Oliver Yuwen

Attribution shows correlation and the path a customer took, they say, but not incrementality — what would not have happened without that channel.

• It also does not model budget reallocation with statistical rigor or produce the confidence intervals a CFO or board member trusts on seven-figure decisions.
• Yuwen's answer is holdout groups and causal modeling, not better tracking pixels.
• A companion post splits the work: multi-touch attribution for short-term tactical calls like ad copy and landing pages, marketing mix modeling for channel mix and long-range budget planning, and incrementality testing to calibrate both.
• A third walks through seven attribution models — last-click, first-click, linear, time-decay, position-based, GA4 data-driven and custom machine learning — and names how each can distort a budget decision.

A DTC ecommerce client's CMO cut spend on a search channel that had looked like a steady revenue driver.

• Sales did not move.
• A holdout test showed negligible lift: customers who would have searched and bought anyway still did.
• Yuwen says consolidating budget into channels with real incremental lift delivered a 20x return on the analytics investment in two months.

Read original · Original 2 · Original 3 · Original 4

ROAS sees attributed revenue only, missing return rates, fulfillment costs and the fact that scaling paid spend reaches lower-intent buyers who purchase once on a discount and never come back.

• The fix: measure contribution margin by channel — revenue minus COGS, fulfillment, returns and ad spend — alongside LTV by cohort.
• A channel with expensive clicks may be buying the best customers, while the best-ROAS channel buys the worst.
• Yuwen makes the same case to finance audiences: forecasts that ask a CFO to trust a long payback fail budget reviews, while a model tying CAC to contribution profit by cohort, SKU and promo cycle gets approved.

A 90-day LTV pull by acquisition channel showed that two campaigns with nearly identical ROAS had cohorts worth roughly half of each other — one had been quietly compressing margin. In another post, Yuwen describes a channel showing 4x ROAS on the dashboard that turned negative once fulfillment cost, return rate and the discount stack were included: it had been buying revenue, not profit.

Read original · Original 2 · Original 3

Dashboard and CDP vendors may hold customer data on their own servers: cancel and the history may not export, get acquired and the data follows the acquirer, face a price increase and you leave with nothing.

• Most brands also fail to configure their sources properly — broken UTMs, email taking credit for sales from paid search, GA4 missing a chunk of Shopify orders because a pixel misfires — and end up optimizing inside a system they do not own with numbers that were already wrong.
• Yuwen recommends a cloud data warehouse (BigQuery, Snowflake, Redshift) as permanent, queryable ground, with pipelines, transformation logic and a deduplication method the executive team has explicitly agreed on.

A $14M outdoor apparel brand's Klaviyo dashboard showed $1.8M in email-attributed revenue for a quarter, while Shopify order data showed $1.1M in email-originated orders. Pulling orders, ad spend, email events and session data into a warehouse and building a deduplication model with last-touch as the baseline brought email to about $803K and surfaced a paid channel with a 4.2x blended return once attribution overlap was removed.

Read original · Original 2 · Original 3

A "we miss you" email sent after a customer stops buying wastes lifecycle spend: by the time someone lands on a churned list, they have mentally moved on.

• Instead, predictive decay signals — a sudden drop in session frequency, a shift in support-ticket sentiment — trigger a rescue sequence while the relationship is still cooling.
• Yuwen makes the logic actionable: find the inactivity window that precedes cancellation, then trigger a value-first email (a tip or product highlight, not a discount) at a fraction of that threshold.
• A separate post applies the same reasoning to repeat purchases: work a channel or product with a predictable buying cycle before the cycle ends, not after.

In a dialogue sketch between a retention lead and a COO, ten days of inactivity during month 4 implies an 85% probability that the subscriber will not renew in month 6.

• Month 4 is the deadline; the intervention is a value-first email at 50% of the inactivity threshold.
• Elsewhere, Yuwen describes pulling a 60-day repeat-purchase window forward to day 52 — unremarkable on its own, but an annualized revenue gain without touching acquisition cost.

