Kustomer

What it is

Kustomer is a CX platform built around one continuous thread per customer instead of a ticket queue. Email, chat, SMS, voice, WhatsApp, Instagram, Facebook and TikTok Shop land on the same timeline alongside order events and custom objects, so an agent opens a person rather than a ticket. That data model is the product; the rest is built on top of it.

What Kustomer sells in 2026 is not what this entry described a year ago. On 14 October 2025 it relaunched as an AI-native platform, and the AI layer now ships as four named products: Concierge (customer-facing AI agents), Envoy (real-time assist for human agents), Architect (the studio where teams design, test and evaluate agents) and Data Explorer (AI analytics). The March 2026 release added Procedures — deterministic branching inside a generative flow, so a refund rule is enforced as logic instead of coaxed out of prompt wording — and an Evaluations Assistant that generates and runs regression tests against agents from the Test screen.

Ownership, since it still gets asked: Meta acquired Kustomer for roughly $1B (closed February 2022), then spun it back out in 2023 at a reported ~$250M valuation. It is independent today, led by co-founder Brad Birnbaum, and raised a $30M Series B led by Norwest in August 2025.

Why it shows up in Customer Success stacks

Timeline, not tickets. For a support org whose recurring pain is “which of these six tickets is the same person,” the data model is the entire argument. Channel context does not fragment, because there is only one thread.

Agent access via MCP, read-only. Kustomer runs a hosted MCP server at https://server.mcp.kustomerapp.com/mcp, authorized by OAuth and listed on the pricing page as included in the package. Roughly 40 tools span search, search-conversations, get-customer-by-id, get-messages-by-conversation, search-kobjects, get-workflow-revisions and the get-*-search-fields helpers, plus ping for connection checks. There is no create, update or delete tool in the set. Setup is two-sided: an admin installs the Kustomer MCP server app from the in-product Application Store, then a workspace admin registers it in Claude or ChatGPT.

Read that scope before you budget against it. You can ask Claude why a conversation escalated and get the SLA data, tags, sentiment and full message thread back. You cannot have it write the reply or change the status.

Transactional AI deflection. Concierge targets retail and DTC contact mixes — order status, returns, subscription changes — where a large share of volume resolves without a human. Kustomer’s published customer results: 70% of inbound chat fully automated, 45% handle-time reduction, and Aplazo at 98% of WhatsApp conversations AI-handled.

Pricing

Kustomer no longer publishes a price. kustomer.com/pricing is now an ROI calculator: “Every CX team’s setup is different, so we don’t post one-size-fits-all pricing.” You answer industry, company size, agent count and monthly conversation volume, get a directional savings estimate, and then talk to sales. The feature table on that page describes a single package at what used to be Ultimate-level limits — 2,000 RPM API, 300 brands, 20 custom permission sets, all languages, 100M custom objects.

The vendor-published rates that do survive sit on the comprehensive-pricing-details page, last updated 19 December 2025:

• AI Agents for Customers — $0.60 per engaged conversation
• AI Agents for Reps — $40 per user/month
• HIPAA-enabled subscription — $25 per user/month
• Voice — from $0.02/minute; SMS, transcription and WhatsApp are pay-as-you-go
• WhatsApp — Meta’s per-template message fee plus a Kustomer markup of 20% of that fee
• Data storage — $50 per GB per month

Seat pricing is absent from all of it. The $89 (Enterprise) and $139 (Ultimate) per-seat figures still repeated across 2026 buying guides are the vendor’s own withdrawn rate card, and that same page now carries two further generations marked legacy — seat-based Professional/Business, and usage-based ProfessionalAI/EnterpriseAI/UltimateAI — all priced “Talk to Sales.” Three packaging generations in the open is the real signal: what you get quoted depends on when you bought.

For an anchor, use contract data instead of list price. Vendr’s marketplace dataset (31 purchases, last updated February 2026) puts the median Kustomer contract at $138,073/year, in a $60,588-$441,342 range, with 15-30% off list on multi-year commitments and 25-35% in competitive evaluations. Third-party guides put implementation at $18,000-$30,000 over 12-16 weeks.

Best for

A support or CX leader at a retail, DTC or fintech company running 50,000+ monthly conversations across chat, WhatsApp and voice, where the contact mix is transactional enough that Concierge closes a real share of it, and where the single-thread model replaces a ticket tool that keeps splitting the same customer across records. The scoped win: an omnichannel consumer brand that wants one customer record read by both its human agents and, through MCP, whatever assistant its ops team already works in.

Not for

Teams under about 15 agents, and any team that needs a number before a sales call. The first was already true under the old 8-seat minimum; withdrawing published pricing makes the second one worse. Below 15 agents, Help Scout, Front or Intercom at $39-139/seat/month put a working figure in front of you the same afternoon.

Versus the alternatives

• Zendesk — the incumbent and the safe procurement answer. Pick it when you need published seat prices ($19 Support Team through $115 Suite Professional), the deepest third-party app marketplace, and an MCP server of its own. Pick Kustomer when the ticket model is the thing that is breaking and marketplace depth is not the constraint.
• Intercom — pick it when the deliverable is resolution rate on inbound rather than a customer record. Fin bills at roughly $0.99 per resolution against Kustomer’s $0.60 per engaged conversation, and that difference is the argument: Intercom charges when the bot closes the issue, Kustomer charges when it engages. On a contact mix with heavy escalation, Kustomer’s meter runs on the failures too.
• Decagon — the fastest-growing entrant in AI support. Pick it when the AI agent is the purchase and you are keeping the help desk underneath it; a ~$50,000 platform fee plus ~$0.50/resolution costs less than replatforming. Pick Kustomer when you are replacing the platform rather than bolting an agent onto it.
• Gladly — the closest structural analog, same people-not-tickets thesis. Its roughly $180-210/agent/month entry is about double Kustomer’s withdrawn seat rate, so agent headcount decides this one on price.

Watch-outs

• You cannot price this without a sales cycle, and every anchor in circulation is stale. The $89 and $139 figures in 2026 buying guides are a withdrawn rate card, not a discount target. Guard: run the ROI calculator to get segmented, then demand the quote broken into seats, AI meter, voice/SMS/WhatsApp usage, storage and implementation as separate lines. Negotiate against Vendr’s $138,073 median, not the blog figures.
• AI is metered on engagement, not resolution. At $0.60 per engaged conversation, a bot that touches a contact and escalates still bills. Guard: model your escalation rate against your real contact mix before signing, and ask for a resolution-based meter or a capped engaged-conversation pool in writing. If the vendor will not cap it, price the worst case at 100% engagement.
• The MCP server reads; it does not write. Buying Kustomer expecting an assistant to update conversations ends in a build project against the REST API at 2,000 RPM. Guard: put the exact tool names into the trial plan — search-conversations, get-conversation-by-id, get-messages-by-conversation — and test the question you intend to ask in production. Anything that mutates a record is API work; scope it before signature.
• OAuth authorization inherits the authorizing user’s visibility. The connector sees what that account sees, which becomes a data-exposure question the moment an admin authorizes it for convenience. Guard: authorize from a purpose-built account carrying the narrowest permission set that satisfies the use case, and call get-current-user to confirm the scope before rollout. Review it on the same cadence as agent permission changes.
• Three live packaging generations make your renewal a repackaging event. Customers on legacy seat-based or legacy usage-based plans get moved onto the current package. Guard: get the migration path and the current package’s price in writing at signature rather than at renewal, and ask in the first call which plan generation your quote sits on.

Related: Gladly vs Kustomer works the closest head-to-head, and the AI support agent stack prices the layers around it.