Long-form thinking,
written by operators.
We write because we have to think, and we publish because we think it might be useful. Essays as record of how we approach the work — not blog posts as content marketing. No 400-word listicles, no SEO filler. Things you might actually want to read.
Recent essays.
Fourteen questions your agency should be able to answer.
Most agency reporting answers the question the platform is built to answer: did the ads get credit? These fourteen questions ask what the business needs to know instead, grouped into money, truth, and growth, with what a real answer requires and what an evasive one sounds like.
Why Meta, Google, GA4, and Shopify report different revenue.
Meta says one number, Google another, GA4 a third, and Shopify books something smaller than all of them combined. The reasons are mechanical, and once you know them, the gap between the four stops being a problem and starts being a diagnostic.
What a new customer really costs.
Blended ROAS makes every established brand look efficient, because it credits the ads for customers the business already won. Separating new from returning, against your own customer list, is how acquisition cost becomes a number instead of a guess.
Why we left automatic execution switched off.
The ability to carry out changes on ad accounts automatically is built and tested in our platform. It is switched off. Switching it on is a decision about client money, and a system being capable of something is not a reason to let it.
Paid acquisition for subscription apps and digital products.
Subscription businesses inherit an ecommerce playbook that does not fit them. The platforms optimize toward an event on day one, while the money arrives over months. This guide covers the metric order, trial design, the signal to send the platforms, cohort reading, and payback.
Free trial or paid: what your ads should optimize toward.
A no-card trial produces cheap trial starts. A card-required trial produces expensive ones that pay. The ad account cannot tell the difference unless you tell it, and the way you tell it is the event and value you send.
Subscription payback: what you can afford per subscriber.
Payback is the month a subscriber has earned back what they cost. Working it out takes a cost per paying subscriber, a monthly contribution, and a retention curve, and it turns every budget conversation into a question the business can actually answer.
Six attribution models, one set of orders: how to read the disagreement.
Every attribution model is a rule for splitting credit among the touches you observed. Run several on the same orders and the disagreement between them tells you more than any one of them can: which channels open journeys, which close them, and which credit is solid.
First-party data for paid media: what to collect and where to send it.
First-party data is not a strategy deck. For paid media it is a short list of things to capture on your own domain and a short list of places to send them. This guide gives both, in the order to do them, with the consent rules that decide what you may send.
Shopping profit by product: the losers campaign ROAS hides.
A Shopping campaign at 3.7x can contain products earning well and products losing money on every sale, and the average hides which is which. Breakeven ROAS differs by product because margin does. This is how to find the losers and restructure around margin.
TikTok ads for ecommerce: when the demographic actually fits.
TikTok can mint demand that search never saw. It can also burn a catalog that sells to a buyer who is not in that audience. This is the fit test, the measurement honesty, and the cases where you should not run it at all.
Google Ads keyword economics for SaaS.
SaaS teams still rank keywords by cost per click and cost per lead. Those are input prices, not the cost of a customer. This is the arithmetic that prices a query by pipeline: close rate, ACV, payback, and the volume you should refuse.
LinkedIn ads: the honest defense of $18 CPCs.
LinkedIn will charge you more per click than any other major paid channel. Sometimes that is the correct price for reaching the buyer who can sign. Sometimes it is a vanity audience with no pipeline behind it. Here is how to tell, without pretending the CPC is cheap.
How to scale an ecommerce business: the revenue ladder.
Most scaling advice is a list of tactics with no diagnosis attached. This guide is the diagnosis: the five rungs from $250k to $5M a month, the constraint that binds on each, the arithmetic that decides how a store doubles, and a test for which rung you are really on.
Scaling ecommerce ads past $100k a month: what changes.
At $100k a month in spend, the game changes: marginal CAC replaces average CAC, single channels saturate, creative becomes a production problem, and measurement becomes infrastructure.
The high-spend monthly review, done properly.
A monthly review of a six-figure ad budget should not be a dashboard tour. It should interrogate the blended scoreboard, the marginal dollar, measurement integrity, structural hygiene, and next month’s allocation — in that order. The agenda we actually run.
Are Bing ads worth it in 2026? A data-grounded answer.
