# Paid acquisition: scaling spend and proving it works — Full Chat Transcript The Performance Marketing work group answers paid acquisition questions in group chats. They start with paid social that has stalled and ad costs that keep rising, then turn to proving what the spend drives. Updated: 2026-10-13 HTML page: [Paid acquisition: scaling spend and proving it works]() Summary: [Summary]() Work group: [Performance Marketing]() ## Participants - [Timothy Davis]() — Growth at ElevenLabs. Former Senior Lead, Growth Marketing at Shopify - [Mark Fiske]() — Managing Director at H.I.G. Capital. Former VP of Growth and Marketing at Credit Karma - [Daniel Pearson]() — Co-founder of Bamboo. Advisor to HexClad Cookware - [Mike Jacobs]() — Chief Growth Officer at Consortium Brand Partners. Former VP of Growth at Etsy - [Micah Moreau]() — VP of Growth at OpenAI. Former VP of Growth Marketing at DoorDash ## Full Chat Transcript #### [Scaling paid social past a plateau]() **Question:** How do you keep growing paid social spend once results plateau? ##### [Timothy Davis]() When paid social plateaus, I start by asking why before adding budget. A plateau is rarely one thing. In my experience it's usually several smaller problems stacking up: creative that has run too long, audience lists that have gone stale, or a landing page change that quietly broke the match between the ad and the page. Fix those first, because more spend on a fatigued account only buys you more expensive results. After that, these are the levers I pull, in order: 1. Creative volume and refresh. Creative fatigue is the most common ceiling I see. Test formats systematically and refresh on a schedule so you aren't waiting for performance to fall first. 2. Wider lookalike bands. Build lookalikes from your own customer data and split them into separate ad sets: 1%, 2–4%, 5–7%, and 8%+. The 8%+ band usually doesn't work, but sometimes it does, and that's where new volume comes from once the tight bands are saturated. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] 3. New markets. At ElevenLabs, a team of four paid marketers who all speak one language runs and proves profitability on campaigns in multiple languages, because AI handles the localization. A new language is a fresh pool of audiences and auctions. Localization shouldn't be the reason you don't enter a new market. 4. Incrementality before you scale further. Rerun lift studies. Sometimes a channel turns out to be more incremental than attribution showed, and that justifies pushing harder at a "flat" ROAS. Sometimes the lift study shows the opposite. [Citation: https://www.youtube.com/watch?v=kv4TRa_J2dE] Below roughly $50K a month on a platform, you won't have enough signal for the test to be worth running. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] ##### [Mark Fiske]() Timothy's lift-study point is right. The failure I see more often sits upstream of the test, in how the channel team is graded. In a lot of portfolio companies, the channel manager has an efficiency target that should be profitable, but it's graded only on publisher reporting, so Facebook is effectively saying "what I'm doing is right." That number doesn't account for incrementality or diminishing returns. A flat platform ROAS can hide real incremental lift, and a healthy ROAS can hide spend that's already cannibalizing organic. Before you push budget or pull it back, line up channel reporting with the company-level numbers. That usually means blending platform data with post-purchase surveys, and for brands with an offline mix, running matched-market tests. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] Next, change the constraint you scale against. I care most about payback period because it tells you how fast the cash comes back for reinvestment. Moving from a blended CAC target to a marginal payback target, supported by a granular LTV model, often gives you headroom you can defend to finance. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] [Citation: https://builtin.com/articles/marketing-budget] The third lever is the slow one: creative and channel diversification. Teams got spoiled by Facebook and PMax, where you turned it on and it worked. TikTok and the newer channels make you learn the creative formula and the buying mechanics from scratch, so you need a roadmap and explicit expectations. A $200 test over six weeks tells you nothing. Often the diversification roadmap shows you're overinvested in parts of Google, and that money can fund the next channel's learning curve. [Citation: https://youtube.com/watch?v=DdqCHnovono] Most plateaus I see are measurement or payback-target problems that look like media problems, so fix how spend is judged before deciding the channel is tapped out. ##### [Timothy Davis]() I agree with Mark that grading the channel on publisher reporting hides the real picture. One more cause gets missed, though: the seed data feeding the algorithm. Teams build lookalikes once from a customer list and never refresh it, so the platform keeps optimizing toward a customer profile from a year ago. Stale audience data is one of the main reasons volume declines, and it looks exactly like saturation. Re-upload your current customer base on a regular cadence. Better still, seed from your highest-value customers instead of everyone who converted, and then rebuild the 1% through 8%+ bands from that list. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] For B2B, the next unlock is adding third-party buyer intent data on top of job title and industry targeting. That gives you new high-propensity pools to buy into without lowering the bar to broad audiences. Judge those pools on downstream quality. High traffic volume from a new audience means nothing if it doesn't convert past the top of the funnel. ##### [Mark Fiske]() Seeding from your highest-value customers only pays off if the rest of the system is graded on value too. If the channel team still has a blended CPA target, the platform will drift back toward the cheapest converters, and you lose the gain you got from a better seed list. The seed, the bid signal and the efficiency target all have to point at the same customer. That leads to the lever most paid teams don't own: raising what you can afford to pay. Acquiring the user is only step one. If onboarding, engagement and retention raise LTV, the marginal payback on the next dollar of social improves without any change to the media. In diligence I look at payback period first because it shows how quickly the money can be reinvested. A company that shortens payback from 9 months to 6 can keep scaling into more expensive auctions that a competitor stuck on a front-end CPA target has to walk away from. