Reeto M.

The problem is not simply budget or audience targeting but the quality of signals sent back to ad platforms, making both platform-reported and blended new-customer CAC unreliable.

• Reeto's preferred North Star is blended new-customer CAC rather than platform attribution.
• The proposed change is to replace a conventional advanced CAPI provider with predictive CAPI.

Reeto describes an ecommerce brand where more than half of daily orders came from returning customers.

• The brand synced customer and CRM lists to Meta hourly and excluded those audiences from prospecting, yet platform-reported and blended new-customer CAC kept increasing.
• It spent seven figures per month across Meta at about 70% share, Google Ads, and affiliate, keeping Google and affiliate spend fixed within about 5% during the experiment.
• On 2/19 it switched to Angler's predictive CAPI for Meta.
• Reeto reports that blended new-customer CAC dropped even as Meta spend rose, and marginal new-customer CAC became more efficient.

Read original

A two-cell A/B test can compare predictive CAPI with the brand's existing setup, while Meta's incremental attribution reporting measures incrementality within that same test. Reeto flags a statistical caveat: treat the readout carefully, not as a simple, clean result.

The incumbent setups Reeto names include Elevar, Blotout, Triple Whale Sonar, Shopify, Segment, and homegrown systems.

• In their reported observations, campaigns optimizing on 7-day click showed incremental attribution improvement exceeding the lift in standard 7-day click attribution; some view-through lift driven by predictive CAPI appeared incremental; and not all 7-day click-reported conversions were incremental.
• Reeto also points out that campaigns can now optimize on incremental attribution, letting the auction adjust based on estimated incrementality.

Read original

That creates two server-side connections and one browser pixel.

• Meta's deduplication depends on matching event IDs between browser and server, so two CAPI sources firing with mismatched IDs cannot be reliably deduplicated.
• The result can be artificially favorable in-platform cost per order, CAC, and ROAS, while ad optimization weakens because platform gains do not match storefront-level results.
• Reeto adds that when Meta receives two similar server-side events, it often discards the second, making the extra signals redundant rather than additive.

Reeto gives a diagnostic sequence: open Events Manager, select the dataset where Sonar is active, open the Purchase event, and compare browser and server event counts.

• If server-side events are roughly double browser events, two server-side connections are likely firing.
• Then check the Event Deduplication tab for overlap between browser and server events.
• Reeto says overlap should ideally be 75% or higher, and preferably 90% or higher in well-set-up systems.

Read original

Returns cluster in larger baskets and often post two to four weeks late, so the algorithm has already learned from a bad example before the brand notices. Drawing on years of building predictive models before founding Angler, Reeto proposes predicting returns at high resolution at the moment of purchase, then sending expected net revenue as Value and expected net margin as Profit at order time.

Reeto contrasts a $1,190 new-customer order returned in full, netting zero revenue, with a $642 order the customer kept entirely, which produced the highest net revenue of the group. The auction's value modeling almost ignored the second order while learning to find more customers like the first.

Read original

Even for low-stakes purchases, consumers wanted to feel they had made a choice; for anything emotional, brand-driven, or expensive, the consideration process is part of the purchase, not friction to remove.

• The opportunity for agentic AI in commerce is compression: helping a buyer decide faster rather than deciding for them.
• Reeto calls this an unpopular opinion and a personal read, and invites builders with a different view to say what they think is being missed.

Reeto points to Amazon's $5 Dash button, designed solely for reordering toilet paper, and notes that it failed.

• Even for something as mindless as laundry detergent, consumers wanted to feel they made a choice.
• The conclusion is a design principle: help me decide faster, do not decide for me.

Reeto presents this as an open question for the field, not a settled finding, setting it apart from their more confident measurement arguments.

Read original

Reeto announced that Kim Yarnell joined Angler AI as Chief Revenue Officer, bringing more than twenty years in growth roles across marketing, data, and technology, including a decade at Meta working with large retailers and marketers, and several months as an Angler advisor. The stated mandate covers business development, customer success, and marketing.

