Marketing Automation Stack Notes 2026

What changed in 2026, in one paragraph

The CRM wars are over. Salesforce and HubSpot stopped fighting over who owns the system of record and started fighting over who owns the agent layer on top of it. The composable CDP won the data layer — Hightouch was named a Leader in the 26 January 2026 Gartner Magic Quadrant for Customer Data Platforms, and Forrester named it a Strong Performer in its B2B CDP Wave. And Model Context Protocol (MCP) — the open standard originally developed by Anthropic — became the integration glue between your CRM, your data warehouse, and the AI tools your reps are already using. The 2026 stack has four layers, but the agent layer has moved inside the CRM and now ships with the system of record, not beside it.

The four layers, in real product names

CRM core. Two platforms own mid-market and enterprise mindshare. HubSpot discloses it is “trusted by 306,000 customers worldwide” (HubSpot, 2026). Salesforce claims more than 18,000 companies run on Agentforce as of mid-2026 and was named a Leader in the 2026 Gartner Magic Quadrant for Conversational AI Platforms (Salesforce, 2026). Pipedrive still wins on simplicity for sub-50-seat sales orgs but is no longer a relevant MAP/reporting layer at scale.

MAP / lifecycle. The boundary is dissolving. HubSpot Marketing Hub is the default for mid-market. Marketo Engage still anchors legacy enterprise stacks. Customer.io describes itself as an “AI-powered customer engagement platform” and now ships a native AI Agent that builds campaigns from a prompt; the product page confirms it “works with any AI tool using MCP, CLI, API or webhooks” (Customer.io, 2026). Braze sells in four named editions — Go, Select, Pro, Enterprise — priced on Monthly Active Users plus Action Credits for AI features, with BrazeAI Agent Console as a Pro+ capability (Braze, 2026). Iterable’s Nova layer (Decisioning + Agents) is now the headline product. ActiveCampaign discloses 180,000+ customers across Starter, Plus, Pro, and Enterprise tiers, with Claude and ChatGPT listed as native integrations via MCP (ActiveCampaign, 2026).

Data and intent. Three buckets matter. (a) Intent: Bombora was named a Leader in the Forrester Wave for Intent Data Providers for B2B (Q1 2025) and discloses that 86% of its Company Surge co-op data is exclusive to Bombora (Bombora, 2026). (b) Data vendors: ZoomInfo (35,000+ customers, Forrester-cited Leader in B2B marketing and sales data), Apollo (600,000+ customer companies, 240M+ contacts, 30M+ companies), and Clay (500,000+ GTM teams, 200+ data vendors, $5B tender-offer valuation reported by the New York Times on 28 January 2026) all compete for the contact/company layer. (c) Composable CDP / Reverse ETL: Hightouch is now the default reference vendor. It connects to Snowflake, Databricks, BigQuery, Redshift, and Azure Synapse, syncs to 300+ destinations, and runs 1M+ daily sync jobs at 99.99% uptime (Hightouch, 2026). Databricks invested in Hightouch in 2025; Snowflake named it 2026 Marketers & Advertisers Product Partner of the Year (Hightouch, 2026).

Agent layer. Two models now coexist. (a) Agents inside the CRM: HubSpot Agent Hub (formerly Breeze Agents) and Salesforce Agentforce. (b) Agents glued to the CRM from the outside: Claygents, Apollo MCP, ZoomInfo Copilot, and a long tail of point solutions. The interesting 2026 move is that the agents-inside-the-CRM camp is now outcomes-priced — you can forecast agent spend the way you forecast media.

The outcome-priced agent budget

This is the change that has real 2026 financial consequences.

HubSpot Agent Hub ships with three named outcomes, all priced per resolved outcome rather than per seat: $0.50 per Customer Agent resolution, $1.00 per Prospecting Agent outreach drafted, and $0.10 per Data Agent answer (HubSpot, 2026). Agent Hub is included with Professional and Enterprise tiers; custom agents built in Agent Builder consume HubSpot Credits, and the public pricing catalog links from the product page (HubSpot, 2026).

Salesforce Agentforce sells in four parallel models (Salesforce, 2026):

• Salesforce Foundations: $0. Includes Agentforce Builder, Prompt Builder, Agent Script, Agentforce Coworker, and Agentforce Vibes for any Salesforce customer.
• Flex Credits: $500 per 100,000 credits (i.e., $0.005 per credit). Each standard Agentforce action costs 20 credits, so $0.10 per action. Agentforce Voice actions cost 30 credits each ($0.15 per action).
• Conversations: $2 per resolved conversation. Optimized for customer-facing agents.
• Per-user licensing: Agentforce add-on at $125/user/month for Sales/Service/Field Service; Agentforce Industries add-on at $150/user/month; Agentforce 1 Editions from $550/user/month with 2.5M Flex Credits per org per year; Agentforce User License at $5/user/month (requires Flex Credits).

