Dovetail

What it is

Dovetail is the interview and conversation corpus that sits underneath a feedback stack. Where Canny collects feature requests and Productboard turns them into a roadmap, Dovetail holds the raw material those decisions are supposed to rest on: call recordings, interview transcripts, support tickets, survey responses and sales calls, stored as searchable, tagged, quotable evidence with the timestamp and the speaker still attached.

The company was founded in Sydney in 2017 and opened a San Francisco office around its US$63M Series A, led by Accel with Blackbird, Felicis and Grok Ventures participating, announced 20 January 2022. It spent its first years as the default UX research repository. In 2026 it repositioned as a customer intelligence platform, and the Summer Launch on 14 July 2026 shipped the features that make the claim testable: digital twins, a queryable stand-in for a segment built from that segment’s own calls and tickets; AI Agents that watch incoming signal and write back into Salesforce, Slack, Linear or a ticket; Channels 2.0 (beta), which classifies tickets, surveys and sales calls into ideas and weights them by the ARR of the accounts raising them; and AI Dashboards (beta).

Why it shows up in Customer Success stacks

• The MCP server is what lifts this above a searchable archive. The remote endpoint is https://dovetail.com/api/mcp, authenticated by OAuth 2.1 or a bearer API token, exposing 50-plus tools across workspace search, projects, docs, highlights, channels, contacts, tags and file import — read and write. ChatGPT, Claude, Figma and Microsoft Copilot connect as first-party connectors; Cursor and Windsurf attach over Streamable HTTP; Claude Desktop runs a local script. The practical effect: a CSM drafting a QBR narrative pulls three verbatim quotes from that account’s own calls inside the session where the drafting is already happening, rather than in a tool someone has to remember to open.

• Revenue weighting is what makes it a Customer Success tool rather than a research tool. Channels 2.0 enriches each classified idea with account and ARR data pulled from Salesforce or HubSpot. A theme raised by 40 trial users and a theme raised by four accounts worth $2M in renewal do not deserve the same roadmap treatment, and that ranking — not the clustering — is the output a CS leader takes into a prioritisation meeting.

• Evidence stays bound to the claim. Every theme resolves back to the highlight, the transcript timestamp and the person who said it. In an escalation or a renewal, that is the gap between “customers are asking for this” and a named quote from that customer’s own director, which is the version that survives being challenged.

Pricing reality

The self-serve tier is gone. Dovetail’s pricing page now lists exactly two plans: Free — $0, no card, one project, one channel, AI chat with basic summaries, basic AI clustering, integrations, search, contacts, templates and API access — and Enterprise, custom-quoted, with unlimited projects, channels, docs, agents and dashboards, semantic search, redaction of text, audio and video, access controls, and a dedicated CSM. Only Enterprise plans are invoiced. The per-seat Professional plan that used to sit between them was retired in 2026, so a comparison article quoting you a $15-to-$39 per-user monthly figure is describing a plan you can no longer buy.

That makes Free an evaluation sandbox rather than a starting point — one project and one channel will not hold a real team’s corpus, and there is no paid step that does not involve a sales cycle. Vendr’s marketplace data, updated February 2026 across 45 purchases, puts the median Dovetail contract at $21,600/yr, ranging $10,800 to $61,200. Treat $20-25K as the honest mid-market band. HIPAA is a paid add-on; ISO 42001, ISO 27001 and SOC 2 Type II come with Enterprise.

Best for

Customer Success and product leaders at roughly 200 employees and up, where customer conversations are already being recorded through Zoom, Google Meet or Gong, and the recurring failure is that nobody can locate the quote six weeks later. The fit sharpens when a research or insights function already exists and Customer Success wants to read from the same corpus instead of standing up a second one. AWS, Visa, Breville, Atlassian, Canva and Johns Hopkins appear on the public customer list.

Do not buy it if

Your feedback is already structured and your question is “what do we build next.” Canny answers that with a public board and vote counts for a fraction of the price, and Dovetail’s strength — unstructured conversation — is wasted on a queue of tidy feature requests. The same holds if you record no calls: the differentiation starts with transcripts, and without them you have bought a costly tagging tool. It is also not a survey platform or a participant recruiter. Dovetail reads Qualtrics, Sprig and Usersnap; it does not run the study or source the panel.

Versus the alternatives

• UserTesting — the largest install base in the segment, holding UserZoom and EnjoyHQ after Thoma Bravo and Sunstone Partners took it private in a $1.3B deal. Pick it when you need the participant panel and moderated testing on one contract; there the repository is a feature of a testing platform, not the product.
• Qualtrics — the largest budget line in experience management and the one most likely to already be in the building. Pick it when survey collection, program governance and case management have to sit on one contract. Pick Dovetail when the material already exists and the problem is reading it.
• Enterpret — the fastest-growing entrant, with a $20.8M Series A led by Canaan Partners announced 4 December 2024. Pick Enterpret when the volume is support tickets and reviews at scale and nobody is running interviews. Pick Dovetail when the highest-value evidence is a conversation a human had.
• Condens — still publishes prices: €15 per user per month on Lite, €500 per month flat for Business at five users and up. Pick it when a small research team wants a repository with a number on the website, which Dovetail no longer offers.
• Marvin — founded 2020 in Oakland, about $3.8M raised. Pick it when research ops — scheduling, participant tracking, incentive handling — matters more to you than corpus depth.

Watch-outs

• The pricing floor moved underneath new buyers. With Professional retired, a team outgrowing Free has exactly one path: a sales-quoted annual contract. Guard: before you load the corpus, get the seat tier, the renewal uplift cap and the bulk data-export format written into the first order form. A research repository is the textbook definition of switching cost, and every bit of negotiating room you have sits before signature.
• Free-tier limits are structural, not stingy. One project and one channel means you cannot pilot two teams in parallel or compare two channel configurations. Guard: scope the trial to a single question answerable in 30 days inside one project, and ask for a time-boxed Enterprise sandbox rather than trying to stretch Free.
• Agents and digital twins write into other systems, and both shipped in July 2026. An agent that updates a Salesforce record or opens a Linear ticket on a false positive becomes somebody’s backlog. Guard: run agents in report-only mode into a private Slack channel for two weeks and count the false-positive rate before granting write scopes. Channels 2.0 and AI Dashboards are still labelled beta — do not build a QBR dependency on them this quarter.
• A digital twin models your corpus, not your customer. If the recorded calls skew toward the accounts your CSMs enjoy talking to, the twin reproduces that bias fluently and without warning. Guard: check segment coverage of the underlying calls before quoting a twin to an exec, and require every twin answer to cite its source highlights.
• The headline reach metric is vendor-published. “Trusted by 40% of the Fortune 500” is Dovetail’s own claim, not an audited figure, and the platform rates 4.5 on G2 across 168 reviews on the vendor’s listing. Guard: ask for a reference call with a customer at your seat count and conversation volume, not at AWS’s.