Read original · Original 2

When a brand's leadership defends a high-frequency, one-size-fits-all email program, they run an incrementality test against a control half of the list and report what changes rather than argue best practice. The post closes by asking readers how personalized their own CRM program is and inviting disagreement — consistent with Yuwen's approach to claims: measure before asserting.

A fast-growing ecommerce brand sent five emails a week to its entire list; open rates were declining, unsubscribes creeping up and email-attributed revenue plateauing.

• Half the list kept the usual blast, while the other half received behavioral triggers, past-purchase history and churn-risk scoring at one to two emails a week.
• Yuwen reports a 21% lift in email revenue, a 38% increase in click-to-open rate and a 22% drop in unsubscribes, reaching statistical significance within two weeks.

Read original

For most businesses, simple segmentation still wins — recency, frequency, monetary value — run monthly and fed into paid, CRM and offline campaigns to identify who is about to churn, who is primed for upsell, and who does and does not need a discount.

• They call RFM "old reliable" and say most teams do not need flashy AI models.
• On customer value, Yuwen answers a product VP's objection about needing three different LTVs by assigning each horizon to a decision: a 6-month LTV for CAC payback and cash flow, a 1-year LTV for retention and channel profitability once lifecycle and CRM work takes effect, and a 2-year LTV for product strategy.
• Little lift beyond one year points to a churn or retention problem.

Yuwen illustrates the LTV question with a VP of Product asking why the company needs three versions of the same metric.

• Their answer: a single lens leaves money and insight on the table.
• If 2-year LTV shows no lift over 1-year LTV, the business most likely has a retention problem rather than a product-strategy one.

The corpus has a clear tension: the same writer who tells teams they do not need flashy models also recommends incrementality testing, holdout groups and custom machine-learning attribution. The consistent thread is that simple segmentation is the floor a team should stand on before reaching for anything more elaborate, not a substitute for testing whether a channel or message actually changed a decision.

Read original · Original 2

Lead attribution shows where a lead came from; revenue attribution shows whether it was worth anything.

• Most teams solve the first and ignore the second.
• Yuwen's lessons: model simplicity beats sophistication nobody trusts; sales data has to be in the room because marketing touchpoints alone do not explain revenue; marketing, sales and RevOps must agree on definitions before anything is built; and attribution is directional rather than definitive — useful for budget decisions, not for winning arguments about credit.

Yuwen describes a recurring scene in data teams: a polished lead dashboard tracking opens, clicks, MQLs and SQLs, but no clear answer when the CFO asks what the spend returned.

Read original

This detailed example shows the problem and one way to rebuild the numbers.

Oliver Yuwen describes a brand whose Klaviyo dashboard showed $1.8M in email-attributed revenue for the quarter, while Shopify order data showed $1.1M in email-originated orders.

• The gap came from each platform claiming credit under its own rules: Klaviyo counted purchases within five days of an open, including sales driven by a Meta ad that same week, while Meta claimed view-through attribution.
• The team pulled Shopify orders, ad spend, email events, and session data into a warehouse and built a deduplication model using last-touch as the baseline.
• That reduced email attribution to about $803K.
• Some paid channels that had looked less attractive showed a 4.2x blended return once the overlap was removed.

For brands planning a campaign

For a brief about attribution or marketing measurement, this is a concrete case of platform dashboards overstating channel value—and an agreed deduplication model changing budget choices.

This is a clear starting point.

Oliver Yuwen contrasts multi-touch attribution tools such as Northbeam, Rockerbox, and Triple Whale with the question executives eventually ask: “If I move $50k from Meta to Google, what happens to revenue?”

• Attribution shows touchpoints and the path a customer took, but not incrementality—what would not have happened without that channel.
• It also does not model budget reallocation scenarios with statistical rigor or provide the confidence intervals a CFO or board member trusts for seven-figure decisions.
• Attribution tools reduce arguments; they do not eliminate uncertainty.

For brands planning a campaign

Useful for a campaign about measurement maturity, with an angle on incrementality testing, holdout groups, and causal modeling—not just better tracking pixels.

This piece explains the mismatch and what to measure instead.

Oliver Yuwen argues that ROAS only captures attributed revenue.