Reeto describes the CRO mandate as scaling revenue efficiently, deepening customer impact, and translating product innovation into durable, long-term growth as Angler enters a new expansion phase. This provides context for how the measurement arguments are being brought to market, not evidence of measurement practice itself.

Read original

Reeto describes a two-cell A/B test comparing Angler's predictive CAPI with a brand's existing setup, whether that's Elevar, Blotout, Triple Whale Sonar, Shopify, Segment, or a homegrown system.

• Meta's new incremental attribution reporting lets the team measure incrementality within the test.
• Early observations are specific: when campaigns optimize on 7-day click, the improvement in incremental attribution outperforms the lift shown in standard 7-day click attribution; some of the view-through lift from predictive CAPI appears incremental; and not all conversions reported by 7-day click attribution are incremental.
• The post also says campaigns can now optimize on incremental attribution, letting the auction adjust based on estimated incrementality.

For brands planning a campaign

A DTC brand testing a new CAPI setup should use a controlled A/B test and incrementality reporting—not platform-reported lift alone—to determine whether the new signal source is adding customers.

Reeto flags a pattern among ecommerce brands using Triple Whale Sonar on top of Shopify pixel and Shopify CAPI: the setup creates two server-side connections and one browser pixel.

• Meta deduplication relies on clean, matching event IDs between browser and server, so mismatched IDs from two CAPI sources can prevent reliable deduplication.
• The post says this can inflate in-platform cost per order, CAC, and ROAS while weakening ad optimization, because platform improvements don't match storefront-level metrics.
• Reeto gives exact checks: open Events Manager, select the dataset where Triple Whale Sonar is active, and check the Purchase event.
• Compare browser and server event counts; if server-side events are about 2x browser events, two server-side connections are likely firing.
• Then check the Event Deduplication tab for browser/server overlap: ideally 75% or higher, and preferably 90% or higher in well-set-up systems.

For brands planning a campaign

Before trusting Meta ROAS or CAC, a DTC brand should audit Purchase event deduplication and check for duplicate server-side connections. Fixing the stack can improve reporting accuracy and optimization quality.

Reeto argues that performance marketing is more complicated than showing ads to the right prospects and converting them, particularly when measurement and attribution are involved.

• The post describes an ecommerce brand where over 50% of daily orders came from returning customers.
• The brand wanted to dedicate 100% of paid spend to acquiring new customers.
• It synced customer and CRM lists to Meta hourly and excluded those audiences from prospecting, yet platform-reported new-customer CAC and blended new-customer CAC kept increasing.
• The brand spent seven figures per month across Meta at 70% share, Google Ads, and affiliate.
• During the experiment, Google and affiliate spend stayed fixed within about 5%.
• On 2/19, the brand switched from an advanced CAPI provider to Angler's predictive CAPI for Meta.
• The North Star was blended new-customer CAC.
• Early results showed blended new-customer CAC dropping despite higher Meta spend, while marginal new-customer CAC became more efficient.
• The next stated frontier is LTV:CAC optimization using pLTV signals to acquire higher-value customers.

For brands planning a campaign

If new-customer CAC rises despite clean audience exclusions, a brand should test predictive CAPI while tracking blended and marginal new-customer CAC. This example shows why platform-reported CAC alone can mislead acquisition decisions.

Reeto gives two contrasting orders.

• A $1,190 new-customer order was returned in full, leaving $0 in net revenue, but the auction still learned to find more customers like that one.
• A $642 new-customer order was kept entirely and had the highest net revenue of the group, yet the auction's value modeling almost ignored it.
• The post says most retailers and DTC brands send gross revenue as the Value signal, so Meta and Google optimize toward what people buy, not what they keep.
• Returns cluster in big baskets and often post two to four weeks late, so the algorithm trains on the wrong customer before the brand notices.
• Reeto says returns are predictable with high resolution at the moment of purchase and recommends sending expected net revenue as Value and expected net margin as Profit at order time.

