Abhi Agnihotri

Historical LTV looks back at past transactions and informs reporting, forecasting, and budget planning.

• Predictive LTV estimates future value shortly after a first interaction and informs segmentation, audience building, and retention prioritization.
• The third level, pLTV activation, sends calibrated predicted values into the ad platform’s bidding stream through server-side APIs such as Meta CAPI or Google OCI, so the algorithm can acquire high-value customers from the start.
• Most conversations stall between the second and third levels: a score in a spreadsheet, dashboard, or even an audience segment tells the platform whom to prefer, but the ranking model still optimizes for its own conversion signal.
• Activation changes what the platform optimizes for.

The post offers a diagnostic for the common claim, “we already do LTV”: ask where the score lives.

• A report is the first category, an audience segment the second, and a conversion-value field the bidder trains on the third.
• The third is credited with up to 60% contribution margin improvement.

A recurring frustration runs through the writing: brands that say they already do LTV usually mean historical LTV, treating reporting as activation.

Read original

Google is described as exceptional at predicting conversion rates because it sees real-time browser, device, and query signals a brand never will.

• But over three, six, or twelve months, it is described as poor at predicting value because it cannot see CRM records, profit margins, or retention rates, so it defaults to the campaign’s historical average order value.
• Replacing that default with a predicted LTV score keeps the auction mechanics identical but changes the optimization target.
• Asked in demos why Google or Meta do not build this themselves, they answer that it is partly a choice—trillion-dollar companies pursue much larger markets—and mainly a design issue, because the value-side data sits with brands.

A side-by-side image shows what the algorithm learns with default settings and what it learns when pLTV signals feed the conversion-value field.

Read original · Original 2

The cited figures show a 61–72% loss of mobile conversion visibility for brands running Pixel-only on iOS after ATT.

• The auction kept running and CPMs kept rising as the algorithm optimized on increasingly incomplete data.
• In this account, brands that rewrote creative or restructured ad sets missed the point.
• Those that adapted replaced the missing signals with first-party data and treated signal engineering as a growth lever.

The post names four replacement signals: hashed customer identifiers through CAPI, subscription enrollment events, replenishment-cycle data, and predicted LTV scores calculated from internal CRM data.

Read original

Many high-value customers never reach platform audiences, and offline conversions connect poorly to attribution. AdZeta’s identity and match layer is described as always-on infrastructure that builds one identity per customer across CRM, ecommerce, web, and offline data; hashes and normalizes IDs to each platform’s specifications; and packs audiences and conversions with multiple identifiers so better conversion and pLTV signals flow back.

The post contrasts a one-off list upload with an always-on identity layer, claiming consistently above 90% match rates for the latter. Its closing point: having a lot of data is not a moat; knowing how to activate it is.

Read original

The role requires hands-on work with at least one of Google Ads OCI, Meta CAPI, TikTok Events API, or server-side conversion tracking. It is pitched as execution-first and AI-native, with direct collaboration with a founding machine learning engineer.

The post names three layers and describes the activation layer as a must-have, not a nice-to-have.

Read original

The gap between how leaders measure success internally and how ad platforms optimize externally is described as massive.

• A panelist from U BEAUTY argued that the path to a second purchase matters most, since third and later purchases become much easier, and called discounting in luxury skincare a drug and a race to the bottom.
• Sabrina Cherubini of FULLBEAUTY described CLV as a conversation that lives with CRM and analytics teams but rarely reaches the core of how a company operates.
• Adrian Rohr of Fabletics, working with three million identified members, said the constraint is activation, not data, and described AI moving brands from personalization toward what he called personal relevance.
• Agnihotri concludes that the data and measurement exist, but most brands still lack the activation layer connecting customer value to the systems that spend money.

The sharpest illustration comes from beauty and wellness: CLV insights sit in a dashboard, inform a quarterly board slide, and never reach the ad auction, bidding logic, or acquisition strategy in real time.

Read original

At a Klaviyo Appy Hour during Shopify DotDev week in Toronto, conversations focused on technology partners creating more value for shared customers and on growth and retention when the data layer works end to end. The conference-season post invites readers to connect in person and bring strong opinions about what is broken in performance marketing.

The post lists ChargeX East on September 17 in New York, RetailClub at Huntington Beach on September 22–24, Shoptalk in Nashville on September 29–October 1, ADWEEK in New York on October 5–8, plus private dinners and after-hours events.

Read original · Original 2

The post defines three types of LTV: historical (past transactions), predictive (forward-looking estimates) and pLTV activation.

• It argues that performance improves when predictive values feed directly into the ad platform's bidding stream via CAPI/OCI, rather than sitting in a dashboard segment.
• This shifts the platform's optimization goal.

For brands planning a campaign

For a DTC growth team, this roadmap traces the move from basic LTV reporting (Category 1) to the technical implementation needed for a high-level acquisition strategy (Category 3).