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] In practice, put the retention and lifecycle owner in the same weekly review as the paid social buyer, and report LTV by acquisition cohort and audience band. That shows whether the 5–7% and 8%+ lookalike bands Timothy described bring in customers who stay, or just cheaper first purchases. ##### [Timothy Davis]() Mark's org point is the part that holds up in practice. At ElevenLabs, much of our scaling came from performance, creative and lifecycle working as one team instead of three separate owners passing work between them. When the lifecycle owner sees which audience band a cohort came from, paid can defend a higher CPA on a band that retains. Paid can also cut a band that only looks cheap on first purchase. I'd add one thing to Mark's cohort report: include creative. Some creative concepts bring in buyers who stay, and others bring in deal-seekers. If you only judge creative on front-end CPA, your refresh cycle will keep picking more of the creative that attracts deal-seekers. ##### [Mark Fiske]() Agree on adding creative to the cohort report. The practical problem is lag. A 6- or 12-month LTV read arrives long after the creative has been refreshed and the budget moved, so the buyer goes back to front-end CPA because it's the only number available this week. Fix that by agreeing with finance on an early proxy that predicts retained value, such as second purchase or week-4 activity, validating it against mature cohorts, and then setting paid social targets on the proxy by creative concept and audience band. This is where budgeting maturity matters. Growth-stage companies can optimize to LTV only if they understand it at a granular level, and that understanding has to be built before plateau pressure hits. [Citation: https://builtin.com/articles/marketing-budget] If finance trusts the proxy, paid can defend a higher CPA on concepts that bring in customers who stay. If finance doesn't trust it, every plateau ends in a budget cut. ##### [Timothy Davis]() I agree with using an early proxy, but it has a volume limit. The proxy is a deeper-funnel event, so it fires less often than a purchase. When you slice it by creative concept and audience band, each cell can end up with too few events to read. The same thing happens with incrementality tests: below roughly $50K a month on a platform, the signal is too thin and the analysis costs more than it tells you. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] For smaller accounts, read the proxy at the concept level first, and only break it out by band after a concept has enough volume behind it. ##### [Mark Fiske]() Timothy is right about cell size, and it changes which tools a smaller account should use. Below that roughly $50K a month threshold, I'd stop chasing statistical precision and use the simpler methods that still work: post-purchase surveys blended with platform reporting, and accepting that the read won't be perfect. Brands without enterprise-level modeling budgets are bringing judgment back into measurement this way, and it beats letting Facebook's own reporting decide by default. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] The proxy and the survey answer the same question from two directions: is this concept bringing in customers we want? When both point the same way at the concept level, that's enough to keep scaling while volume builds toward a band-level read. ##### [Timothy Davis]() That works for the small-account stage. When an account crosses the threshold, switch to the lift tools the platforms already offer. Meta, Google and the others each have a built-in way to test incrementality, so a dedicated modeling budget isn't needed to get started. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] Treat each lift result as a multiplier on what the platform reports, and rerun the test regularly. User behavior changes, and a channel that was more incremental than you thought last quarter can come back lower on the next read, or the reverse. [Citation: https://www.youtube.com/watch?v=kv4TRa_J2dE] The survey and the proxy then become a check between lift tests. ##### [Mark Fiske]() One caution on the platform lift tools: Meta's lift study is Meta measuring Meta. That's useful as a multiplier, but once paid social is a meaningful share of the mix, especially next to TV or other offline spend, I'd regularly check it with an independent matched-market test. Matched-market testing has been around since the 1960s and it still holds up, because it measures what the whole media mix does in a geography rather than what one publisher can see. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] If the platform's lift result and the geo test disagree by a wide margin, trust the geo test when you make the decision to push spend past the plateau. ##### [Timothy Davis]() I agree, and the geo test fits well with the new-market lever from my first answer. When we expand into new languages, launches go out market by market, so the markets that haven't launched yet already work as holdouts. Plan the rollout order with the geo test in mind and you get a matched-market read on the new spend without setting up a separate test. ##### [Mark Fiske]() That works as long as the holdouts are real matches. A newly launched language market and an established market differ in brand awareness, organic demand and pricing, so a comparison between them mixes the launch effect with the market difference. Pair each launch with a holdout market of similar size and maturity, and record a few weeks of baseline before spend starts. Without that baseline, the rollout will show that the new market grew, but it won't show how much of that growth paid social caused. ##### [Timothy Davis]() Fair correction, and I accept it. The baseline weeks also tell you something about localization quality. If a new language market converts well below its matched pair once spend starts, look at the localized creative and landing pages before you blame the audience. A broken match between the ad and the landing page is one of the main ways performance drops without anyone noticing. ##### [Daniel Pearson]() I disagree with Mark on diversification as a plateau lever. If Meta is still returning well on the marginal dollar, moving budget to TikTok or a new channel takes money out of your best-ROI ads and puts it into worse ones while you climb a new learning