For brands planning a campaign

For DTC brands with meaningful returns, value-based bidding should use expected net revenue and expected net margin, not gross revenue, so the platform optimizes toward customers who keep their orders. This changes the value signal and performance readout.

Reeto points to Amazon's $5 button, whose only job was to reorder toilet paper, and notes that it failed.

• The argument is that agentic commerce may have a narrower path than the hype suggests.
• Even for something as mindless as laundry detergent, consumers wanted to feel they made a choice.
• For anything emotional, brand-driven, or expensive, the consideration process isn't friction; it's the point.
• Reeto thinks the real win for agentic AI in commerce is compression, not automation: help me decide faster, don't decide for me.

For brands planning a campaign

A DTC brand should use agentic AI to shorten decision time and support choice, not remove the customer from the purchase decision—especially in higher-consideration categories. This affects where AI belongs in the funnel.

In a detailed account of one seven-figure-per-month brand spending across Meta (about 70% of paid budget), Google Ads, and affiliate, the brand had already synced customer and CRM lists to Meta hourly and excluded those audiences from prospecting, yet platform-reported new-customer CAC and blended new-customer CAC kept climbing.

• Reeto describes swapping an advanced CAPI provider for Angler's predictive CAPI on 19 February and tracking blended new-customer CAC as the north star rather than platform-reported attribution.
• The stated early result was falling blended new-customer CAC even as Meta spend rose, with marginal new-customer CAC also improving.

Held non-Meta channels within roughly +/-5% during the test to maintain a stable baseline, and chose an audience-level outcome (blended nCAC) that does not depend on any single platform's attribution claims.

For brands planning a campaign

For a brand whose platform numbers and storefront numbers disagree, the clearest entry point is a fixed-budget, single-variable CAPI swap measured against blended new-customer CAC. Ask whether the brand already has clean CRM-to-platform list syncing; Reeto treats it as a prerequisite.

Examples: Source 1

Common incumbents include Elevar, Blotout, Triple Whale Sonar, Shopify, and CDPs such as Segment or a homegrown setup.

• Reeto notes that platform conversion lift studies are an alternative but take more time, budget, and setup.
• The post argues that Meta's incremental attribution reporting lets brands measure incrementality within the existing two-cell test.
• It reports three observations: campaigns optimizing on 7-day click showed incremental attribution improvement exceeding the lift in standard 7-day click attribution; some of the view-through lift driven by predictive CAPI appears incremental; and not all 7-day click-reported conversions are incremental.
• Reeto also points out that campaigns can now optimize against incremental attribution, letting the auction adjust on estimated incrementality, and says raw data comparing against Elevar was shared separately.

Two-cell A/B test: predictive CAPI against the incumbent setup, measured through Meta's incremental attribution reporting rather than a platform lift study. Reeto flags the statistical caveat directly.

For brands planning a campaign

A brand with an incumbent CAPI or CDP can run this head-to-head rather than rebuild. The key is to choose the decision metric in advance: 7-day click, view-through, and incremental attribution give different answers about the same test.

Examples: Source 1

That setup produces two server-side connections and one browser pixel.

• Because Meta's deduplication depends on matching event IDs between browser and server, two CAPI sources firing with mismatched IDs cannot be reliably deduplicated.
• The stated consequences are artificially favorable in-platform cost per order, CAC, and ROAS, alongside weaker ad optimization because in-platform gains do not show up at storefront level.
• Reeto adds that when Meta receives two similar server-side events, it often discards the second, making the extra signals redundant rather than additive.

A concrete diagnostic sequence: open Events Manager, select the dataset where Sonar is active, open the Purchase event, compare browser and server event counts (server-side counts roughly double browser counts when two server-side connections are firing), then check the Event Deduplication tab. Reeto gives an overlap target of 75% or higher, preferably 90% or higher in well-configured systems.

For brands planning a campaign

This is the cheapest fix in the material and the first to check. A brand seeing platform ROAS that does not reconcile with storefront revenue should verify event counts and deduplication overlap before investing in new measurement infrastructure or blaming Meta's modeling.

