ABM revenue stack — intent signals to closed-won

Between February and August 2026, every layer of this stack shipped an MCP server. Outreach reached general availability on 2026-02-24. Demandbase announced MCP delivery alongside its April 14, 2026 platform relaunch. Common Room documents read-and-write MCP on paid plans. Gong ships one on any plan. 6sense opened its own in beta on 2026-07-14 with GA set for August 2026.

That changes what this stack is expensive at. The integration tax — six enterprise systems pointed at one account record, each with its own connector, sync cadence, and field-mapping argument — was the reason teams bought fewer layers than they wanted. It is now a policy problem instead of a plumbing problem, and the policy is harder than the plumbing was.

What has not changed is the reason to buy any of it: six-figure deals, 6-12 month cycles, buying groups of 8 or more stakeholders, where intent-data depth is the differentiator. Heavier and more expensive than the outbound stack, and only worth it when the deal economics support the investment.

How the pieces fit

• 6sense is the intent and predictive layer. It identifies in-market accounts, surfaces buying-stage signals, and scores accounts by probability-to-close — the demand layer the signal-based selling discipline runs on. RevvyAI went to open beta for Revenue Marketing customers and general availability for Sales Intelligence customers at no additional cost on 2026-05-12. The 6sense MCP Server exposes account insights, predictive buying stages, 6QA status, keyword intent, and ad campaign performance to Claude, ChatGPT, Writer, and Agentforce. All three of 6sense’s published plan shapes are contact-sales.

• Demandbase is the second intent source, and it is now sold as one product or none. The April 14, 2026 relaunch at GO London put a conversational interface on top of the platform — Demandbase AI Chat, a Site Customization Agent, and a Context Intelligence layer — on top of the earlier Agentbase agents. Company, contact, technographic, and intent data reach ChatGPT, Claude, Copilot, and Gemini through MCP. Two buying constraints matter here: Demandbase states it does not split Demandbase One into a marketing-only or sales-only purchase, and its pricing shape is a platform fee plus a flat fee per user.

• Salesforce is the CRM substrate and the only layer with a published rate card. Checked 2026-08-08: Starter Suite $25, Pro Suite $100, Enterprise $175, Unlimited $350, and Agentforce 1 Sales $550 per user per month on annual billing. Enterprise is the floor this stack assumes — Advanced Pipeline Management and the integration surface below it are what an account-based motion needs from a CRM.

• Outreach is the sequence engine, and it is no longer sold as a sequencer. The spring 2026 relaunch announced 2026-04-27 moved the company to outreach.ai and repositioned it around agents: Omni as a conversational agent in Slack and mobile, Agent Studio for building custom agents on a visual canvas, and named agents including Meeting Prep and Deal Agent. Pricing gained a consumption layer — Amplify Core, Plus, and Pro carry 25,000 / 50,000 / 100,000 AI credits respectively, on top of a per-seat licence. When 6sense surfaces an in-market account, Outreach runs the multi-touch orchestration across the buying group.

• Common Room is the person-level signal layer. Community activity, product-led signal, and job changes land on the same account record 6sense and Outreach work from. RoomieAI Spark auto-surfaces prospects ready for outreach with one-click actioning in Common Room, Slack, and email. It is the second layer here with a published floor: Essential is $2,500/mo billed annually and includes 5 seats, 5,000 RoomieAI research credits, and 2,500 Prospector credits. Advanced (15 seats) and Enterprise (30 seats) are quoted.

• Gong is the deal-cycle conversation layer. Every call across the 6-12 month cycle gets analyzed for MEDDDPICC progression, competitive mentions, and stakeholder coverage. Gong MCP ships both a client and a server, works on any Gong plan, and is configurable only by a technical administrator through Admin Center > Settings > Ecosystem > API > Integrations. It reaches call recordings, transcripts, meetings, users, and conversation analytics.

Named handoffs

1. 6QA fires in 6sense → account is flagged in Salesforce → Outreach enrolls the buying group. The trigger is a buying-stage transition, not a lead score, which is why the enrollment covers 8 stakeholders rather than one form-fill.
2. Common Room detects a person-level signal → upserts the contact and logs a timestamped activity → the rep sees it against the same account. Upsert is the operative behavior: a duplicate create updates the existing record instead of forking it.
3. Outreach books the meeting → Gong records and analyzes the call → MEDDDPICC gaps route back to the AE. This is the handoff that tells you whether the intent signal was real.
4. An external agent queries all six over MCP → and hits a different write contract at each one. That one is not a handoff so much as a hazard.

The write asymmetry nobody diagrams

Six MCP servers do not add up to one agent surface. They disagree on what an agent is permitted to change, and the disagreement is deliberate on at least one side.

