per lead pushes teams toward cheap, low-quality leads. If you flip the question and target high-value customers, cost per lead may rise but ROI rises too. Getting there, he says, requires joining revenue data to channel data.

Thrive built an internal ETL tool called Thrive Stack, and he points to Supermetrics as a commercially available data connector. The idea is to pipe CRM revenue into a database, join it with channel data, and see which audiences produce real revenue. A lead scoring model then lets you bid in real time on audiences likely to become high-value customers.

The problem it solves is slow gratification. When opportunities turn into revenue after two, six or even twelve months, you still need to decide today how aggressively to bid. A lead scoring model supplies a predictive basis for that decision.

Becker defines attribution as determining the relationship between what you did and what happened. He cites John Wanamaker's 1919 line that half of marketing spend is wasted and nobody knows which half. Classic cookie-based models include last-click, first-touch and weighted multi-touch, and choosing one is subjective, driven by whether a business prioritizes profitability or growth.

Apple's IDFA changes in mid-2021 let users opt out of sharing identifiers, which removed core data Facebook relied on to validate attribution. Teams became less sure which audiences to target and how much revenue campaigns produced. The modern response combines cookie-based attribution with statistical modeling, customer and population surveys, and media mix modeling, a regression-based approach dating from the 1950s. He mentions Recast as an off-the-shelf tool, and Lenny discloses that he is an investor in it.

His advice is that anyone claiming to be a single source of truth is not accurate. Attribution is an ongoing investigation for evidence. His team has built sophisticated models that MMM or other tools later invalidated, showing campaigns that did not work, which he calls an important finding for deciding whether to keep investing.

Becker notes that the industry has been shaped by AI for over a decade because Google, Facebook and Microsoft automated their platforms. The result is displacement of work, not of people: his team is larger than ever but does more modeling, validation and creative strategy and less keyword-level bid management. He suggests this may foreshadow other fields, such as architecture offices doing less drafting.

On a practical level, ChatGPT generates copy variants for testing and drafts RFP responses from past ones. He says the output is about 80% good and still needs ten hours of editing, but replaces a week of work by five or six people. His creative group uses DALL-E and Midjourney for mockups in about 1% of the previous time, and can re-prompt live on a client call.

Asked whether AI takes away the fun parts, Becker says it democratizes visual expression and that his team, which has not let anyone go because of these efficiencies, is excited. He compares it to the printing press and the loom, accepts that some roles will be displaced, and expects a later generation to see AI as simply normal.

Becker says agencies are not an alternative to in-house staff. Thrive almost always reports to a professional marketer, and he advises starting with in-house expertise and then adding an agency. At earlier stages agencies help with experimentation and scaling faster, while at later stages they maintain a sophisticated machine of testing, attribution and reporting that small internal teams rarely have the resources to build.

For hiring, he looks first for a technical background, citing a nuclear physicist, mathematicians and finance people on his team, then for hands-on experience across Meta, Google, TikTok and Amazon, and an understanding of creative and attribution. Years of experience matter less than demonstrated ability to run channels competently.

He also borrows a Jeff Bezos-style interview question, asking candidates how many windows there are in New York City. He is less interested in the number than in whether the candidate thinks on their feet and stays composed when surprised, which predicts how they handle clients. Job titles vary widely, so he suggests borrowing descriptions from sites like LinkedIn and Glassdoor.

Becker's second landing story comes from a 2015 or 2016 RFP from Snap in Los Angeles. The night before, he and his team struggled at 2:00 a.m. to turn a long list of zig-zagging stakeholder questions into a coherent presentation. They worked until 6:00 and decided to build something bespoke instead.

In the pitch he told about 20 executives he would not answer any of their questions and would instead explain what they needed to do to get where they wanted to go. If they agreed with his logic they should hire Thrive, and if not, they should not. Two hours later, while driving to San Francisco, he got the call that they were hired.

His lesson is to trust yourself and your strengths. When a well-meaning client does not know what it needs, be brave enough to tell them rather than conforming to what they ask for.