Unwrap

What it is

Unwrap reads the customer feedback a company already collects and groups it into themes nobody had to define in advance. It ingests support tickets and chats from Zendesk, Intercom, Freshdesk, Kustomer, Gladly, Gorgias, Front and Help Scout; call transcripts from Gong, Zoom, Aircall and Talkdesk; reviews and social posts from Trustpilot, Amazon, Google Maps, Discord, X and Instagram; and product signals from Salesforce, HubSpot, Jira, GitHub Issues, Pendo and UserVoice. Snowflake, BigQuery and S3 cover the warehouse, and Shopify and Mixpanel attach order and usage metadata so a theme can be cut by customer segment.

The product is a handful of named modules on one corpus. Auto Tagger classifies every entry against a taxonomy the model builds and keeps current. Assistant, rebuilt on Claude in March 2026, answers plain-English questions with verbatim evidence and can now construct taxonomy groups from a description. Alerts push emerging themes to Slack and email, and Responder drafts replies at volume. SupportIQ is the newer piece and the one that changes who buys it: it reads whole support conversations, extracts the actions agents took, groups similar cases and shows which resolution paths actually closed the issue.

The company is based in Santa Barbara, founded by Ryan Millner (CEO) and Ashwin Singhania (CTO), both formerly on Amazon Alexa’s question-answering teams. It raised a $12M Series A led by Scale Venture Partners in January 2025, with Atlassian Ventures, ScOp VC and the AI2 Incubator participating. The public customer list runs heavily consumer: Lululemon, Deckers, Oura, WHOOP, jetBlue, Southwest, Lyft and DoorDash, alongside GitHub, Microsoft, Perplexity, Clay and Adyen.

Why it shows up in Customer Success stacks

• SupportIQ turns VoC into a support-operations tool. Most feedback-analytics products tell you what customers complain about. SupportIQ tells a support leader which agent behavior resolves it — the most common resolution paths across similar cases and whether the customer acknowledged the fix. That is the input a coaching program and a macro library both need, and it is why the 15 September 2026 partnership with Solidroad matters: SupportIQ finds the recurring issue and the winning resolution pattern, Solidroad turns it into a training simulation for human and AI agents.

• The MCP server makes the corpus reachable from where analysis already happens. Shipped in March 2026, it exposes more than a dozen read-only tools, including search_actionable_groups, sample_feedback_entries, classify_feedback_entries, plot_chart and discover_filter_schema, at https://nlp.api.production.unwrap.ai/mcp. Auth is OAuth 2.0 per user, or static OAuth credentials an admin generates for automations. Documented clients are Claude (web, desktop and Code), ChatGPT, Codex, Cursor, Gemini CLI and GitHub Copilot. “What are the top complaints from enterprise customers this month” becomes a question asked inside the assistant the team already has open.

• Consumer channels are first-class, not bolted on. Amazon reviews, Gorgias, Gladly, Shopify metadata and Instagram are native connectors. For a DTC or marketplace brand, that is the difference between a theme model trained on the channels your customers actually use and one trained on tickets alone.

Pricing reality

Unwrap publishes a floor: from $24,000/yr, billed annually, with a 30-day trial run on your own data. There are no seat charges — the price moves with monthly feedback volume and the number of integrations connected. Vendr’s marketplace data, from six purchases, puts the median contract at $27,600/yr with a range of roughly $15,000 to $47,500. That sits under Enterpret’s $36,302 Vendr median and far under a Qualtrics or Medallia enterprise program. The honest mid-market band is $24–35K/yr; budget toward the top of it if you plan to connect calls and warehouse data in year one.

Best for

Product, CX and support-operations leaders at consumer, travel and marketplace companies handling roughly 5,000 or more feedback items a month across support, reviews and social, who want both “what is breaking” and “which resolution works” from one system. The fit is sharpest when a support leader, not only a product manager, owns the budget — SupportIQ is the part Enterpret and Chattermill do not match.

Do not buy it if

Your feedback lives in one helpdesk and arrives at under about 1,000 items a month: native Zendesk or Intercom intelligence plus a monthly export into a Claude project answers the same questions for close to nothing, and $24K is a meaningful slice of an analyst’s cost. Skip it, too, if you need to collect feedback — Unwrap has a Research Suite, but it is no substitute for Qualtrics when survey design, panel management and program governance are the job. And if your problem is coaching on QA scores rather than finding issues, start with a QA tool and add Unwrap later.

Versus the alternatives

• Enterpret — the closest AI-native peer and the stronger pick for B2B SaaS. Its Customer Context Graph ties feedback to account revenue, and its Agent Automations act on signals. Pick Enterpret when theme-to-ARR sizing drives the roadmap; pick Unwrap when consumer channels and support-resolution analysis matter more, or when the budget sits below Enterpret’s median.
• Qualtrics — the largest install base in experience management, now carrying Press Ganey Forsta after closing that $6.75B deal in May 2026. Pick it when survey collection and enterprise program governance belong on one contract.
• Medallia — the other incumbent by install base, under creditor control since its April 2026 debt-for-equity restructuring. Pick it for contact-center speech analytics at enterprise scale, with a written roadmap commitment before a multi-year term.
• Chattermill — the AI-native rival for retail and consumer CX, stronger on cross-brand benchmarking. Pick it when competitive benchmarking is the deliverable.
• Dovetail — not a substitute but a neighbor: it owns interview and research synthesis. Teams that run both send Unwrap’s high-volume themes into Dovetail as hypotheses for the next study.

Watch-outs

• Theme quality depends on volume, and small corpora disappoint. Reviewers and analysts note the model needs real feedback volume to cluster well. Guard: use the 30-day trial on your three highest-volume sources, and hand-label 200 entries before kickoff; if Auto Tagger disagrees with your labels on more than about 10% of them, rescope the sources rather than sign.
• Metering on volume and integrations means expansion is a pricing event. Guard: get the volume band, the per-integration cost and the overage rate on the first order form, and connect a new channel only when a named, unanswered question requires it.
• The MCP tools are read-only, but a sample persists. sample_feedback_entries and classify_feedback_entries create named samples that live server-side. Guard: use per-user OAuth for analysts so every query is attributable, keep admin-generated static credentials to service accounts, and review persisted samples quarterly for anything containing personal data the PII redaction missed.
• SupportIQ’s resolution patterns describe what worked, not what is policy-compliant. A fast resolution path can be a refund nobody approved. Guard: route SupportIQ’s top resolution paths through the support policy owner before they become macros or Solidroad training scenarios.
• The headline outcomes are vendor-published. Unwrap cites a 25% productivity increase on GitHub Copilot and review time dropping from 15–20 hours a month to 1–2. G2 shows 4.8 out of 5 on roughly 26 reviews — a strong score on a thin base. Guard: ask for a reference at your feedback volume and in your vertical before trusting the case-study numbers.