Microsoft Ads in 2026 is a more serious platform than the one advertisers dismissed years ago — PMax with real transparency, native experiments, Copilot inventory. Whether it earns your budget depends on math, not sentiment.
Audience exclusions and retargeting in a consolidated Meta era.
In a consolidated, automation-first Meta account, targeting is largely the algorithm’s job — but exclusions are the control that decides whether you pay to acquire new customers or to re-reach ones you already had. How to structure exclusions and retargeting that is genuinely incremental.
Meta account structure in 2026: consolidation, Advantage+, and control.
For years, sophisticated Meta accounts meant elaborate segmentation: an ad set per audience, per placement, per stage. That playbook now actively hurts. The modern skill is consolidation: fewer campaigns, more signal, and a clear framework for when to trust automation.
Performance Max examples: what good actually looks like.
What does a well-run PMax setup actually look like? Three worked examples — high-SKU ecommerce, considered purchase, lead gen with offline import — with the reasoning made explicit.
Meta reporting beyond Ads Manager.
Meta’s in-platform ROAS is generous by design — attribution windows, view-through conversions, and blended prospecting all flatter it. How to build a reporting layer on a blended metric that reconciles against real revenue instead of taking Ads Manager at its word.
Google AI Max for Search: should you turn it on?
Google AI Max for Search promises broader reach by letting AI find new queries and adapt your ads. Sometimes that is real incremental growth. Sometimes it is just a more expensive route to traffic you already had. Here is how we decide, and how we test it without risking the account.
The measurement stack for high-spend advertisers.
At scale, measurement is not a dashboard — it is a layered system: clean signal into the platforms, a blended metric anchoring reality, and incrementality tests settling the questions attribution cannot. What the stack contains and why each layer earns its place.
How we scaled a global book retailer 300% without breaking ROAS.
Awesome Books came to us wanting to grow fast without sacrificing profitability, across an inventory of millions of titles and a global customer base. We grew sales 300% in six months while holding their ROAS and CPA targets. Here is how the engagement was built.
Freelancer, expert, or agency: who should actually run your ads?
"Google Ads expert" and "Google Ads agency" are usually typed by the same person on the same afternoon, and they lead to very different places. Here is what each option is genuinely good at, the spend range where each stops making sense, and the failure mode specific to each.
Scaling Meta creative production without burning out.
A high-spend Meta account consumes creative faster than any individual can produce it. The production system that keeps pace: concept pipelines, modular assembly, a disciplined testing cadence, and a method for turning winners into the next round of tests.
Creative testing: you cannot afford significance.
Every creative testing guide tells you to let tests run until the results are significant. Do the arithmetic and you discover that significance costs roughly four hundred conversions per variant, more than many accounts produce in a month across everything. So the honest system is not a better experiment. It is ranking, floors and retirement rules that make good decisions while individual verdicts stay noisy.
Facebook Ads cost benchmarks in 2026 — and why they mislead.
Realistic 2026 ranges for Meta CPM, CPC, and CPA — presented with the caveats that make them usable. The real lesson is methodological: a benchmark is a distribution, not a target, and your own economics decide whether your cost is good.
Performance dropped: a systematic method for finding out why.
Performance drops trigger a predictable panic: bids get changed, campaigns get restructured, budgets get shuffled, all before anyone has established what actually happened. We work a fixed diagnostic order instead: measurement, market, platform mechanics, then the account. In that order, for a reason.
Server-side tracking and the Conversions API: a practical 2026 setup.
Server-side tracking is no longer optional. The browser pixel misses too much, and the gap is widening every year. This is a practical look at how Conversions API and Enhanced Conversions work, what a trustworthy setup looks like, and where implementations go wrong.
Meta bidding: cost cap, bid cap, and when to use which.
Meta’s bid strategies are widely misused. What highest volume, cost per result goal, ROAS goal, and bid cap each tell the auction, how to set a starting goal from unit economics, and the adjustment rules Meta itself publishes.
SaaS marketing metrics that matter.
Median B2B SaaS CAC payback now sits around 15-16 months — a different world from the 12-month rule of thumb. The metrics that matter, current benchmark ranges, and how to derive paid media targets from them.
Questions to ask before hiring an ads agency.