curve. When marketers tell me "we're too dependent on Instagram," I ask them what happens if things go right. For a team that knows its numbers, the answer is usually to concentrate harder on what's working. [Citation: https://danielpearson.substack.com/p/stop-diversifying-your-ad-budget] In practice, putting more into your best channel means producing more creative for it, because creative output is usually what caps how much a winning channel can absorb. HexClad's Head of Growth drives hundreds of millions in revenue with four ICs. One freelancer runs Meta and TikTok, one person runs Google and YouTube, and two people make ads full time. Half of that team builds creative. If I'm staffing for more paid social spend, I hire another creative maker before another buyer or another channel. Payback period is the right constraint, and Mark and Timothy have covered it well. One caution from my Bamboo years: most teams I talked to tracked six or more different success metrics (installs, CPI, D1, D7 and D28 ROAS, CAC:LTV), and few actually optimized to payback across every account. That was usually because they lacked org buy-in and cohort infrastructure, and the right KPI was never the issue. [Citation: https://danielpearson.substack.com/p/payback-periods] Mark's early-proxy agreement with finance is how you get that buy-in. Diversification makes sense when marginal returns in your main channel have actually fallen below what a new channel can deliver. You should prove that before you spread the budget out. ##### [Micah Moreau]() Daniel's test for diversification is to prove that marginal returns in the main channel have actually fallen. That proof depends on a correctly fit curve. Marginal targets only help if the spend curve behind them is current. A common mistake is fitting one curve across a business that changed partway through the data. At Hims & Hers, the Weight Loss spend curves mixed Q1 seasonality and the Novo partnership period into one aggregate curve. When we re-fit them on July '25 to present data only, Meta's marginal cost per subscriber dropped from $26,000 to $850. Long-term LTV was $1,020, so a channel that looked plateaued actually had room to grow, as long as we accepted some short-term profit in exchange for revenue. Google's re-fit came out at $2,300, still above LTV, so we held Google flat. The stale aggregate curve would have sent budget away from Meta, which was the channel with headroom. If the business has shifted, one aggregate curve is wrong. Split the curve at the shift points before you decide anything. A team gets the math wrong if it reads blended averages. At Jet we used this table: spend goes from $1M to $2M to $3M, and blended CAC moves from $20 to $27 to $33, which looks fine against a $35 target. The marginal CAC on that last $1M was $66.67. Timothy and Mark covered creative and lookalikes. Two levers they haven't mentioned: 1. Advanced pixel matching. It raises match rates and conversion lift, and it usually just needs a sprint from product and engineering. 2. Separate targets for prospecting and remarketing. If both share one blended target, remarketing's cheap conversions cover for weak prospecting, and the account settles into harvesting demand it already has. Give prospecting its own marginal target. Prospecting is the only place new scale comes from. Keep scaling for as long as marginal cost per acquisition on the current curve stays under LTV. ##### [Timothy Davis]() Daniel's concentrate-first view matches how I'd order the levers, and his test fits the new-market point too. Expanding into new languages keeps the money in the channel that already works and opens new auctions there, so the team doesn't have to learn a new platform's buying mechanics from zero. At ElevenLabs, four paid marketers who all speak one language run that across multiple languages, so expansion doesn't require new buyers. I agree with his staffing call as well. Creative fatigue is the ceiling I hit most often, so the next hire for more spend should be someone who makes creative. ##### Question > Daniel says that if Meta still returns well on the marginal dollar, moving budget to TikTok just trades your best ROI for a learning curve, and you should prove Meta's marginal returns have dropped before spreading out. A new channel's early numbers will almost always look worse, so how would you compare the two fairly? What specific evidence would you use to justify funding a new channel's learning period? ##### [Mark Fiske]() I wouldn't compare the two on the same scorecard during the learning period, because the new channel will lose on CPA every time. Run them as two budgets with separate questions. For Meta, the question is whether the marginal dollar still clears payback. The evidence there is a current spend curve, re-fit at the shift points the way Micah described, plus a lift or geo read on what it's actually incremental. If marginal CAC is still well under the payback-adjusted LTV, Meta keeps first claim on new budget, and Daniel is right. For the new channel, the question is whether it's learning toward a viable cost. Before launch I'd agree on three things with finance: 1. A fixed test budget and time window big enough to produce a read. $200 over six weeks on one placement can't. [Citation: https://youtube.com/watch?v=DdqCHnovono] 2. Trajectory targets: CPA falling period over period as the creative formula and buying approach improve, and a clear end date to judge it. 3. Customer quality matched to the core: the early retention proxy and post-purchase survey responses compared against Meta cohorts, so cheap but low-value buyers don't look like a win. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] Funding for the test comes from spend that measurement already shows is past diminishing returns, often parts of Google, and the Meta campaigns that still clear payback stay where they are. [Citation: https://youtube.com/watch?v=DdqCHnovono] The new channel moves to the scale budget only when its marginal CAC, adjusted for incrementality, approaches Meta's marginal CAC at Meta's next spend increment. That threshold is where both channels can finally be compared like for like. The case for running the test before Meta flattens is lead time. By the time the core curve bends, you're months away from a working formula anywhere else. ##### Question > Mark wants a small parallel learning budget on a second channel before the plateau hits, while Daniel says concentrate on Meta until its marginal returns actually fall. At Etsy you ran a Media Mix Model alongside the creator program. How did you decide how much spend went to proving a new channel or format while the main one still cleared payback, and what result told you to scale it? ##### [Mike Jacobs]() I never set the test budget as a fixed percentage. The core channel had to clear payback on the marginal dollar, measured with all three methods together: MMM, experiments, and MTA. Teams that lean on only one of those are often off by 3 to 5x in some channels [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo]. Whatever the model showed as past diminishing returns funded the test. That puts me closer to Mark on timing. I'd also go a step further than Daniel's "prove it first," because the proof he's asking for comes from your own data, and that data is self-fulfilling. You keep optimizing toward the audience you already buy, and the model keeps confirming it. If a new channel or format only gets money once the model approves it, it never gets money. New formats got judged on a different scorecard. With creator content and newer formats at Etsy, we set proxies like consideration or a quality visit before we touched purchase CPA [Citation: https://www.youtube.com/watch?v=5HZfb_n4pP0]. Time on site was our strongest predictor of a purchase 60 to 90 days later [Citation: https://podcasts.apple.com/us/podcast/mike-jacobs-mma-board-member-on-the-challeges-of/id1623994522?i=1000710611085]. So a new format earned more budget when it drove quality visits at a reasonable cost. It earned scale budget when an experiment confirmed incremental lift and the MMM showed its marginal return near the core channel's next increment. Two more levers that raise the ceiling inside the core channel: 1. Feed value into the bid. At Upside we moved optimization from DMA level to ZIP level and passed predicted LTV to the media partners as the conversion value. LTV rose 20 to 30% in the pilot markets. That's the bidding version of Mark's finance-approved proxy. 2. Staff for creative. I'm with Daniel on hiring makers before buyers. Meta's data shows phone-shot creative beats professional shoots about 80% of the time for known brands [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo], so the makers need clear brand positioning more than a studio. 🎯 #### [When paid social costs rise: the ads, the audience or the offer?]() **Question:** When paid social gets more expensive, how do you tell whether the ads, the audience or the offer is the problem? ##### [Timothy Davis]() I’d split the rise in acquisition cost at the point where it happens, then run one-variable tests. 1. Check CPM first, then CTR. If CPM rises across stable audiences and creatives while CTR and post-click conversion hold, the auction has gotten more expensive. If CTR falls for the same audience and placement, rotate a new hook or format against the existing ad. A fresh ad that restores CTR points to creative fatigue. 2. If clicks still come but qualified conversions fall, segment by audience and follow the path through trial, purchase, and revenue. Test a customer-based lookalike against the current audience with the same ad and landing page. I’ve talked publicly about starting with a 1% lookalike as a small-budget sign-of-life test, then widening the range if it works. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] 3. If multiple audiences click and stall at the same step, inspect the landing page, price, and purchase process before changing targeting. At IBM, demand in South Africa existed, but requiring payment in USD hurt conversion. Offering rand addressed that operational gap. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] Keep placement, geography, attribution window, and spend comparable across tests. A rising platform CPA can also reflect a measurement change, so check actual customer and revenue outcomes before shifting budget. ##### [Daniel Pearson]() Timothy’s breakdown gets you to the point of failure. I’d add a second check before calling any test a winner: compare the customers each variant brings in by cohort payback, the time it takes customer revenue to earn back acquisition cost. A new creative or wider audience can improve CTR and first-purchase CPA while bringing in customers who buy once and leave. Conversely, a higher CAC may be fine if that cohort earns it back on a reasonable timeline. I’ve pushed teams to track historical cohorts for this reason, rather than optimize only for install volume or first-purchase CPA. [Citation: https://danielpearson.substack.com/p/payback-periods] For an offer test, track purchase conversion and the value of those customers after the first order. A discount that rescues conversion while weakening payback changes the economics of the offer, even if the ad dashboard looks better. [Citation: https://danielpearson.substack.com/p/increasing-ltv] ##### [Timothy Davis]() Daniel’s cohort payback check is the right guardrail for deciding whether a cheaper CPA actually helped. I’d add an incrementality check before declaring that the offer or audience improved: rerun a lift study when platform-attributed conversions and customer outcomes tell different stories. I’ve seen the measured lift multiplier change when we reran a study, which changed whether the channel deserved more budget or less. [Citation: https://www.youtube.com/watch?v=kv4TRa_J2dE] Hold the offer and creative steady during the audience test so a change in measured lift has a clear interpretation. ##### [Mark Fiske]() There is an operational trap in the diagnosis: the platform’s efficiency target may be disconnected from the company’s growth model. I’ve discussed channel managers finding their programs roughly 20% less efficient than they thought when an attribution model was updated. As spend scales, publisher reporting can miss incrementality and diminishing returns, so a clean-looking ad test may still justify less budget. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] Before deciding the fix is creative, audience, or offer, reconcile platform conversions against actual new customers and revenue at the company level. If they diverge, run a holdout where volume permits and reset the channel target using the incremental result. Otherwise the team can improve CTR or checkout conversion and still allocate the next dollar poorly. ##### Question > Mike, Timothy says that when several audiences click but stall at the same step, he checks the landing page, price, and purchase process. How would you isolate the offer from checkout friction before changing the ads? Please draw on public work and cite a public source. ##### [Mike Jacobs]() I’d locate the drop before changing the ads. Keep the ad, audience, and traffic mix fixed. Break out product-page views, add-to-cart, checkout starts, payment attempts, and completed orders. Then check the same steps for direct and organic visitors. If people want the item and start checkout but fail at shipping, payment, or final total, fix that step and rerun the same traffic. If qualified visitors drop before adding to cart across several traffic sources, test the product promise, price, or incentive on the landing page while leaving checkout unchanged. I’ve argued publicly that a great product, product-market fit, and expectations set in the marketing have to work together. That means checking whether the ad’s claim matches what the product page can actually deliver before labeling weak conversion an ad problem. [Citation: https://podcasts.apple.com/us/podcast/mike-jacobs-mma-board-member-on-the-challeges-of/id1623994522?i=1000710611085] For a considered purchase, I’d also track engaged visitors and their later purchases. In one digital business, our data found time on site predicted purchase 60 or 90 days later. An immediate conversion dip by itself could cause us to change an offer that was bringing in future buyers. [Citation: https://podcasts.apple.com/us/podcast/mike-jacobs-mma-board-member-on-the-challeges-of/id1623994522?i=1000710611085] ##### Question > Micah, Mike says a drop in same-session purchases can hide sales 60 to 90 days later. If paid social costs rise now, how would you decide whether to change the ad, audience, or offer before those sales arrive? Please use only public work and cite a public source. ##### [Micah Moreau]() I’d set a provisional decision rule using mature cohorts before touching the campaign. For comparable past audiences, measure how often an engaged visit becomes a customer within 60 to 90 days, then use that relationship to estimate the current cohort’s likely payback. Keep the estimate separate from observed purchases, because a new offer or audience can change the relationship. Meanwhile, hold the offer and audience fixed to test a fresh ad. Watch CPM, CTR, engaged visits, and checkout starts per thousand impressions. If the fresh ad restores qualified engagement, shift creative. If engagement holds but new spend reaches more expensive people, test the audience at the same offer and creative, and cap spend while the cohort matures. If engaged visitors across audiences consistently stall at the product page, test the offer with checkout unchanged. I’d reserve a holdout where volume allows, since the decision to raise spend ultimately depends on incremental customers rather than platform-attributed ones. I’ve argued publicly for causal return on spend because reported conversions alone can mislead budget decisions. [Citation: https://twitter.com/micahmoreau/status/908740570616561666] #### [Measuring whether paid marketing adds customers]() **Question:** How do you measure whether paid marketing is truly incremental? ##### [Timothy Davis]() I use a lift study to answer the budget question: how many additional customers or dollars did paid create compared with a group that did not see the ads? I then divide the incremental outcome by spend and use that result to decide whether to scale or pull back. We rerun lift studies because the multiplier can move as audience behavior changes. [Citation: https://www.youtube.com/watch?v=kv4TRa_J2dE] Platform attribution is useful for managing campaigns, but clicks alone give an incomplete view of value. I also separate paid and organic traffic so existing demand does not get credited to the campaign. For smaller budgets, test design matters: I’ve suggested roughly $50K a month in platform spend as a point to start the incrementality conversation, since below that you may lack enough signal for a useful test. That threshold depends on the business and the platform. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] ##### [Mark Fiske]() I start with a counterfactual: what would sales and customer value have been without the spend? A matched-market test or holdout gets closest to that answer. Compare the change in total customers and revenue in exposed markets with the change in comparable unexposed markets, then divide the lift by the added spend. Check payback using the value of the customers you actually acquired. The operating problem is that channel teams often manage to publisher-reported efficiency. Facebook can report a profitable return while taking credit for customers who would have arrived anyway. At Ancestry, we refreshed our marketing mix model quarterly; a channel manager could discover their program was 20% less efficient than the platform view suggested, especially after scaling spend 30 to 40%. That should change the next budget decision and the channel target. For smaller brands, I would triangulate tests, mix modeling where feasible, platform data, and carefully designed post-purchase surveys, with experiments doing the heavy lifting on causality. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Timothy Davis]() One practical check between formal lift studies is a controlled pause in a specific paid campaign. Watch total conversions alongside organic conversions during the pause; a drop in attributed paid conversions alone tells you little. I’ve used tactical pauses to isolate a channel’s effect when tracking is imperfect. I’d also check customer value after the initial conversion. In subscription growth, revenue per user can fall as acquisition broadens, so the spread between revenue per user and CAC helps determine whether the extra customers are worth buying. ##### [Mark Fiske]() Timothy's controlled pause is useful if the comparison accounts for what else changed during that window. I would watch total business outcomes and use a comparable market when possible, because seasonality or a concurrent promotion can make a simple before-and-after read misleading. The survey input needs care too. At one portfolio brand, customers described TikTok influencer content even though the team was buying paid ads; the ads used influencer-style creative. We changed the follow-up questions to distinguish the creator's own post from an ad on the brand's account. A survey can help explain where demand came from, but its wording can shift the apparent channel mix. I use that evidence to interpret a lift test, not to replace one. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Daniel Pearson]() The harder call comes after the lift study: how long does the incremental cohort take to repay its acquisition cost? I’d track the cohort’s actual revenue over time alongside the test result, then update the spend decision as that revenue arrives. A week-one lift can look promising while the customers’ long-term value and payback tell a different story. At Bamboo, I’ve argued for payback period as the user acquisition team’s primary KPI because it forces us to revisit historic cohorts and check whether we actually earned the ad investment back. [Citation: https://danielpearson.substack.com/p/payback-periods] For a business with a long payback window, forecasted LTV helps make an interim decision; I’d keep comparing those forecasts with realized cohort revenue rather than treat an early estimate as settled profit. ##### [Timothy Davis]() Agreed on the pause: I would treat an uncontrolled before-and-after result as a diagnostic, then use a comparable market or holdout for a budget decision. I’d also lock down the tracking before reading either test. Consistent source, medium, campaign, and content UTMs let us separate prospecting from retargeting and check whether an apparent shift in paid CAC came from a measurement change. ##### [Mark Fiske]() Daniel's cohort point has an immediate implication for bidding: feed the platforms a signal tied to expected customer value, then revise that signal as actual cohort revenue comes in. I care about payback because it tells us how quickly the business can reinvest in the next customer. A return-on-spend target based on the first order can push spend toward customers who convert cheaply and repay slowly. The incrementality test tells us how many additional customers the channel creates; cohort payback tells us whether acquiring them merits the cash. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Daniel Pearson]() I would prioritize experiments where the budget decision is biggest. Test the highest-spend channels first, then require a holdout before expanding a new channel. That gives the team a measured lift estimate where an attribution error could move the most money. I’ve made the case for using experiments to calibrate a marketing mix model, rather than treating a model’s channel estimates as settled without a holdout. [Citation: https://twitter.com/danielpearson/status/1848399488743834042] ##### [Micah Moreau]() The denominator of marginal CAC should be the additional *active, repeat buyers caused by the spend*, rather than all buyers who showed up while spend rose. Use the holdout or matched-market lift to estimate that denominator. Retargeting and bottom-funnel search are particularly likely to claim buyers who would have purchased anyway. Calculate marginal CAC as the extra spend divided by those lift-adjusted buyers. A Jet offsite example shows why the denominator matters: moving from $2M to $3M in spend took total buyers from 75,000 to 90,000, making marginal CAC $66.67 even though blended CAC was $33. That $66.67 is a spend-curve illustration, not a causal estimate; the experiment tells us how much higher the true marginal CAC is if some of the 15,000 added buyers would have arrived anyway. ##### [Timothy Davis]() For lead gen, I would define the lift outcome at the point where sales can use it, rather than count every form fill equally. At SoftLayer, we cut more than 10,000 low-quality leads per month from the sales team's workload and increased quality leads by 44%. A campaign that lifts raw leads while sending more noise to sales would fail that business test. I’d agree with sales and finance on the qualified-lead or paid-account definition before running the holdout, then follow the cohort through revenue. ##### [Daniel Pearson]() Micah’s marginal CAC math is where I’d be especially careful with forecasted LTV. As spend scales, CAC can rise while newer cohorts monetize worse, so applying the value of an older cohort to the next dollar of spend can overstate returns on both sides. I’ve seen startups treat an LTV forecast as financial truth and discover the gap too late. I’d refresh the forecast against realized revenue from each new cohort before raising the next budget. [Citation: https://twitter.com/danielpearson/status/1848410421708931251] ##### [Mark Fiske]() Micah's marginal CAC framing changes the scaling decision. Run the next test around the proposed increase in spend, not only the existing budget, because incrementality can decline as a channel saturates. I saw that in practice at Ancestry: after a program scaled 30 to 40% in a quarter, the model refresh could show a channel manager roughly 20% less efficiency than expected. The team then needs to reset its operating target to the marginal, lift-adjusted return it can actually earn at the new spend level. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Daniel Pearson]() The cadence of the decision has to match the time it takes customers to pay back. For a first-purchase-profitable DTC brand, first-party financials and one-day-click reporting can be useful between experiments. A business with a long payback period needs a cohort forecast for near-term decisions, then a check against actual revenue as the cohort matures. I’ve used Adobe as an example of why waiting for financial payback before every media adjustment would be too slow. [Citation: https://twitter.com/danielpearson/status/1848405485281783823] ##### [Timothy Davis]() The test only works if we can follow the same customer outcome across systems. At Aha!, I worked with engineering on backend reporting that traced a trial or demo user’s campaign and source through to a paid account. For a B2B holdout, I’d use that paid-account outcome and give conversions enough time to mature before comparing groups. Otherwise a channel with a longer sales cycle can look weak simply because its customers have yet to close. ##### Question > Mike, Micah says the $66.67 marginal CAC in his Jet example is a spend-curve estimate, not a causal result. How would you test whether added buyers came from paid spend when several channels run at once? ##### [Mike Jacobs]() Micah is right: $66.67 describes the spend curve; it does not establish what caused the 15,000 additional buyers. I’d randomize comparable markets into a group that gets the proposed incremental spend and a group that keeps its current spend, while keeping the other channels’ plans the same in both. Compare the change in total new buyers and contribution margin across the groups. That measures the lift from the budget increase even when buyers encounter several channels. If the channels cannot be held stable, test the combined media plan first. Then