Abhi Agnihotri outlines Google's ad bid calculation: P(Conversion), E[Value | Conversion] and Target ROAS.

• They explain that while Google is effective at predicting who will convert (P(Conversion)), it cannot see internal CRM data or long-term retention, so it defaults to a simple Average Order Value.
• Growth teams can replace that assumption with custom predictive LTV scores.

For brands planning a campaign

A performance marketing agency can use this approach to show clients how high-fidelity long-term value signals can give them a bidding advantage on Google Ads, which currently lacks access to those signals.

Many brands upload basic CRM lists to ad platforms, producing low match rates and failing to connect offline conversions and LTV. Abhi Agnihotri details an always-on identity and match layer that normalizes and hashes customer IDs to Meta and Google specifications, packing multiple identifiers to maximize conversion data and pLTV signals.

For brands planning a campaign

The post identifies identity matching as an infrastructure requirement for signal engineering and effective signal activation.

At CommerceNext, the discussion highlighted a gap between available data and measurement and the activation layer that connects customer value metrics like CLV to spending systems. Adrian Rohr said AI is moving brands from general personalization to 'personal relevance,' which requires connecting internal value data to real-time ad execution.

For brands planning a campaign

The post frames this work as a strategic business shift, not just a technical integration, and a priority for modern DTC growth.

Abhi Agnihotri describes AdZeta's full stack for turning predictive LTV scores into real-time value signals. It includes ingesting first-party data into BigQuery, establishing multi-tenant GCP infrastructure and developing Python packages to dispatch pLTV signals via Ads APIs (Meta CAPI, Google Ads OCI).

For brands planning a campaign

The post outlines the technologies and processes needed to build a signal engineering solution for Google and Meta.

The first two describe or segment customers; the third changes what the auction optimizes toward before acquisition. They say this drives up to 60% contribution margin improvements.

Diagnostic question: where does the score live — in a report, an audience segment, or the conversion value field?

For brands planning a campaign

For a DTC growth team that already tracks LTV, the first diagnostic is where the score lives.

• If it sits in a dashboard or audience segment, the brand has reporting or segmentation, not activation.
• A campaign plan should route predicted value into the conversion value field the bidder trains on.

Examples: Source 1

Google is described as very strong on conversion probability because it sees real-time browser and device signals, but weak on expected value because it cannot see a brand’s CRM, margins, or long-term retention.

• By default, the value variable is the campaign’s historical average order value.
• Replacing it with a predicted 3-, 6-, or 12-month LTV score is presented as a structural bidding advantage.

For brands planning a campaign

Evaluate a Google program on whether it feeds predicted long-horizon value into the auction, not only on target ROAS. Specify which internal value signals replace the default AOV assumption and how often they are refreshed.

Examples: Source 1 · Source 2

The answer is partly choice and mainly design: the platforms pursue very large markets, while value-side data — CRM records, margins, retention — sits with brands. They locate the arbitrage for growth teams in targeting high-value cohorts on day one rather than after a long conversion window.

For brands planning a campaign

Agency or vendor positioning can frame pLTV activation as filling a structural gap in platform visibility. For a brand brief, separate what the platform already optimizes well from the value data the brand must supply.

Examples: Source 1

They cite 61–72% mobile conversion visibility loss for Pixel-only brands after ATT.

• In their account, brands that adapted replaced those missing signals with first-party data: hashed customer identifiers through CAPI, subscription enrollment events, replenishment cycle data, and predicted LTV scores from CRM data.
• Meta’s Andromeda ranking engine is described as learning from signal quality rather than targeting audiences the old way.

For brands planning a campaign

Assess a Meta program by which conversion and value signals it sends to the platform, not only by creative or ad-set restructuring. Treat signal engineering as a growth lever alongside media buying.

Examples: Source 1

AdZeta’s identity and match layer is always-on infrastructure that unifies customer identity across CRM, ecommerce, web, and offline data, normalizes and hashes IDs the way platforms expect, and packs audiences and conversions with multiple identifiers. AdZeta claims consistently above 90% match rates and says better conversions and pLTV signals feed back into platform bidding and attribution.

For brands planning a campaign

Before investing in pLTV activation, test the identity layer. Match-rate quality determines whether high-value customers reach the platform’s bidding system.

Examples: Source 1

The role requires hands-on work with Google Ads OCI, Meta CAPI, TikTok Events API, or server-side conversion tracking. Abhi Agnihotri frames the ideal hire as execution-first and AI-native, working directly with a founding ML engineer.

For brands planning a campaign

A signal-engineering roadmap needs an owner for ingestion, infrastructure, and activation, not just a media plan. Scope whether that capacity sits in-house, with a vendor, or in a combined team.

Examples: Source 1

The U BEAUTY panelist argued that the path to a second purchase matters most and that discounting in luxury skincare is a drug and a race to the bottom. The recurring panel theme was that data and measurement exist, but most brands still lack the activation layer connecting customer value to ad spending.