A worked example from the Salesforce pricing page: a customer-facing “Where is my order?” agent that authenticates the customer and retrieves the order runs two actions, costing $0.20 per use case. Salesforce models 20 requests per day across 30 days at $120/month. A service case-management agent running 100 users, 3 cases per day, 20 days per month — three actions per case at $0.30 each, $1,800/month (Salesforce, 2026). A 10-rep field service team running 3 appointments per day runs five actions per appointment at $0.50 each, $300/month. These are vendor-published examples; actual consumption varies.

The practical difference: HubSpot’s per-outcome pricing caps your downside because a “resolution” or “answer” is a defined thing. Salesforce’s per-action pricing is more granular and tracks work rather than results — a 5-action troubleshooting flow costs five times what a 1-action lookup costs. If your agentic use case is high-volume and low-complexity (data lookup, simple routing), Salesforce’s per-action model is cheaper. If your agents handle messy multi-turn work and your goal is outcomes, HubSpot’s per-resolution model is more predictable. Treat both like media budgets: forecast a monthly consumption, set an alert, and renegotiate quarterly.

The composable CDP won, quietly

The 26 January 2026 Gartner Magic Quadrant for Customer Data Platforms elevated Hightouch to the Leaders quadrant for the first time. Gartner’s evaluation criteria are execution and vision; the report was authored by Lizzy Foo Kune, Rachel Dooley, Suzanne White, Benjamin Bloom, and Audrey Brosnan (Hightouch via Gartner citation, 26 January 2026). Forrester named Hightouch a Strong Performer in its Customer Data Platforms for B2B Wave (Hightouch, 2026). The 2026 IDC MarketScape for AI-Enabled CDPs for B2C Users also names Hightouch a Leader.

What “composable CDP” means in 2026: the data lives in the warehouse (Snowflake, Databricks, BigQuery, Redshift, Synapse). The activation layer sits on top, queries that warehouse with SQL, and syncs rows into the tools that need them. No data copy. No parallel identity store. No months-long implementation. Hightouch publishes that 90% of customers are fully onboarded in days or weeks (Hightouch, 2026) and contrasts that with “6–9 month implementation” timelines for traditional CDPs (Hightouch, 2026).

The sales pitch is the same one it has been for two years, but in 2026 the analyst recognition finally lines up. Treat Segment, Treasure Data, and ActionIQ as the legacy holdouts for now; they remain reasonable choices when you cannot get the warehouse funded. Census is the closest head-to-head to Hightouch and is the second-place reference for warehouse-native activation. mParticle remains the default if you have a mobile-first product with strict identity-stitching needs; Twilio Engage lives at the edge of Segment and Twilio’s communications stack and is rarely the right answer for a B2B revenue team.

MCP is the integration layer your AI tools already want

Model Context Protocol is an open standard, originally developed by Anthropic, that defines how AI models connect to external tools and data sources. Salesforce’s framing — “USB-C for AI” — is the cleanest (Salesforce, 2026).

In 2026, the major GTM platforms shipped an MCP surface:

• Salesforce Agentforce MCP Support connects Agentforce to MCP servers from AgentExchange partners, with a centralized MCP server registry and enterprise-grade auth (Salesforce, 2026).
• Apollo MCP exposes Apollo’s data and workflow engine to any MCP-aware client, with budget guardrails so reps can use it without ops losing control (Apollo, 2026).
• Clay MCP lets reps find, enrich, and route contacts from inside Claude, ChatGPT, or Codex, with 200+ data vendors reachable from a single prompt (Clay, 2026).
• Customer.io explicitly markets “Built for agents, not just humans” and supports “any AI tool using MCP, CLI, API or webhooks” (Customer.io, 2026).
• ZoomInfo MCP brings ZoomInfo’s contact and intent data to AI tools; ZoomInfo has a parallel ChatGPT integration that surfaces the same data (ZoomInfo, 2026).
• ActiveCampaign lists Claude and ChatGPT (via MCP) in its integrations matrix (ActiveCampaign, 2026).

Why this matters: MCP lets your reps do their actual job (“find me ten VP-Engineering contacts at Snowflake, Databricks, or MongoDB, enrich them, push to my sequencer, and log the activity”) in a single natural-language prompt, instead of three Zapier chains and a brittle Python script. The architecture question is no longer “do we have an API?” — it is “do we have an MCP server?” If you don’t, an AI tool can’t reach your data, and your team will keep paying for the same manual orchestration.