A good question has a property almost nobody checks for: a weak agency cannot answer it well by being articulate. That single test disqualifies most of the questions in circulation — "what is your process," "can you show me case studies" — and it produces a much shorter list where each question comes with the answer that should reassure you and the answer that should end the meeting.
The first 90 days: how we approach a foundation rebuild.
The interesting work in paid media, the bid strategy and creative testing and audience expansion, only compounds once the foundations are right. So our first 90 days on any account are deliberately spent on the unglamorous work first. Here is the sequence.
Managing Google Ads at $100k a month and beyond.
The tactics that grow a $20k account are not what governs a $200k one. At scale the job shifts to marginal-return allocation, incrementality, and structural discipline — and the failure modes get more expensive. What management actually looks like at six figures a month.
Incrementality testing: work out what your test can detect first.
Every attribution model is an allocation of credit among ads; none can tell you what would have happened without the ad. Only a controlled test can, but an underpowered test is worse than no test, because its nulls all point the same direction: toward "nothing is incremental". Here are the three designs a marketing team can run, and the arithmetic that tells you in advance which questions your data is capable of answering.
Advantage+ Shopping campaigns in 2026: an operator’s guide.
Advantage+ Shopping now lives inside the Advantage+ sales campaign. The settings that decide whether Advantage+ is on, the new way to limit spend on existing customers, a worked split of the return by customer type, and a decision rule for the budget.
How to scale Meta ads in 2026.
Meta scaling in 2026 has three levers left: consolidated Advantage+ structure done properly, a creative pipeline that refills faster than fatigue drains it, and conversion signal the ranking models can trust.
Why most Google Ads audits miss the real issue.
Most accounts we audit have great-looking dashboards and broken foundations. The interesting issues are never bidding strategy or keyword choice. They are upstream of those: in measurement, attribution, and architecture decisions made years ago and never revisited.
The Facebook Ads audit checklist for high-spend accounts.
Meta has fewer knobs to misconfigure and more ways to starve the delivery system of what it needs. This is the audit with numbers attached: the conversion arithmetic that sets a hard ceiling on how many ad sets you can support, the reason a reconciliation that balances can be two opposite errors canceling out, the thresholds that make creative fatigue a finding rather than a bad week, and why fixing signal raises the return on every dollar rather than redirecting a slice of them.
Lead gen bidding when not every lead is equal.
Cost per lead is the most seductive metric in lead generation and the most dangerous. Optimize to it and the algorithm learns to find people who fill in forms, not people who buy. The way out is closing the loop between the CRM and the ad platform.
PPC management pricing models, compared.
A percentage-of-spend fee does not just "create an incentive to spend more" — at 12% on a $10,000 budget increase, it pays the agency $1,200 to make a recommendation that can cost you $2,800. That ratio is calculable for every model, and once you have it, the choice stops being philosophical. The four models with their conflicts sized in dollars, plus the specific contract term that neutralizes each.
Attribution after iOS 14: what actually works.
Pixel attribution is not coming back, and the difficulty has moved on from missing data to something subtler: platforms now fill the gaps with modeled estimates displayed alongside observed conversions. Platform claim, observed correlation, modeled contribution, measured incrementality. Four numbers, four meanings, and most budget mistakes come from confusing them. Here is the stack that separates them.
How much should you spend on ads?
There is a minimum spend below which paid media cannot learn, and a maximum beyond which marginal returns go negative. How to find both numbers for your business, and the cadence for moving between them.
Creative volume is the new targeting.
For ten years, performance marketing was won on targeting. iOS 14 weakened that game, and Meta has since rebuilt ad delivery so the system picks the audience from the creative. The edge now is a steady supply of genuinely different concepts, and you can calculate how many your spend needs.
How much does a Facebook Ads agency cost in 2026?
The creative line in a Meta quote is not a percentage of anything. It is a fixed monthly cost set by how many net-new concepts the account has to produce to keep a winner in rotation, and that number can be derived rather than guessed. Once you can derive it, the whole shape of Meta agency pricing becomes obvious: why sub-$15k budgets rarely support a full engagement, why the same fee is 20% of spend at one budget and 3% at another, and what improvement any fee has to buy.
