SAP Bet Its €21.9B Cloud Backlog on Claude. That’s How AI Wins Enterprise Distribution.

In April 2026, Hightouch closed a $150 million Series D at a $2.75 billion valuation, making it one of the most highly-valued companies in the marketing data infrastructure space. The headline number matters less than what CEO Tejas Manohar said about where the money is going: not toward more connectors, not toward a fancier UI, but toward an "agentic marketing platform" — a system where AI agents handle the full cycle of audience selection, campaign execution, and optimization without human intervention between iterations.

This is a meaningful bet. Traditional marketing automation built on Marketo, HubSpot, and Salesforce Marketing Cloud was designed around a human in the loop: a marketer defines a segment, chooses a channel, writes the copy, schedules the send, and reads the report. Agents collapse that loop. The question Hightouch is answering is: what does your data infrastructure need to look like when the optimization loop runs 1,000 times a day instead of once a week?

What Hightouch Actually Built

Hightouch launched in 2020 as a Reverse ETL company — the idea being that once you had moved your customer data into a warehouse (Snowflake, BigQuery, Databricks), you should be able to push it back out to operational tools without copying it into a separate CDP first. That was a $15 billion market insight: the warehouse was winning as the system of record, and legacy CDPs like Segment were at risk of becoming redundant middleware.

The Reverse ETL thesis held. By 2023 Hightouch had replaced Segment and mParticle at dozens of mid-market and enterprise companies. But the company saw a second wave coming: not just moving data from warehouse to tools, but using that data to drive autonomous decisions at the campaign level.

The agentic layer Hightouch built sits on top of its data activation infrastructure. It has three core components:

1. Audience Intelligence Engine The agent continuously re-segments users based on behavioral signals from the warehouse. Rather than a marketer building a static cohort ("users who signed up 30 days ago and haven't invited a teammate"), the agent maintains dynamic segments that update in real time as warehouse events stream in. Snowflake and BigQuery both support change-data-capture streaming now, which makes this technically feasible at scale for the first time.

2. Experiment Orchestrator For any given segment, the agent generates candidate campaign variants — different messages, different channels, different timing — and runs them against each other using a multi-armed bandit rather than traditional A/B testing. The key difference: a bandit allocates more traffic to winning variants in real time, so you're not burning 50% of impressions on a losing variant for two weeks while a test runs to statistical significance.

3. Attribution Feedback Loop Results flow back into the warehouse as structured events, which the agent reads to update its priors for the next campaign cycle. This closes the loop without requiring a human to open a report, interpret a dashboard, and manually apply learnings to the next campaign.

Why Traditional Marketing Automation Is Losing

The market timing for this thesis is real. Marketing automation platforms built in the 2010s were designed around a set of assumptions that have all weakened simultaneously.

Email deliverability has collapsed. Gmail and Outlook's AI-powered filtering now deprioritizes bulk sends from any domain that cannot demonstrate engagement. The playbook of "blast your list monthly" produces diminishing returns. You need to send the right message to a small, highly-relevant segment — which requires the kind of real-time behavioral data that lives in your warehouse, not in a separate email platform's contacts table.

Channel proliferation makes single-platform automation obsolete. A modern SaaS growth team runs programs across email, in-app, SMS, push, paid retargeting, LinkedIn, and increasingly conversational AI touchpoints. Coordinating timing and frequency across these channels from a single workflow builder is a scheduling nightmare. Agents that read a shared context layer (the warehouse) can coordinate across channels natively, eliminating the problem of a user getting an email and an in-app notification and a paid retargeting ad on the same day about the same thing.

The economics of human campaign managers do not scale. A mid-market SaaS company with 50,000 users and five segments might run 25 active campaigns at any time. A full-stack growth engineer can manage 5-10 campaigns at depth. The rest get neglected, run stale, and generate unsubscribes. Agents don't get fatigued, and they do not have context-switching costs.

The competitive gap is not incremental. It is architectural.