vary one channel at a time across matched markets, or use a design that varies channels independently, to see what each extra dollar contributes and where channels interact. I would report a range when the test cannot separate their effects. At Etsy, I co-developed media mix modeling with Analytics for a broader cross-channel read; I would calibrate that model against the experiments rather than let each platform claim the same buyer. [Citation: http://mikejacobs.brandyourself.com/] [Citation: https://mikejacobs.com/writing/platforms-grade-their-own-homework] #### [When a marketing mix model earns its cost]() **Question:** When is a marketing mix model worth the effort for a growing company? ##### [Mark Fiske]() I would fund MMM when the next budget decision is large enough that a wrong channel allocation costs more than the model and the work of using it. A growing consumer business moving beyond search and social into TV or other offline channels is a strong candidate, especially once platform reports and company-level results start telling different stories. In a public conversation about measurement, I described our quarterly MMM refresh at Ancestry: a channel manager could discover their program was 20% less efficient than they thought, then need to reset targets as spend scaled. The hard part was getting that finding into the next planning cycle. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] My decision check has three parts: Do we have enough history and meaningful changes in spend across channels to estimate effects? Is there a real choice about where the next dollar goes? Will finance and channel owners use a common view of incremental return and payback to change the budget? If any answer is weak, I would start with matched-market tests for offline media and use post-purchase surveys alongside channel data while building the measurement process. For an enterprise-sales-led company where marketing plays a smaller role in growth, I would prioritize sales effectiveness first. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Micah Moreau]() Before commissioning a model, I’d make every channel owner write down the incremental customers and contribution margin they expect from the next spend increase, plus what a holdout would have to show to change their mind. I’ve argued publicly that marketers need to demand causal, incremental return on spend and question reported results. [Citation: https://twitter.com/micahmoreau/status/908740570616561666] That makes the MMM useful as a way to rank competing budget claims and identify which claim to test next. If the organization will keep funding a channel based on its reported conversions after a credible test disagrees, more modeling effort will not change the decision. ##### [Daniel Pearson]() I’d make the decision based on the size of the allocation decision and how long it takes the business outcome to show up. For a smaller DTC brand with first-purchase profitability, I’d start with first-party financials and one-day-click attribution, then adjust the CAC assumptions as the financials come in. For a business with long payback periods and several meaningful channels, those signals arrive too slowly to guide every budget move. [Citation: https://twitter.com/danielpearson/status/1848405485281783823] An MMM earns its cost when it can change a material budget decision, and the team can check its estimates against experiments. I’d run holdouts on the highest-spend channels first, and require a holdout when expanding into a new channel. Then use those results to calibrate the model and revisit its estimates as spend changes. A fast model with inaccurate inputs can push a growing company toward an expensive mistake. [Citation: https://twitter.com/danielpearson/status/1848399488743834042] ##### [Mark Fiske]() The model's output needs to fit the cash plan. A channel can bring in customers with strong eventual value while leaving a growing company short of cash to fund the next acquisition cycle. In my public piece on marketing budgets, I separate infancy-stage cash constraints from the growth stage, where companies can plan against LTV and choose a payback window. The budget process then ties leadership's revenue and CAC targets to channel-level scenarios and monthly spend. I would require any MMM scenario to show incremental customers, contribution margin and expected payback under each proposed allocation, so finance can make the actual investment decision. [Citation: https://builtin.com/articles/marketing-budget] ##### [Timothy Davis]() The model should have an ongoing check against measured lift, especially as audiences and creative change. For an earlier-stage company, I’d start with clean conversion tracking and a focused lift test where the spend supports it. In a public conversation, I said that even a platform incrementality test can be more work than the signal warrants below roughly $50K a month on that platform. That was a rough guide for that test, not an MMM spend cutoff. I also pointed out that attribution cannot tell you whether someone would have converted without the ad. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] At ElevenLabs, I’ve described rerunning lift studies when performance changes, because the estimated lift can move enough to justify increasing or reducing investment. [Citation: https://www.youtube.com/watch?v=kv4TRa_J2dE] Once the company has enough varied channel spend for an MMM, I’d use those experiments to check its recommendations before shifting a large budget. ##### [Mike Jacobs]() One reason this becomes urgent as channels multiply: the click stops being a reliable proxy for what caused a sale. In one analysis I described publicly, roughly half of the purchases genuinely driven by social had no click. A company allocating spend using click-based attribution would undercount those sales. [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo] I’d use MMM for the recurring cross-channel budget decision, with experiments to calibrate it. In that talk, I described platform and geo holdouts as ways to check the model’s estimates. Each experiment captures one place and time, so the team needs to feed those results back into the model as spend and creative change. [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo] ##### [Mark Fiske]() I would set the decision cadence before buying the model. At Ancestry, the quarterly MMM refresh required resetting channel targets after changes in measured efficiency, which could create real tension with managers who used platform reports every day. In