Examples: Source 1

The illustration: a $1,190 new-customer order where the customer returned everything, netting zero, versus a $642 order that was fully kept and produced the highest net revenue of the group — yet the auction's value modeling almost ignored the second order. Because returns cluster in larger baskets and often post two to four weeks late, the algorithm has already trained on the wrong customer before the discrepancy becomes visible.

Drawing on years spent building predictive models before founding Angler, Reeto proposes predicting returns with high resolution at the moment of purchase, then sending expected net revenue as the Value signal and expected net margin as the Profit signal at order time. That keeps value optimization from cutting both ways.

For brands planning a campaign

A brand with meaningful return rates in large baskets should treat its Value and Profit signals as a data-quality project, not a bidding setting.

• The testable claim is that returns can be predicted at purchase time with enough resolution to feed the auction.
• Check whether the brand's order-time data supports that, or whether returns only become known weeks later.

Examples: Source 1

The argument: even for something as low-stakes as laundry detergent, consumers wanted to feel they had made a choice.

• For anything emotional, brand-driven, or expensive, the consideration process is the point, not friction to remove.
• Reeto frames the opportunity for agentic AI in commerce as compression rather than automation — helping a buyer decide faster rather than deciding for them — and invites builders with a different view to say what is being missed.

Reeto calls this an unpopular opinion and a personal read, framing it as a question to the field rather than a settled conclusion. Treat it as a point of view, not a finding.

For brands planning a campaign

Useful as a positioning input rather than a measurement tactic. A brand evaluating agentic shopping tools or AI-assisted product discovery can take away a testable design principle: assist and shorten the decision, but be wary of interfaces that remove the choice entirely, particularly in higher-consideration categories.

Examples: Source 1

Yarnell brings more than twenty years in growth roles across marketing, data, and technology, including a decade at Meta working with large retailers and marketers, and several months as an advisor to Angler before the appointment. The stated mandate covers business development, customer success, and marketing, with a focus on scaling revenue efficiently, deepening customer impact, and translating product innovation into durable growth.

For brands planning a campaign

Background on how the measurement arguments are being brought to market: the go-to-market function sits with someone whose experience is on the platform side of retail advertising, which shapes how incrementality and attribution claims are likely to be framed in a sales conversation.

Examples: Source 1

They say they spent years building predictive models before founding Angler.

• Posts cover onboarding brands to Angler's predictive CAPI, running A/B tests against incumbents, auditing client event setups, and using expected net revenue as a value signal.
• A separate post announces Kim Yarnell joining Angler AI as Chief Revenue Officer and describes the company's growth and her role, not Reeto's.

View example

Reeto describes switching a brand from another advanced CAPI provider to Angler's predictive CAPI for Meta, testing Meta's incremental attribution reporting against a two-cell A/B test, and auditing how brands running Triple Whale Sonar alongside Shopify's CAPI create duplicate server-side events that Meta cannot deduplicate. These posts describe platform mechanics and measured results, not a sponsorship or formal partnership with Meta.

View example

Both lay out an ordered audit sequence and warn that fixing duplicate events will not resolve an attribution gap, or vice versa.

Bivas focuses on tracking and measurement plumbing, especially EU consent states, Consent Mode and payment-provider flows, writing as an independent consultant. Reeto goes further into what the auction optimizes for, using predictive CAPI and incrementality tests from inside a measurement vendor.

Read original · Original 2

Samuel's pre-holiday audit list covers CAPI deduplication, new versus returning customer signals and offline order reconciliation, overlapping directly with Reeto's deduplication and value-signal posts.

Samuel frames the work around agency operations and data infrastructure, including self-hosted automation pipelines and per-task cost traps, for an audience of agencies running client volume. Reeto focuses on auction-level signals, incrementality testing and value modeling.

Read original · Original 2

Both argue that revenue-based metrics mislead, that margin is the measure that matters, and that where the value signal lives is the deciding question.