• It misses return rates, fulfillment costs, and the fact that scaling paid media can reach lower-intent audiences who buy once on a discount and do not return.
• When that happens, CAC creeps up and margins fall even if ROAS barely moves.
• The recommended fix is to measure contribution margin by channel—revenue minus COGS, fulfillment, returns, and ad spend—then layer in LTV by cohort.
• In one example, two campaigns had nearly identical ROAS, but 90-day LTV by acquisition channel showed one cohort was worth roughly half the other.

For brands planning a campaign

For a B2B SaaS or ecommerce retention brief, this supports a campaign about moving from ROAS to contribution margin and cohort LTV as the basis for profitable growth.

This post outlines a specific trigger model and the reasoning behind it.

Oliver Yuwen argues that sending a “We Miss You” email after a customer has stopped buying wastes lifecycle resources: the customer has mentally moved on.

• Automated retention interventions use predictive signals, such as a sudden drop in session frequency or a shift in support ticket sentiment, to trigger a rescue sequence before disengagement becomes cancellation.
• In a dialogue example, a 10-day inactivity gap during month 4 predicts an 85% probability that a subscriber will not renew in month 6.
• The recommendation is to identify the engagement cliff and trigger a value-first email at 50% of that inactivity threshold.

For brands planning a campaign

For a retention analytics brief, this is a practical example of using predictive churn triggers and value-first interventions instead of waiting to send a discount.

This piece shows how one test answered that question and changed where the money went.

A DTC ecommerce client’s CMO cut spend on a search channel that had looked like a consistent revenue driver.

• Sales did not move.
• Oliver Yuwen ran an incrementality test: hold out a segment, cut spend, and measure the delta.
• The lift turned out to be negligible because customers who would have searched and bought anyway were still searching and buying.
• The team recalibrated its investment, found other channels with no incremental lift, and consolidated budget into channels that actually moved the needle.
• The savings delivered a 20x return on the analytics investment in the first two months.

For brands planning a campaign

Good for a campaign on incrementality testing and budget reallocation, especially when the goal is to show how analytics work pays for itself.

This post answers with a specific model and a before-and-after business outcome.

Oliver Yuwen argues that forecasts saying “trust us, this will LTV out” fail budget reviews when they take three quarters to prove.

• The alternative is a model that ties CAC to contribution profit by cohort, SKU, and promo cycle.
• In one case, a SaaS CMO needed to justify a budget increase after a failed Q2 scale.
• The team built a three-variable model that broke down CAC by intent segment, adjusted for trial conversion variance, and mapped payback in weeks instead of months.
• Forecast accuracy rose from 70% to 85%, and the CFO approved incremental spend.

For brands planning a campaign

For a B2B SaaS brief, this supports a campaign angle around CFO-ready measurement, CAC payback, and contribution profit rather than brand-level marketing claims.

This original shows a controlled test that changed the answer.

A fast-growing ecommerce brand’s CMO said the team sent five emails a week to everyone.

• Open rates were declining, unsubscribe rates were creeping up, and email-attributed revenue was plateauing.
• Yuwen ran an incrementality test: half the list got the usual blast; the other half got fewer, more relevant emails based on behavioral triggers, past purchase history, and churn risk scoring, reducing frequency from five emails a week to one or two.
• The test produced a 21% lift in revenue from email, a 38% increase in click-to-open rate, and a 22% reduction in unsubscribes, reaching statistical significance within two weeks.

For brands planning a campaign

For a retention or CRM brief, this is a useful example of testing frequency and personalization instead of assuming more email means more revenue.

In one post, they name Northbeam, Rockerbox and Triple Whale as genuinely useful for seeing the customer journey, then argue that these tools answer the wrong executive question: "If I move $50k from Meta to Google, what happens to revenue?"