The Composable CDP Advantage

Hightouch's structural advantage in building this agentic layer is that it never owned the data itself. Segment built its CDP as a system of record — your customer data lives in Segment's infrastructure, which means Segment controls access, pricing, and portability. That worked when Segment was the dominant player, but it created a lock-in dynamic that enterprises increasingly rejected when the warehouse became cheap enough to use as the primary data layer.

Hightouch's composable model means the agentic marketing system runs on your data, in your infrastructure, with your security controls. For enterprises in regulated industries — fintech, healthcare, SaaS companies with EU customers under GDPR — this is not a nice-to-have. It is a hard requirement that rules out any system that copies customer behavioral data into a third-party store.

The PLG activation ceiling problem is fundamentally a data freshness problem: by the time a human sees the engagement signal, opens the campaign builder, and queues the nudge email, the user has already churned or moved on. The composable architecture removes the latency. A user who fails to complete onboarding triggers a warehouse event, which the agent reads within seconds, which routes a personalized message within minutes — not 24 hours later when the batch sync catches up.

How Enterprise Teams Are Deploying This Today

The early adopter profile for Hightouch's agentic layer skews toward companies with three characteristics: they have already consolidated customer data in a cloud warehouse, they have more than 30,000 users (enough to make agentic optimization statistically meaningful), and they have outgrown what a four-person growth team can manage manually.

A SaaS infrastructure company in Hightouch's publicly shared case studies reported that switching from HubSpot workflow automation to the agentic layer cut their average time-to-first-meaningful-engagement from 8 days to 1.4 days. The mechanism: HubSpot was syncing user behavior data on a 24-hour batch schedule, so "recently activated" users were actually more than a day stale. The agent running against the warehouse read events as they happened.

The deployment pattern that is emerging in enterprise accounts looks like this:

1. Define success events Work backward from retention data to identify the 2-3 behavioral signals that predict 90-day retention. These become the agent's objective function. This step is non-negotiable — agents optimize for what you measure, and if you measure email opens instead of activated features, you will get aggressive subject lines and no improvement in actual retention.

2. Grant warehouse read/write access The agent needs read access to the behavioral events table and write access to an activation_events table where it logs what it sent, when, and to whom. This is the feedback loop. Without it, you are running experiments and destroying the results.

3. Set frequency and channel guardrails The most common early-adopter mistake is letting the agent run without sending frequency caps. Users who receive three messages in 24 hours from a "personalized" system do not feel understood — they feel spammed. Set hard limits: no more than one message per user per 48 hours across all channels, excluding triggered transactional messages.

4. Run in shadow mode for two weeks Before the agent sends anything live, run it in simulation — generating what it would have sent, to whom, across which channel — and audit the outputs for brand safety and coherence. Agents trained on engagement metrics can learn to be aggressive in ways that damage long-term brand equity.

5. Graduate to live sends with escalation logic Once you are comfortable with shadow mode outputs, flip to live sends with an escalation rule: if any segment's unsubscribe rate exceeds 2% in a rolling 7-day window, pause that segment and route to a human reviewer.

The Competitive Landscape

Hightouch is not building in a vacuum. The agentic marketing space is attracting serious competition from multiple directions, and the outcome of the next 18 months will define which architecture wins.

Braze has been moving toward AI-powered campaign orchestration for two years. Its Sage AI layer adds predictive audience modeling and send-time optimization on top of its existing messaging infrastructure. But Braze owns the data layer — you sync into Braze, not query your warehouse — which limits segment freshness and creates the same lock-in problem Hightouch is exploiting. Braze's retention numbers are strong among mobile-first companies, but its architecture is fundamentally at odds with the warehouse-centric data stack that enterprise SaaS companies are standardizing on.

Salesforce Data Cloud plus Agentforce is the obvious enterprise competitor. Salesforce acquired its own data cloud infrastructure, and Agentforce is its bet on autonomous campaign execution. The Salesforce advantage is the CRM relationship — most enterprise sales teams already live in Salesforce, which means the customer data is there too. The Salesforce disadvantage is the Salesforce tax: the platform is expensive, complex, and notoriously slow to deploy. A company that needs to run agentic activation experiments in two months will not choose Salesforce.