Profit on ad spend: bidding to margin, not revenue.
A 4x ROAS on a 15% margin product loses money. A 2.5x ROAS on a 70% margin product prints it. Revenue-based bidding cannot tell the difference, which means most accounts are systematically over-buying their worst economics. The fix is to bid to margin.
MER vs ROAS: the argument is a decoy.
Every version of this article tells you to stop trusting ROAS and start trusting MER. That advice is not wrong so much as beside the point: MER and blended ROAS are the same economics inverted, and neither is safer than the revenue figure you feed it. Here is the ladder that ranks your revenue sources by trustworthiness, the arithmetic that turns one week of data into three contradictory decisions, and how to set the floor from your own P&L.
How much does a Google Ads agency cost in 2026?
A fee quoted as 12% of spend can consume 20% of the profit your paid channel produces, or 60%, depending entirely on your contribution margin — and the same fee may only need a 7.5% performance improvement to pay for itself. Both numbers come from three lines of arithmetic almost nobody runs, and they answer different questions: whether the agency can plausibly earn its fee, and whether it is worth paying even if they do. The market fee models are here too, but the arithmetic is the part that decides.
In-house vs agency: the honest comparison.
In-house teams win on context and focus; agencies win on cross-account pattern exposure and infrastructure. The honest cost math, where each model breaks, and the hybrid most scaling brands converge on.
Smart Bidding in practice: tROAS vs tCPA and when each wins.
Target ROAS or target CPA is one of the most common questions in Google Ads, and most answers are dogmatic. The honest answer depends on your business model and, far more importantly, on whether your conversion data is clean enough for Smart Bidding to work at all.
Landing pages: where paid media is actually won or lost.
Every efficiency gain inside the ad account is capped and local. A conversion rate improvement on the landing page compounds across every channel, every campaign, and every future click at once. Most teams still optimize in the wrong order.
Brand vs non-brand: the error that grows as you succeed.
Counting branded search as paid acquisition is the most common measurement error in paid search, and the usual account of it (that it inflates ROAS) understates the problem twice over. It makes every other figure in the account uninterpretable, and it is self-concealing: the error grows precisely as your non-paid marketing succeeds. Here is the full decomposition, why separation always leaks, and the cases where the split is honestly ambiguous.
The Google Ads audit checklist for high-spend accounts.
Most audits fail not because the checks are wrong but because nothing tells you when an observation becomes a finding — so the output is a list of things that could be improved, which is every account ever built. This version attaches a threshold to every check, states the significance floor underneath it (including the Poisson arithmetic that explains why a term spending one target CPA with no conversions is a coin flip rather than a problem), and ends with how to rank what you found by dollars you can actually recover.
You are buying pipeline, not leads.
Optimizing lead gen to CPL is how you fill a CRM with people who will never buy anything. The feedback loop, bidding setup, and reporting structure that aim spend at pipeline instead.
The paid-media account standard, v1.0.
Every "healthy account checklist" is a list of things no competent operator would dispute, which is exactly why accounts fail them: agreement is free and thresholds are not. This is the same discipline rebuilt as a real standard: 22 numbered criteria each with a pass condition you can fail, weighted scoring that gates on the layers which corrupt everything downstream, an explicit list of what we left out and why, and a version number so an account scored today stays comparable next year.
Google Shopping feeds: the optimization layer most retailers ignore.
In Shopping and Performance Max there are no keywords. The product feed is the targeting. Which auctions you enter, what you pay, and what converts are all decided by data most retailers export once from their store and never look at again.
Performance Max in 2026: how to actually control it.
Performance Max is the default for most Google Ads accounts now, and most advertisers run it on faith because Google frames it as a black box. It is not a black box. It is a budget router across six surfaces that reports a single average, and the job is not to control the algorithm, it is to make that average decompose.
How to scale Google Ads without breaking it.
Before any sequencing question, one input predicts whether an account can absorb more money: the direction efficiency is already moving on its biggest spenders. If that slope is flat or negative, more budget buys less efficiency and no amount of careful staging fixes it. This is the readiness test we run first — four gates with thresholds, the marginal-versus-average arithmetic that explains why average ROAS keeps saying "you have room" through two consecutive value-destroying steps, and the difference between headroom that exists in a report and headroom the account can actually reach.