• They say the tools show correlation and touchpoints but not incrementality, cannot model budget reallocation with statistical rigor, and do not produce the confidence intervals a CFO trusts on seven-figure decisions.
• The answer, they argue, is holdout groups and causal modeling.
• Another post walks through seven attribution models (last-click, first-click, linear, time-decay, position-based, GA4 data-driven, custom machine learning) and how each distorts budget decisions, ending with a warning: if your channel reports add up to more than your revenue, the attribution conversation is overdue.
• A third pairs multi-touch attribution with marketing mix modeling, assigning MTA to short-term tactical choices like ad copy and landing pages and MMM to channel mix and long-range budget planning, with incrementality testing used to calibrate both.
• The sharpest illustration is a client whose "top-performing" search channel showed negligible lift in a holdout test; consolidating spend into channels that actually moved the needle reportedly returned 20x on the analytics investment in two months.

Opens with the budget question an executive would actually ask, then shows why the popular tool category cannot answer it.

For brands planning a campaign

A strong fit for campaigns selling measurement, incrementality testing, MMM or analytics services to growth and marketing-ops teams. The content reframes the buyer's question from "which channel gets credit" to "what would have happened without it" — useful if your product's differentiator is testing rather than reporting.

Examples: Source 1 · Source 2 · Source 3 · Source 4

ROAS sees attributed revenue only; it misses return rates, fulfillment costs, and the fact that scaling paid reaches lower-intent buyers who purchase once on a discount and never return.

• Their prescribed fix is contribution margin by channel — revenue minus COGS, fulfillment, returns and ad spend — layered with LTV by cohort, because a channel with expensive clicks may be buying the best customers while the best-ROAS channel buys the worst.
• One post describes two campaigns with nearly identical ROAS where a 90-day LTV pull showed one cohort worth roughly half the other.
• Another says most clients are surprised to find a channel they believed was profitable was actually losing money once fulfillment, return rate and the discount stack were added in — "buying revenue but not profit."
• A third ties the same logic to budget approval, describing a CAC-to-contribution-profit model broken out by cohort, SKU and promo cycle that moved a SaaS forecast accuracy claim from 70% to 85%.

Contrasts two metrics that look similar and shows a client case where they diverge.

The client results, including the 20x and 70%-to-85% figures, are self-reported from Yuwen's own engagements and are not independently verified in Yuwen's posts.

For brands planning a campaign

Good fit for ecommerce profitability, finance-adjacent analytics and CFO-facing budget content. If your product computes margin or payback rather than channel credit, Yuwen's framing gives you the argument and vocabulary.

Examples: Source 1 · Source 2 · Source 3

One post describes dashboard and CDP vendors holding customer data on their own servers: cancel, and the history may not export; get acquired, and the data follows the acquirer; face a price increase, and you leave with nothing.

• Before that risk even bites, Yuwen says most brands never configured their sources properly — broken UTMs, email taking credit for sales that came from paid search, GA4 missing 15–20% of Shopify orders because a pixel is misfiring — so they optimize inside a system they don't own using numbers that were already wrong.
• The recommendation is a cloud data warehouse (BigQuery, Snowflake, Redshift) as the permanent, queryable foundation.
• A detailed example follows a $14M outdoor apparel brand where Klaviyo reported $1.8M in email revenue for a quarter while Shopify order data showed $1.1M in email-originated orders; a warehouse deduplication model with last-touch as the baseline and assisted touches layered on brought email to about $803K and surfaced a paid channel with a 4.2x blended return once overlap was removed.

Uses a concrete platform-versus-order-data discrepancy and the reconciliation model as proof.

For brands planning a campaign

Well suited to data-warehouse, CDP and measurement-infrastructure campaigns. The pitch angle Yuwen's posts support is data ownership plus clean pipelines as a prerequisite for analytics and AI, not dashboards as the endpoint.

Examples: Source 1 · Source 2 · Source 3

One post argues that a "we miss you" email sent after a customer stops buying is wasted lifecycle spend, because by the time someone reaches a churned list they have mentally moved on.

• The alternative is predictive decay signals — a sudden drop in session frequency or a shift in support-ticket sentiment — triggering a rescue sequence while the relationship is still cooling.
• A dialogue sketch makes the logic concrete: 10 days of inactivity during month 4 implies an 85% probability of non-renewal in month 6, making month 4 the deadline and a value-first email (a tip or product highlight, not a discount) the intervention, triggered at 50% of the inactivity threshold.
• A separate post applies the same reasoning to repeat purchasing: identify the product customers buy on repeat and the median gap before they do, then match lifecycle touchpoints to that window — if customers reorder at 60 days, nurture them from day 30.
• Pulling a 60-day repeat window to day 52 sounds minor but annualizes into revenue without touching acquisition cost.