Abhi builds the argument from conference panels and retail leaders' stated positions, stressing the organizational gap between CRM and analytics teams. Reeto draws on experience building predictive models and describes a product-level fix for order-time value signals.

Read original · Original 2

The two posts walk through the same buyer task from the audit side: reconciling Meta Ads purchase counts against Shopify, separating a tracking discrepancy from an attribution discrepancy, checking whether the Meta Pixel and Conversions API are double-counting a purchase through mismatched event IDs, and confirming deduplication in Events Manager.

• Bivas explicitly serves EU businesses and agencies and names Consent Mode V2, GDPR, PayPal and Klarna redirects, and consent-state measurement as part of the audit.
• A brand whose numbers are wrong in the first place, especially one selling into the EU where consent handling changes what data is available, gets a more directly useful starting point here than from a value-signal or incrementality argument.
• The stated method is a sequence — tracking, then deduplication, then data quality, then consent, then attribution — giving a brand a checklist to run before spending on new measurement software.

The tradeoff is that this material stops at clean measurement; it does not show how to use the corrected signal.

• Where Reeto's posts connect the fix to a business decision — blended new-customer CAC falling as Meta spend rose, or sending expected net revenue as the Value signal so the auction stops chasing orders that later net to zero — Bivas's posts end at an accurate purchase count.
• A brand that trusts its tracking but wants the algorithm to bid toward higher-value customers would need a second engagement.
• The posts offer no campaign-level benchmarks or test designs; ask for a controlled experiment with a named decision metric if that is what the brand needs.

Examples: Source 1 · Source 2

Both posts address agencies rather than a single brand.

• One frames pre-holiday tracking audits as margin protection, describing how Ads Manager can report a strong ROAS, the media buyer can scale budget, and the brand can later find it barely broke even.
• It then lists three layers to audit: CAPI deduplication with matching event IDs, new-versus-returning signals passed through the server-side payload, and offline reconciliation of settled orders against ad platform click data.
• The other post addresses agency operations, explaining why pay-per-task automation platforms become a variable cost as client volume grows and describing a self-hosted alternative.
• A brand buying through an agency, or an agency seeking a service definition to sell, gets a more operationally shaped answer here.

The tradeoff is that the posts provide less detail on measurement mechanics.

• The peak-season post gives three audit headings but no worked numbers, diagnostic steps inside Events Manager, or example of an actual discrepancy traced to a cause — the kind of detail Bivas's posts supply.
• The agency-operations post is about infrastructure cost, not ad measurement, so a brand evaluating tracking audits should weigh only the first post.
• The posts also do not discuss incrementality testing or value-based bidding; a brand focused on what the platform optimizes toward, rather than whether the pipeline is clean, is better served elsewhere.

Examples: Source 1 · Source 2

The posts make that the whole argument.

• One distinguishes three kinds of customer value — backward-looking historical LTV used for reporting, predictive LTV used for segmentation, and predictive LTV pushed into the platform's bidding stream through CAPI so the auction acquires high-value customers from the start.
• It states plainly that a score sitting in a dashboard or an audience segment decorates the bidding system, while a score in the conversion value field re-aims it.
• The other post reports from a CommerceNext panel where retail leaders defined customer value as margin rather than revenue and described activation, not data, as the blocker.
• A brand looking to move from clean numbers to auction-level optimization, and comfortable with the vendor framing, finds a sharper argument here than in the audit-oriented alternatives.

The tradeoff is that this evidence argues a position rather than showing a repair.

• It offers no diagnostic sequence or event-level check, and says nothing about duplicate purchases or consent handling.
• A brand with a broken CAPI setup should fix that before optimizing on a signal it cannot trust.
• The panel post is secondhand commentary on other people's remarks, not the author's tested work; the LTV post cites an improvement figure without showing the underlying data or the test that produced it.
• Before changing bidding logic, ask for a measurement audit and the data and test behind the LTV figure.

Examples: Source 1 · Source 2