Snowflake Cortex is the wildcard that Hightouch investors need to be watching carefully. Snowflake has been building native ML and orchestration capabilities directly into the warehouse layer. If Snowflake ships a "campaign agent" feature that connects natively to the data already inside Snowflake, the middleware layer Hightouch occupies gets compressed. The bull case for Hightouch is that multi-cloud data stacks (Snowflake plus BigQuery plus Databricks) are the norm at enterprise scale, and a warehouse-agnostic orchestration layer has durable value. The bear case is that most companies ultimately standardize on one warehouse, and the warehouse vendor eats the activation layer.

The SaaS retention cliff at month one is fundamentally an activation speed problem: users who do not reach a meaningful outcome within the first week have a dramatically higher probability of churning. Whatever infrastructure lets you close that gap fastest wins the market.

What the $2.75B Valuation Is Pricing In

At $2.75B on $150M raised (ARR not publicly disclosed, but estimated at $60-80M by secondary market trackers at the time of the round), Hightouch is trading at a significant premium. That premium is not for the Reverse ETL business — it is for the agentic platform bet.

The bull case: every company with more than 10,000 users and a warehouse eventually needs this. The market is effectively every SaaS company that wants to run more sophisticated growth programs than a four-person team can manage manually. That is a very large TAM, and the composable architecture means that switching costs compound over time as more data flows through the warehouse.

The bear case: the warehouse vendors have been quietly building native orchestration and ML capabilities. If Databricks releases a campaign agent feature that connects natively to its data warehouse, the middleware layer Hightouch occupies gets compressed from below. The timing question is whether Hightouch can reach a critical mass of enterprise accounts — where the agentic layer is deeply embedded in their growth infrastructure — before the warehouse vendors close the capability gap.

The honest read is that Hightouch has a 12-18 month advantage in production-grade agentic marketing infrastructure, and they are using the $150M to extend that lead through integrations, compliance tooling, and an enterprise sales motion before the warehouse vendors make their move.

What This Means for Growth Teams Right Now

The strategic implication of the Hightouch raise is not that every company needs to buy Hightouch. It is that the campaign-builder paradigm — human writes copy, human selects audience, human schedules send, human reads report — is being disrupted, and teams that do not have a plan for autonomous optimization will be at a compounding disadvantage as the gap widens.

Activation benchmarks across SaaS show that companies hitting sub-2-day time-to-value are almost universally running real-time behavioral triggers, not weekly batch campaigns. The technology to close that gap exists today and is increasingly within reach of teams that have already made the warehouse investment.

For growth teams evaluating this space now:

• If your customer data is not in a warehouse yet, that is step one. The agentic layer runs on warehouse data. Nothing else works.
• If you are running HubSpot or Marketo on a batch sync, you are operating with 24-hour-old data. For activation-phase users, that is often too late.
• If you are already on Reverse ETL (Hightouch, Census, Polytomic), you are well-positioned to evaluate the agentic layer. The infrastructure is in place — you just need the orchestration on top.
• If you are a Braze or Salesforce shop, evaluate the roadmap seriously. Both are building toward this, but neither has production-grade agentic activation today.

The AI-native SaaS retention playbook points to the same structural pressure: companies running continuous optimization loops against real-time behavioral data outperform companies running batch campaigns on stale data. The question for growth teams is whether to build this capability internally, buy it from Hightouch, or wait for their existing MAP vendor to catch up. Given the pace of product development at Hightouch, waiting is a losing strategy.

The composable CDP won the data infrastructure debate of the 2020s. The agentic marketing platform is the next debate, and it is beginning now.

Takeaway: Hightouch's $150M Series D is less about the money and more about the thesis: AI agents will run the activation and retention loop, and companies that built their data infrastructure around a composable warehouse model will be able to plug in and benefit from autonomous optimization. The window to make the architectural decisions that position you for this shift is 12-18 months. After that, you will be retrofitting.