Turns a general principle into a numeric trigger a reader can implement in a CRM.

For brands planning a campaign

Useful for lifecycle, CRM and retention-analytics campaigns where timing or trigger design is the differentiator. Yuwen's posts supply copy-ready thresholds — an inactivity window, a percentage-of-threshold trigger, a repeat-purchase gap — that a brand can adapt to its own funnel.

Examples: Source 1 · Source 2

The observed symptoms were declining open rates, creeping unsubscribes and plateauing email-attributed revenue.

• Rather than argue about best practice, Yuwen describes an incrementality test: half the list received the usual blast, the other half received behavioral triggers, past-purchase history and churn-risk scoring at one to two emails a week.
• The reported outcome was a 21% lift in email revenue, a 38% increase in click-to-open rate and a 22% drop in unsubscribes, reaching statistical significance within two weeks.
• The post closes by asking how personalized the reader's CRM program is and inviting disagreement.

The lift figures are self-reported and the brand is unnamed. Ask for test methodology and sample sizes.

For brands planning a campaign

Relevant to email service providers, CRM and personalization vendors. Yuwen frames personalization as something to prove with a split test rather than a design preference — a brand selling automation should expect that standard.

Examples: Source 1

One post states that for most businesses the simplest segmentation still wins — recency, frequency, monetary value — run monthly and fed into paid, CRM and offline campaigns to identify who is about to churn, who is primed for upsell, and who does and does not need a discount.

• They describe RFM as "old reliable" and explicitly say most teams do not need flashy AI models.
• On customer value, a second post answers a product VP's objection about needing three different LTVs: a 6-month LTV supports CAC payback and cash flow, a 1-year LTV shows retention and channel profitability once lifecycle and CRM work takes effect, and a 2-year LTV informs product strategy — with little lift beyond one year pointing to a churn or retention problem.

Assigns each metric to the specific decision it serves rather than presenting it as a standalone KPI.

For brands planning a campaign

Suitable for segmentation, CDP and customer-analytics campaigns aimed at operators rather than data scientists. Yuwen's framing favors fundamentals first, so a product positioned as an AI upgrade may need a different messenger.

Examples: Source 1 · Source 2

Their distinction: lead attribution tells you where a lead came from; revenue attribution tells you whether it was worth anything.

• Most teams solve the first while ignoring the second.
• Their lessons: model simplicity beats sophistication nobody trusts; sales data must be included because marketing touchpoints alone do not explain revenue; marketing, sales and RevOps must agree on definitions before anything is built; and attribution is directional rather than definitive—useful for budget decisions, not for winning arguments about credit.

For brands planning a campaign

The corpus’s most directly B2B-relevant topic, and a natural fit for demand-generation, revenue-operations and B2B attribution campaigns. The focus is pipeline-to-revenue reporting and cross-team definitions, not ad-platform mechanics.

Examples: Source 1

The posts don’t say whether Yuwen owns the business, works for it, or partners with it. Ask Yuwen to clarify the relationship before engaging.

View example

The posts don’t state a partnership or commercial tie.

View example

The posts don’t state a commercial relationship.

View example

The posts don’t state a partnership.

View example

The posts don’t state a commercial relationship.

View example

Lagace walks through an 11-step bottom-up forecast that converts a business-level aMER target into channel- and SKU-level targets, holds prospecting on every channel to the same financial outcome, and sets targets by first-purchase contribution margin because delivery costs differ by product.

• That matches Yuwen's approach of replacing ROAS with contribution margin by channel and layering in LTV by cohort.
• Both also treat returning-customer behaviour as the stable part of the model and new-customer acquisition as the volatile part.

Lagace publishes a detailed forecasting system for named brands, laid out as a step-by-step build readers can copy into a model.