• Common Room supports create and update with upsert semantics — contacts, organizations, activities and notes, and static segments.
• Outreach is read, create, and delete only. Outreach excludes record updates on purpose, reasoning that model behavior when editing existing records is unpredictable. Access requires a licensed seat with the Amplify add-on plus an admin toggle in organization settings, so it is a paid-tier capability rather than a platform default.
• Gong reaches calls, transcripts, and conversation analytics. Its public API v2 has no deal-board read endpoint — GET /v2/crm/entities returns only objects you previously uploaded through a registered generic CRM integration. Deal risk has to be derived from tracker occurrences joined to party affiliation, not read off a board.

So an agent instructed to “update the deal record” updates cleanly in Common Room, has no update verb in Outreach, and cannot read Gong’s deal board at all.

Guard: write the per-system write policy before you enable the second MCP server, and put it where the agent reads it. The specific failure to pre-empt is an agent reaching delete-then-create as a substitute for the missing update verb in Outreach — that is a permitted sequence of permitted calls, and it destroys the record’s history.

Cost reality

Two of the six layers publish anything. Salesforce Enterprise at $175 per user per month is $63,000/yr for 30 seats; moving that same 30 seats to Agentforce 1 Sales at $550 adds $375 per seat per month, or $135,000/yr. Common Room Essential is $30,000/yr and stops at 5 seats, which puts a 30-AE team in the quoted Enterprise tier.

6sense, Demandbase, Gong, and Outreach publish no prices. Two of them publish the shape: Demandbase is a platform fee plus a flat per-user fee, and Gong is a platform fee sized by user count plus per-seat. Third-party reporting puts Outreach near $100 to $160 per seat per month across the three Amplify tiers, with implementation at $1,000 to $10,000-plus — that is not vendor-published, and belongs in a budget as a negotiating band rather than a number.

The line item that is new in 2026 is metered AI. Outreach sells credits by Amplify tier, Common Room meters RoomieAI research and Prospector credits separately, and Salesforce meters Flex Credits. Three of the six layers now carry a consumption cost that headcount does not predict.

Guard: ask each vendor what happens at credit exhaustion — hard stop, overage rate, or degraded output — and get it in the order form. A stack with three independent credit meters has three separate ways to stop working in the last week of a quarter.

Match rules

Use this stack when: average deal value clears $100K and the cycle runs 6-12 months; the buying group is large enough that account-level orchestration beats contact-level sequencing; you already run Salesforce Enterprise or above; and one named person owns which system holds truth for account state.

Do not use this stack when: deal values sit under $25K, where platform fees and per-seat costs consume the margin the signal advantage was supposed to create; the motion is volume outbound rather than account-based, which the outbound stack serves at a fraction of the cost; the team is under 10 reps, where the 6sense and Demandbase floors dominate everything else in the budget; or nobody will own the MCP write policy above.

Common variations

• One intent platform, not two. Running 6sense and Demandbase together is reserved for programs where advertising and pipeline attribution are separately owned and separately budgeted. Default to one. The rule for choosing: 6sense when predictive buying-stage scoring drives the sales motion, Demandbase when the advertising spend is the larger line item and Demandbase One’s all-or-nothing packaging is acceptable.
• Drop Demandbase, keep Common Room. Take this swap when the signal you are missing is person-level — a champion changed jobs, a user filed an issue, a prospect showed up in the community — rather than account-level. Common Room publishes a floor; the second intent platform does not.
• HubSpot-shop variant. Replace Salesforce with HubSpot Enterprise. Works at mid-market scale, harder to justify at the deal sizes this stack assumes.
• Agentforce instead of a separate agent layer. Agentforce 1 Sales consolidates the agent surface into the CRM at +$375 per seat per month over Enterprise. Take it when the constraint is that agents keep needing data the CRM already holds; skip it when the agents you want are the ones Outreach and 6sense already ship.
• Add an AI workbench layer. Clay plus Claude for per-account research and first-touch personalization — see the outbound stack for the integration pattern.

What this stack does NOT replace

• A demand-generation engine — paid, content, partnerships — that creates the brand awareness 6sense and Demandbase detect
• A sales enablement platform (Highspot, Seismic) at the scale this stack assumes
• A CPQ for complex pricing (Salesforce CPQ, Conga CLM, DealHub)
• A marketing automation platform (Marketo, HubSpot Marketing) for the nurture motion that feeds intent
• A post-close motion — add Gainsight for expansion, which runs on the same signals
• An agent permission model. Six MCP servers give you six vendor-defined write contracts and no single audit of what an agent changed across them. That gap is yours to close, and no component here closes it.