Hightouch

What it is

Hightouch is warehouse-native activation. You write SQL against Snowflake, Databricks, BigQuery or Redshift, define a sync, and the rows land as fields in Salesforce, audiences in Meta or profiles in Braze — without a second copy of your customer data living in the vendor’s storage. That was the reverse ETL pitch in 2021. The company now sells the same substrate as an agentic CDP, and the platform spans identity resolution, a Customer Studio audience builder, Content Assembly for creative, and AI Decisioning, which uses reinforcement learning to choose the message, channel and timing per customer rather than running an A/B test and waiting. Hightouch raised a $150 million Series D on April 29, 2026, led by Growth Equity at Goldman Sachs Alternatives and Bain Capital Ventures at a $2.75 billion valuation, with Iconiq, Sapphire Ventures, Amplify Partners, Y Combinator and TD7 — The Trade Desk’s venture arm — also participating. The company states it has grown more than 100% in each of the past two years. That investor list tells you where the roadmap points: adtech activation, not data engineering.

Why it shows up in RevOps stacks

• The warehouse stays the system of record. Hightouch lists 300+ integrations and does not ask you to re-ingest events into its own store, which is the whole argument of the composable CDP pattern. If your dbt models already define what a qualified account is, that definition is the segment — you do not rebuild it in a marketing tool’s UI and then spend a year reconciling the two. This is the specific case where Hightouch earns its money: one modeled customer table fanning out to a dozen ad, lifecycle and CRM destinations with a single source of truth behind it.
• Seats are not the meter. The Composable CDP tier includes unlimited user seats and bills on monthly tracked rows — the count of unique records synced per month. A 40-person growth team costs the same as a 4-person one at equal data volume, which inverts the cost curve of almost everything else in a GTM stack.
• The MCP server is first-party and carries no surcharge. It connects Claude, ChatGPT, Gemini Enterprise or Copilot to the workspace so audiences, journeys and creative can be built in plain language, and Hightouch’s docs state access costs nothing extra. Read the gating caveat in the watch-outs before you plan around it.
• Gartner recognition landed. Hightouch’s own page states it was recognized as a Leader in the Gartner Magic Quadrant for Customer Data Platforms for the first time, dated January 2026. Treat that as a vendor claim about an analyst placement, but it does move Hightouch through enterprise procurement in a way it could not two years ago.

Pricing reality

There is no published dollar figure anywhere on the pricing page. Three shapes are offered: a free Basic Reverse ETL tier capped at 2 active syncs with unlimited destinations and seats, a usage-priced Composable CDP, and a usage-priced Agentic Marketing Platform sold on a preset allowance of monthly personalized actions. Everything past the free tier is a quote.

Buyer data fills the gap. Vendr reports a median annual contract of $15,000 across 170 analyzed purchases, a range of $9,600 to $75,000, and an average 26% discount off list. The same source puts the bands at roughly $1,000 to $2,500 per month for deployments up to 100,000 monthly tracked rows, $2,500 to $8,000 from 100,000 to 1 million, $8,000 to $25,000 from 1 million to 10 million, and above $25,000 per month past 10 million. Read that as: a mid-market B2B activation use case lands near $15,000 to $30,000 a year, and consumer-scale volume is a six-figure line item.

Best for

RevOps and lifecycle teams at companies that already run Snowflake, Databricks or BigQuery with a modeled customer table and someone who writes SQL — and who want the audience definition to live in the warehouse rather than in a marketing vendor’s segment builder. Above roughly 1 million tracked rows with several ad destinations, the warehouse-native pattern is the cheaper and more auditable option.

Do not buy it if you have no warehouse. Hightouch activates data; it does not create it, and buying activation before you have a modeled customer table means paying for an empty pipe. Do not buy it as an event-collection layer either — Hightouch Events exists, but that is not where the product is strong.

Versus the alternatives

Twilio Segment is the packaged-CDP incumbent and collects events itself. Pick Segment when you have no warehouse, no data team, and need identity and activation to work out of the box in weeks. Pick Hightouch when the modeling capacity already exists and you refuse to maintain segment logic in two places.

Fivetran Census is the closest direct rival, and the strategic picture changed when Fivetran acquired Census in May 2025 and folded it in as Fivetran Activations. Pick it when you already buy Fivetran for ingestion and want activation on one invoice with one vendor relationship. Pick Hightouch when activation is the harder half of your problem — the destination catalog, identity resolution and decisioning layers are deeper here, and you are not betting the activation roadmap on an ingestion company’s priorities.

GrowthLoop is the fastest-growing entrant, an agentic composable CDP on the data cloud that took a follow-on investment from TJC in January 2026 and ships its own Composable AI Decisioning. Pick GrowthLoop when the buying center is a large enterprise marketing org and you want a vendor whose entire product is the agentic loop. Pick Hightouch when you also need the boring, reliable sync layer underneath and the destination breadth to go with it.

If none of these fit — because the real problem is that finance, sales and marketing each define an active customer differently — you have a modeling problem, not an activation one. Fix the dbt models first; any of these tools will faithfully broadcast the wrong definition to 300 destinations.

Watch-outs

• The monthly-tracked-rows meter charges for churn in your data, not for value. A model that touches every row nightly because of a volatile field bills like a model that syncs genuine changes. Guard: before signing, run the free tier against your actual candidate model for two weeks and count distinct synced records, then price that number rather than your table’s row count. Ask for the overage rate per thousand rows in writing.
• Every paid price is a quote, and a 26% average discount means list is soft. Guard: enter the negotiation with the Vendr band for your tracked-row volume, ask for the tier boundaries in the order form, and cap the annual uplift — Hightouch’s own material describes economies of scale, so a volume commitment should buy a rate, not just a bigger bill.
• The MCP server is not on by default. Hightouch’s docs state plainly that it is not available in all workspaces and that you must contact Hightouch to enable it. It costs nothing extra, but “free” and “provisioned” are different facts. Guard: if the agent-driven workflow is part of why you are buying, get MCP turned on in the trial workspace and run a real audience build through Claude before signature — not after.
• The AI Decisioning proof points are vendor-published customer results. WHOOP at a 22% increase in loyalty offer activation and PetSmart at a 10% lift in cross-sell conversions are Hightouch’s own case studies with no independent replication. Guard: insist on a holdout group in the first campaign and measure incrementality yourself. Reinforcement learning against a badly specified goal optimizes hard in the wrong direction, so write the guardrails and the success metric before the first run.
• The positioning has moved twice in five years — reverse ETL, then composable CDP, then agentic CDP — and the Series D pushes further toward marketing execution. Teams that bought the plumbing may find the roadmap attention is elsewhere. Guard: buy the syncs you need today and keep your transformation logic in dbt, in your repo. Your exit cost should be rewriting sync definitions, not rebuilding your data model.
• Warehouse compute is not on the Hightouch invoice. Frequent syncs against large models are queries, and Snowflake or Databricks bills for them. Guard: set sync schedules from business need rather than defaulting to the shortest interval, and track warehouse spend attributable to Hightouch queries for the first 60 days so the true cost of ownership is a measured number and not a surprise.