• It covers forecast construction end to end.
• Yuwen's posts are shorter critiques of measurement practices, usually anchored in one client discrepancy or executive question.
• They focus more on attribution tooling, data ownership and lifecycle timing than forecast mechanics.

Read original · Original 2

Both use blended new-customer CAC or similar deduplicated measures as the North Star instead of platform attribution, describe A/B or holdout tests to establish incrementality, and question view-through credit and attribution stacks that reduce arguments without removing uncertainty.

Reeto writes from inside a predictive CAPI vendor, so the posts focus on testing and proving that product against incumbents such as Elevar, Blotout or Triple Whale, and on what Meta's incremental attribution reporting shows. Yuwen writes as a consultant about the wider stack, including data warehouses, contribution margin and lifecycle timing, without centring on one signal product.

Read original · Original 2

Kiel's example—a customer sees a Meta ad, searches the brand on Google and buys through Google Shopping, so Google gets the conversion and Meta gets nothing—illustrates the same distortion Yuwen describes. Both point readers toward blended metrics and proper incrementality testing rather than more attribution modelling.

Kiel focuses on creative production and ad-account structure for small DTC brands: how many conceptually distinct ads a founder can ship, what the weekly launch routine should be, and why creative capacity becomes the bottleneck before budget.

• Attribution is one topic among many.
• Yuwen writes almost entirely about measurement, data ownership and retention timing, with little attention to creative output or account operations.

Read original · Original 2

The two posts answer the same buyer question from the infrastructure side: a two-cell A/B test of predictive CAPI against an incumbent setup (Elevar, Blotout, Triple Whale Sonar, Shopify, Segment); blended new-customer CAC as the North Star instead of attribution; and early reads on Meta's incremental attribution reporting, including the observation that not every 7-day click conversion is incremental and part of view-through lift is. If the series has to end with a reader connecting their CAPI setup to a measurement product, this creator is closer to that ending than Yuwen: their posts stop at the argument for testing and hand the reader a walkthrough booking.

The material argues from inside a measurement vendor's onboarding experience, so a co-authored series would sit closer to product positioning than neutral teaching.

• It also offers less of the model-by-model explanation that makes Oliver Yuwen's posts useful to a general measurement audience.
• Before booking, ask for audience composition, engagement figures and commercial terms; only two posts appear here, and they provide none of them.

Examples: Source 1 · Source 2

A marketing-analytics platform whose buyers are finance-adjacent needs a series on turning a P&L into ad-account targets.

• The two posts lay out an 11-step bottom-up forecast system attributed to named brand work (M&Ms, Purdy & Figg, WildBird), covering aMER targets, first-time versus returning AOV, recently acquired repeat rate, a delayed attribution multiplier that bridges a 28-day business target to a 7-day click target, and running every SKU to the same first-purchase contribution margin instead of one blended ROAS.
• That is a more operational answer to "what should we spend and against what target" than Yuwen's posts offer: they answer "did the channel cause the sale."

Forecasting and planning are the subject, not incrementality testing: the holdout question appears as a step inside the model rather than the thing being taught, so a measurement platform would sponsor content about a neighbouring job. The system belongs to an agency and is explicitly described as one it charges thousands for, so the series carries a service pitch alongside the teaching.

Examples: Source 1 · Source 2

One post works through the Meta-to-Google last-click problem — a customer sees a Meta ad, searches the brand, buys through Google Shopping, and Meta gets nothing — and gives a tracking fix using custom conversions for Google-referral, Klaviyo-referral, and branded Google Shopping traffic.

• It states plainly that blended metrics matter more than attribution modeling for judging whether marketing works.
• The other post covers a repeatable ad-production routine for a founder running a $30k/month account.
• If the platform also sells to brands spending in that range and wants the series to end in a concrete tracking or creative change, this creator's material is more immediately actionable than Yuwen's.

Creative production and media buying are the centre of the material; holdout testing and analytics instrumentation are not.

• A measurement vendor's product would therefore sit off-topic in most of a series built with this creator.
• The attribution post is one of the two available and reaches a smaller-spend audience than the enterprise measurement buyer Yuwen writes for.

Examples: Source 1 · Source 2