Dynamic keyword insertion without embarrassing yourself.
Dynamic keyword insertion is either a relevance tool or a nonsense generator, depending entirely on the discipline around it. The mechanics, the failure modes, and where it still earns a place.
Ecommerce unit economics: four numbers, one quantity.
Most scaling mistakes are made in the spreadsheet long before they reach the ad account. The reason is structural: CAC, LTV, payback window, and MER are treated as four independent KPIs when they are four views of one quantity, the contribution margin a customer produces. Here is the full waterfall, the denominator error that makes CAC look half its real size, a payback rule for each rung of the revenue ladder, and why a business with textbook ratios can still run out of cash while scaling.
How to choose a performance marketing agency.
There is a spend level above which one in-house hire is simply cheaper than an agency, and it is calculable rather than a matter of opinion: fully-loaded senior salary against a percentage-of-spend fee gives a crossover around $100,000 a month — with three specific reasons the naive crossover is too low. That is decision one. This is the full sequence: whether to hire, what shape, how to select, and how to structure the first ninety days so the arrangement can fail cleanly instead of drifting for a year.
Demand Gen campaigns: when to use them (and when not to).
Demand Gen bundles YouTube, Shorts, Discover, and Gmail into one visually-led campaign. Where it fits between Search capture and PMax, and the setups where it earns its budget.
Google Shopping vs Performance Max: which, and when.
Standard Shopping gives you control and transparency; PMax gives you reach and automation. When each wins, when to run both, and the migration mistakes that cost real money.
The SaaS paid acquisition playbook.
The ratio everyone optimizes — LTV:CAC — is a solvency measure, and companies do not die of insolvency. They die of illiquidity, which is what payback period measures. This is the playbook with that constraint put first: how payback sets your maximum self-funding growth rate, why a change to billing terms can double the budget you can afford while a campaign optimization moves it a few percent, and why in SaaS a month-end review is not just correlational but measuring the wrong quarter.
PMax asset groups: how to structure them properly.
Asset groups are where PMax strategy actually lives. How to theme them, how many to run, when to split campaigns instead, and the creative quality bar that decides your reach.
Google Ads audience segments that do real work.
First-party lists, custom segments built from converting searches, and observation-mode data you actually read: the audience toolkit that changes outcomes, not just settings.
YouTube ads for performance, not branding.
The advertisers winning on YouTube treat it as direct response: conversion-focused formats, intent-built audiences, creative engineered for the skip button, and measurement that respects lag.
How Performance Max actually works under the hood.
PMax is a black box, but not a mystery. How asset combination, surface arbitration, and conversion-value bidding actually interact — and where your inputs steer the machine.
Google Ads metrics that lie to you.
CTR rewards clickbait, conversion rate punishes growth, impression share hides its own denominator, and platform ROAS grades its own homework. What each metric really measures and what to read instead.
PMax brand exclusions: setup, step by step.
Unexcluded brand traffic is the most common source of inflated PMax performance. The exact setup — brand lists, negatives, verification — and what to expect when the crutch is removed.
Microsoft Ads: the second search engine your competitors ignore.
Most advertisers treat Microsoft Ads as an afterthought, which is exactly why it works. The audience economics, who should bother, and how to run it in two hours a week.
Google Shopping ads: the complete operator's guide.
One target ROAS across a mixed-margin catalog is not a simplification, it is two errors running at once: the high-margin products are throttled well above their breakeven while the low-margin products lose money on every sale the algorithm is being rewarded for buying. This guide works the channel in the order that matters — eligible inventory, then feed as targeting, then margin-band targets with the arithmetic laid out, with the thresholds that separate a real leak from ordinary noise.
What is Performance Max? A plain-English guide.
Performance Max is the most misunderstood campaign type in Google Ads. This is the explanation we give clients: what it is, where ads show, what you control, and when to run it.
Written by operators, not content teams.
One essay when we find something worth your time, from people running real accounts. How we research and correct this writing is set out in our editorial policy.
no tracking · no spam · sent when we find something
Want to see this thinking on a real account?
Browse case studies