Attention

What it is

Attention records sales calls, transcribes them in 100+ languages, grades them against a scorecard (MEDDIC and BANT ship as templates), and writes the result back into Salesforce or HubSpot. That part is Gong’s product with a different logo on it.

The part that isn’t: Attention ships agents that take the action instead of reporting it. The Follow-Up Email Automator drafts and sends the email. The CRM Auto-Update Agent fills the fields. The Deal Risk Monitor scans conversations for risk indicators and notifies the manager. The vendor lists 100+ configurable agent templates, including a Knowledge Base Builder that watches calls for new product facts, a Lead Qualification Enricher, and a Competitive Intelligence Tracker. At its $30M Series B announcement on 24 June 2026 — led by RTP Global, with returning investors Aglaé Ventures, Eniac and Alven and new investor Linea — the company said those agents were running more than 20 million actions per month, on 500+ customers including Abridge, Scale, Lovable, Preply, Engine and BambooHR, with ARR up 4x year over year.

The correction to make before you evaluate it: Attention is not a notetaker you export from. It is a thing with write access to your system of record. Price it, scope it and govern it accordingly.

Why it shows up in RevOps stacks

The trigger is almost never “nobody is watching our calls.” It is that reps don’t send the follow-up and don’t update the record, and a second dashboard telling you so changes neither. Attention is bought to close that gap by doing the two tasks rather than measuring them.

The second reason is the depth of its agent surface. Attention publishes an MCP server at https://api.attention.tech/mcp and ships as a Claude Connector, which on its own is no longer a differentiator — Clari, Chorus, Fireflies and Granola all expose one, and Gong still does not. What is different is what the server exposes: 68 tools across 15 groups, covering Calls, AI Analysis, Scorecards, Forecasting, CRM, Intelligence Collections, Reports and Users & Roles, split across two OAuth scopes, mcp:read and mcp:write. Most conversation-intelligence connectors give an assistant read-only transcript search. This one gives it the workspace. If your deal reviews already happen inside Claude, that gap is the entire evaluation.

Pricing reality

Quote-only, and unusually opaque even by this category’s standards. attention.com/pricing returns HTTP 404 as of 2 September 2026, and no price appears anywhere on the marketing site — the path to a number is a demo request. Vendor docs name two commercial tiers in passing, Team and Enterprise, in the instruction that an organization Owner must add the connector under Organization Settings → Connectors before individual users can connect.

Third-party aggregator pages publish confident per-seat tables for Attention. Those numbers appear in no vendor source, they disagree with each other, and at least one of the sites confuses this vendor with Attention Insight, an unrelated eye-tracking company. Do not carry them into a negotiation.

Anchor on published comparables instead:

• Avoma lists Organization at $24/user/month annual, a Conversation Intelligence add-on at $29/seat/month and a Revenue Intelligence add-on at $29/seat/month. Two-way CRM field writes sit in the Revenue Intelligence add-on, so the closest published like-for-like — call recording, coaching and CRM write-back — lands at about $82/seat/month annual.
• Sybill publishes $90/user/month for the Business tier, which is where CRM Autofill starts and caps at 10 fields; unlimited autofill is Enterprise-only.
• Gong is also quote-only and runs $1,200-1,800 per rep/year — roughly $100-150/seat/month — with a 25-50 seat minimum.

Estimate, anchored on those three and marked as one: expect Attention’s agentic tier to quote in the $70-140/seat/month annual band for a 20-50 seat revenue team. It sells the write-back Avoma charges an add-on for, and it does not carry Gong’s seat floor. Ask for the number of agent actions included, not just seats — actions are what scales on this product, and a seat-only quote hides the ceiling.

Best for

A RevOps or sales leader on a 20-150 seat revenue team already standardized on Salesforce or HubSpot, whose measured problem is follow-up latency and CRM hygiene rather than call visibility. The scoped case where Attention wins outright: a team that runs pipeline questions and deal reviews inside Claude and wants the assistant to read call data and change the CRM record in the same session, instead of reading a dashboard and retyping the change by hand.

Not for

Teams under 10 reps whose actual need is meeting notes. Fathom, Granola and Fireflies cover recording, transcript and summary at $0-30/user/month published, and an agent layer has nothing to act on at 40 calls a month. It is also the wrong pick when forecast accuracy is the mandate that funded the purchase — Clari and the forecasting-first tools own that process, and Attention’s forecasting is a feature next to their product.

Versus the alternatives

• Gong — the share leader, and the one your buying committee already knows. Pick Gong when coaching depth and market intelligence across 100+ reps are the deliverable, when procurement wants the vendor nobody gets fired for, or when your CRM is neither Salesforce nor HubSpot. Pick Attention when the deliverable is the sent email and the written field.
• Clari — pick it when the board conversation is forecast accuracy and the buyer is the CRO. Attention writes the record; Clari argues about the number on top of it. Teams above roughly 150 reps often run both.
• Avoma and Sybill — the published-price challengers taking mid-market seats. Pick either when procurement needs a list price on paper, when you want to start at $19-30/user/month and add CI later, or when a 12-month agentic commitment is more risk than the follow-up problem justifies. The tradeoff is real: at Sybill’s Business tier the CRM autofill stops at 10 fields.

Watch-outs

• The MCP server acts as whoever authorized it. Vendor docs are explicit: if you can see a call in the app, the assistant can read it through the connector. A connector authorized from a sales leader’s session inherits that leader’s visibility across the org, and the mcp:write scope includes destructive tools — delete_user, archive_team, delete_label — gated only on admin role. Guard: authorize the org connector from a purpose-built service account holding mcp:read only, and keep mcp:write on one named admin with a written change log. Do not connect it from a super-admin session because that was the fastest way to get the demo working.
• Agents that send are agents that can send the wrong thing. Attention makes human-in-the-loop verification configurable rather than on by default, and the thing being sent goes to a prospect, not to an internal channel. Guard: turn review-before-send on from day one and hold it there until you have read 100 consecutive drafts with no edit. Set the rollback path — who kills the agent, from which screen — before the first send, not after the first bad one.
• Quote-only pricing with no published floor removes your anchor. You will negotiate against a number the vendor picked after reading your headcount. Guard: open with Avoma’s $82/seat/month like-for-like and Gong’s $100-150/seat/month band, request the agent-action allowance in writing, and cap renewal uplift in the first paper rather than the second.
• The write-back exists for two CRMs. Salesforce and HubSpot are the CRMs on the integrations page; dialers (Dialpad, Aircall), meeting platforms (Zoom, Google Meet, Teams) and Zapier fill out the rest. If your system of record is Pipedrive, Dynamics or something in-house, the capability you are buying does not exist for you. Guard: name your CRM in the first demo email and make the write-back a scripted step in the trial, not a slide.

Related: the conversation intelligence stack prices the surrounding layers, and alternatives to Gong covers the wider replacement set.