The four failure modes, in their 2026 form

The same four failure modes that broke 2018 stacks are still breaking 2026 stacks. The difference is that agents now execute on the bad data at machine speed.

1. Duplicate records. Still the most common B2B CRM problem, now compounded by AI-generated enrichment. The “augment data, then merge” loop is the source of most modern duplicate explosions. HubSpot’s Data Hub sells “real-time deduplication” and “custom deduplication” as named features, which is itself evidence the problem persists (HubSpot, 2026). Hightouch’s published contrast — “Traditional CDPs only allow you to access and activate basic customer data… Traditional CDPs are built on duplicative data storage — your database and theirs” — is the data-warehouse-native answer (Hightouch, 2026).

2. Lifecycle drift. Marketing says MQL, sales says SQL, customer success says “active customer.” When those definitions don’t match in the source-of-truth fields, every agentic workflow misroutes. Both HubSpot Agent Hub and Salesforce Agentforce promise to fix this with automation; the underlying problem is documentation, not tooling.

3. Attribution disagreement. MAPs credit first-touch and last-touch; CRMs stitch multi-touch; warehouses compute net-new using custom rules. Each is correct under its own definition; none match. Bombora sells Campaign Measurement as a standalone product because this problem is durable (Bombora, 2026). The pragmatic move is to pick one source-of-truth attribution report and stop debating the others.

4. Platform-of-record confusion. Where does the company record live? Hightouch markets against this directly: “Keep your CRM, email tools, advertising platforms, and analytics dashboards all on the same page by storing and activating data all in one place — your existing data infrastructure” (Hightouch, 2026). The doctrine in 2026 is the same as in 2025: warehouse is the system of record, CRM is the system of engagement, MAP is the system of execution.

A 90-day stack simplification, version 3.0

Days 1–30: stabilize. Pick one source of truth for company records (warehouse or CRM, not both). Stand up dedupe rules on contact, company, and lead objects. Document one lifecycle definition per stage with named owners. Audit your CRM and MAP licenses and cancel what isn’t being used weekly.

Days 31–60: instrument. Stand up a reverse-ETL pipeline (Hightouch or Census) so the warehouse feeds the CRM and MAP. Stop syncing company fields bidirectionally. Pick one attribution report and decommission the others. Connect at least one MCP server so your reps can query your data from Claude or ChatGPT.

Days 61–90: automate cautiously. Turn on one agent at a time. Start with the lowest-risk, highest-frequency surface: HubSpot Data Agent for CRM research at $0.10 per question, or Customer.io’s AI Agent for transactional lifecycle emails. Set a hard monthly budget per agent and a kill switch. Document what “good” looks like before the agent goes live, and measure it after week two.

What to watch over the next two quarters

The agent platform war is the active front. Salesforce’s Foundations free tier is the more aggressive distribution move; HubSpot’s per-outcome Credits pricing is the cleaner unit economics. Mid-market will consolidate on HubSpot because credits are easier to forecast than enterprise contract renegotiations. Enterprise will stay on Salesforce because the Data 360 + Agentforce + MuleSoft stack is the only one that credibly touches the rest of the CRM.

Composability will keep winning on the data side. Hightouch’s investor base — Databricks invested in 2025, Snowflake named it 2026 Partner of the Year — locks in warehouse-native activation as the default for another 18 months (Hightouch, 2026). Census will remain the credible second choice; Segment will continue to lose ground in B2B.

Intent data is bifurcating. Bombora wins on account-level intent but cannot tell you whether a specific person at that account is in-market. The repo is moving to provider blends — Apollo for contacts, Bombora for account surge, first-party product usage for behavioral signal — wired through Clay or a reverse-ETL pipe.

MCP will become table stakes. If your MAP, CDP, or data vendor doesn’t ship an MCP server in 2027, your AI tool can’t reach it, and your team will quietly route around it.

Caveats

Pricing captured from vendor product pages on 11 August 2026. HubSpot Marketing Hub tiered pricing was not directly readable from the public pricing page; HubSpot publishes tier names and feature lists but routes non-Starter tier discovery to sales. Salesforce Marketing Cloud pricing was not directly verified for this version. Customer-count disclosures — HubSpot 306,000, Salesforce 18,000+ Agentforce orgs, ActiveCampaign 180,000+, Apollo 600,000+ companies, Clay 500,000+ teams, ZoomInfo 35,000+ — are unit-incompatible and should not be ranked against each other. Gartner Magic Quadrant citations on vendor sites are vendor-controlled excerpts; the full reports are behind Gartner’s paywall. Forrester Wave citations are similarly excerpted. We treated vendor-published customer outcomes and ROI figures as directional, not as audit-quality evidence.