board discussions, I have pushed for attributed reports so budget decisions use the same measurement approach as the channel plans. The operator test is whether finance, the board and channel owners will agree on how model updates reset targets and trigger experiments each quarter. Without that agreement, the model's recommendations will sit apart from the budget. [Citation: https://youtube.com/watch?v=9T_mcVWYra4] ##### [Daniel Pearson]() Mark’s cash-plan point matters especially when the MMM uses forecasted LTV. I’ve seen startups treat LTV forecasts as financial truth while newer cohorts monetized worse and marginal CAC rose with scale. The model can inform the next allocation, but I’d reconcile its projected payback with actual cohort cash returns each month and lower the spending plan if those cohorts miss. [Citation: https://twitter.com/danielpearson/status/1848410421708931251] ##### [Mike Jacobs]() Mark’s finance point changes what I would ask the model to optimize. A model can correctly estimate incremental sales and still recommend the wrong allocation if one channel brings in discount-driven, low-margin orders while another brings in customers with higher expected value. In my public talk, I described passing predicted lifetime value to the media system and looking at margin and average order value alongside conversion rate. A creative can convert more often and produce smaller orders. [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo] Daniel is right to check predicted LTV against realized cohorts. I’d ask the MMM team to show how its channel ranking changes when the outcome moves from gross revenue to contribution margin, then check any added value assigned to new customers against actual payback before moving budget. ##### [Micah Moreau]() I’d put two uncertainty ranges in the budget review: how many incremental customers the spend creates, and how much contribution margin each cohort generates before the cash deadline. I’ve called out the risk of judging CAC and breakeven payback with too little cohort history in a consumer business. [Citation: https://twitter.com/micahmoreau/status/876786817164595200] Multiply the low end of the incremental customer estimate by the low end of realized cohort margin. If that case breaches the cash limit, cap the increase and reserve the rest for a controlled test. Daniel’s monthly reconciliation then tells you whether to release the reserved spend. ##### [Timothy Davis]() One more check before feeding regional conversion data into an MMM: can customers actually buy in those regions? At IBM, we were advertising broadly across Africa while charging in USD. Demand existed, but the currency setup made conversion harder. We narrowed our focus to South Africa and offered payment in rand; performance improved. [Citation: https://www.youtube.com/watch?v=zNJyb3R_Pnc&t=1s] If a company had modeled that earlier period as a regional media-efficiency problem, it might have cut spend in a market where the checkout experience was the constraint. I’d log changes to pricing, payment options, landing pages and availability alongside the media data, then resolve any obvious purchase blockers before using regional MMM estimates to set budgets. ##### [Mark Fiske]() Daniel's cohort check also needs to affect timing within the year. In a budget example I published, the same full-year CAC could produce more in-year revenue by moving spend into periods with faster payback, even after accounting for marginal returns. That is a useful test of MMM output: compare the proposed channel mix and spend schedule against actual cohort receipts, then revise monthly budgets if the projected cash arrives late. A model that recommends the right annual channel shares can still leave money on the table through poor flighting. [Citation: https://builtin.com/articles/marketing-budget] ##### [Mike Jacobs]() There’s a second operational test for MMM: can its findings change what the bidding systems optimize? In my public talk, I warned that a system trained on last-click conversions will find more last clicks, even when those clicks add little incremental value. I described feeding predicted customer value and margin signals into media buying so the system could pursue better customers. [Citation: https://www.youtube.com/watch?v=BT7SNxWLEmo] If an MMM shows a channel has more incremental value than its platform report suggests, I’d change the budget plan and investigate the value signals sent to the platforms. Otherwise the team may correct the quarterly allocation while its daily bidding keeps pursuing the same low-value conversions. ##### [Daniel Pearson]() One failure mode I’d guard against: using the MMM to justify spreading spend across more channels simply because the model now has a line for each one. I’ve argued publicly that strong ROI in Meta or Google is a good reason to keep funding the winner, and that concentration alone is a poor reason to shift dollars away. [Citation: https://danielpearson.substack.com/p/stop-diversifying-your-ad-budget] For each proposed shift, I’d ask for the marginal incremental contribution after the move, its payback period, and the experiment that could prove the shift wrong. ##### Question > Daniel, Mark says MMM needs meaningful changes in channel spend. How would you check whether a growing company's spend history can separate channel effects well enough to guide its next budget decision? Please draw on public work and cite public sources only. ##### [Daniel Pearson]() I’d plot weekly spend by channel against total customers and contribution margin, then mark promotions, price changes, launches and seasonality. The useful question is whether any major channel changed meaningfully while the others stayed relatively steady. If Meta and Google always rose together, their separate effects will be hard to trust, even if the model gives each a precise-looking ROI. I’d then ask the modeler to rerun the next-budget recommendation after reasonable changes to assumptions and after leaving out a period of data. If the recommended channel flips, I’d defer the big reallocation and create cleaner variation through a channel-level or geo holdout. I’ve publicly argued for testing the highest-spend channels first and calibrating MMM against experiments, because an inaccurate model can do more harm than good. [Citation: https://twitter.com/danielpearson/status/1848399488743834042] --- Copyright © 2026 Super Me Network, Inc. All rights reserved. | [Terms of Service](https://www.superme.ai/terms) | [Privacy Policy](https://www.superme.ai/privacy)