For brands planning a campaign

For beauty, wellness, or other retention-sensitive brands, connect margin-based value signals to second-purchase cohorts. A campaign aimed at acquiring customers should ask how acquisition bidding can favor customers likely to buy again without relying on discounts.

Examples: Source 1

Conversation topics include the integration layer, how partners create more value for shared customers, and what growth and retention look like when the data layer works end to end. They explicitly invite conversations about ad signals and what is broken in performance marketing.

In-person events and partner networking for operator conversations.

For brands planning a campaign

For event, partner, or ecosystem-led campaigns, Abhi Agnihotri is a relevant voice in DTC retention and ad-signal infrastructure. A campaign brief could feature panels, dinners, or partner co-marketing on data-layer integration or performance-marketing repair.

Examples: Source 1 · Source 2

Ask for Abhi’s formal role and ownership at the company.

View example

Ask for the partnership’s terms.

View example

Both treat what a brand sends the ad platform as the lever that changes acquisition outcomes.

Reeto M.'s posts focus on controlled comparisons: a two-cell A/B test of predictive CAPI against an incumbent, and Meta's incremental attribution reporting. Abhi Agnihotri takes a more architectural view, covering Google's expected-value variable and the match-rate layer that determines whether signals arrive at all.

Read original · Original 2

Both explain why platform metrics diverge and why the fix is rarely a settings toggle.

Bivas Kumar Roy focuses on diagnosis and implementation — event IDs, deduplication, consent states, GA4 and GTM under EU privacy rules — and stops at making measurement trustworthy. Abhi Agnihotri carries the argument into bidding, value signals and acquisition strategy.

Read original · Original 2

Both argue that bad data infrastructure quietly erodes profit.

Samuel Adeitan widens the lens to agency operations — self-hosted workflow engines, Python transformation scripts, queueing and the cost of pay-per-task automation. The remedy is operational infrastructure; Abhi Agnihotri's lies in the ad platform's optimization target.

Read original · Original 2

Their posts describe an A/B test design: a two-cell test of predictive CAPI against the current setup, including Elevar, Blotout, Triple Whale Sonar, Shopify, Segment, or homegrown tools, with blended new-customer CAC as the north star instead of attribution-reported numbers.

• They also report patterns from Meta's incremental attribution reporting, including cases where incremental lift outperformed standard 7-day click reporting.
• For a Meta-heavy DTC brand with a large returning-customer base seeking a tested path to pLTV signals and a vendor partner, this is closer to the buying decision than Abhi's explanatory roadmap.

The tradeoff is narrower platform scope and less teaching depth on Google.

• Reeto M.'s posts center on Meta CAPI, predictive CAPI, incremental attribution, and new-customer CAC.
• They do not lay out Google's auction-time bid formula, OCI activation, or the same first-party identity and match infrastructure Abhi describes across both platforms.
• If the campaign needs one coherent Google-and-Meta pLTV roadmap, pair Reeto M.'s proof design with Abhi's platform mechanics rather than replacing one with the other.

Examples: Source 1 · Source 2

Their posts set out a concrete diagnostic order: check attribution windows and click-through versus view-through first, then verify Event IDs and browser/server deduplication, test whether Purchase fires more than once, check order values and currencies, and review consent and privacy limitations.

• They also describe work with EU businesses and agencies across Meta CAPI, Google Ads, GA4, GTM, and Consent Mode V2.
• For a brand that must clean its signal foundation before trusting any pLTV activation, this is the more practical starting point.

The tradeoff is that Roy's posts diagnose and repair tracking; they do not set out a predictive LTV model or show how to inject a pLTV score into Google's or Meta's bidding stream.

• Abhi's material supplies that value-signal strategy.
• Use Roy when measurement hygiene is the unresolved dependency, and Abhi when the decision is how to change what the auction optimizes toward.
• Roy's posts also focus on troubleshooting examples rather than a full implementation roadmap.

Examples: Source 1 · Source 2

Their posts frame a pre-Q4 audit around three technical layers: CAPI deduplication with identical event IDs, new-versus-returning signals passed through server-side payloads, and offline order reconciliation that matches settled Shopify orders against ad platform click data.

• They also address the cost trap of scaling an agency on per-task automation platforms, with a self-hosted alternative using n8n, Python scripts, and a local Redis or database queue.
• For an agency that needs its tracking and data plumbing to handle client volume while protecting margin, this is closer to the day-to-day task than Abhi's predictive LTV activation argument.

The tradeoff is scope: Adeitan's posts focus on preventing margin loss from bad tracking and lowering the cost of high-volume data operations.

• They do not explain how to build or calibrate a predictive LTV score, replace Google's default average order value with a long-horizon value signal, or dispatch pLTV through Google Ads OCI and Meta CAPI.
• If the need is a signal-engineering roadmap for bidding, Adeitan covers the prerequisite audit and automation layer, while Abhi covers the value-signal layer.
• The two can be sequential rather than substitutes.

Examples: Source 1 · Source 2