Kustomer 评论:自 2026 年中期以来,座位价格隐藏,AI 代理现在每次参与对话 0.75 美元。实际成本、出口限制以及 Zendesk 或 Gladly 何时更适合。
Until March 2026 you could price Kustomer from its website. The /pricing/ page captured by the Wayback Machine on March 10, 2026 listed Enterprise at $89 per seat per month and Ultimate at $139, with "a minimum of 8 seats, paid annually." By the July 22 capture both numbers were gone. Today the page says "Every CX team's setup is different, so we don't post one-size-fits-all pricing" and offers a "two-minute assessment" instead. It is an ROI calculator, not a trial.
The AI did not get hidden. It got dearer. The comprehensive pricing page, stamped "Last Updated: October 12, 2026", lists the customer-facing AI agent, now called Concierge, at $0.75 per engaged conversation, up from $0.60. The rep copilot, Envoy, is $50 per user per month, up from $40. Both are 25% rises. The usage-based plans launched in October 2024 are now labelled "Legacy". Every figure in this review was checked on Kustomer's own pages on November 21, 2026.
So the real questions are what it costs now that you have to ask, what the extra 15 cents buys, and how hard it is to leave.
Brad Birnbaum and Jeremy Suriel started Kustomer in New York on August 14, 2015, and the pitch has not moved since: a support platform built like a CRM. Each customer is one timeline holding every email, chat, SMS, call, WhatsApp, Instagram and Facebook message they have sent, next to their orders and account data. A rep sees last week's duplicate charge and this morning's angry DM in one scroll, instead of two tickets assigned to two people.
Facebook bought the company in a deal announced in November 2020 and reported at about $1 billion, which closed in February 2022. It did not last. In May 2023 Kustomer spun back out with $60 million from Battery, Redpoint and Boldstart, at a valuation CNBC put at $250 million, a quarter of what Meta paid. A $30 million Series B led by Norwest followed in August 2025. The homepage claims "600+ B2C & B2B companies". That is a small base next to Zendesk, which means fewer consultants and fewer forum threads from people who already hit your problem.
Concierge is the bot your customers talk to. It answers across channels using the same timeline your reps see, and Concierge 2.0 runs on what Kustomer calls an AI Reasoning Engine: language and intent models for the conversation, plus deterministic rules for steps where a guess is expensive, such as refunds and escalations. "Our reasoning engine allows businesses to toggle between adaptive and deterministic modes," says CTO Jeremy Suriel. That split is the right idea. A refund policy should not depend on a model's mood.
The billing unit deserves more attention than the feature list. In September 2024 Kustomer charged $0.75 per resolution. The October 2026 price is $0.75 per engaged conversation. Same number, different meter: you now pay when Concierge engages, whether or not it solves anything. If it hands half its conversations to a human, your real cost per resolution is $1.50. Gladly bills its AI per resolution instead (see our Gladly review). Ask Kustomer's sales team exactly what counts as "engaged" before you sign.
Envoy, at $50 per rep per month, sits beside the human agent with suggested replies, conversation summaries and next best actions. Architect, a natural-language builder for AI agents and workflows, is included on both current plans, with a claimed three hours to a first deployed agent. Smaller meters stack on top. AI QA is $0.05 per execution and Conversation Intelligence $0.02, voice starts at $0.02 a minute, outbound messages are $0.025 each, and WhatsApp adds 20% on top of Meta's own per-message fee. HIPAA compliance is $25 per user per month.
Two things the pricing page leaves out. Kustomer's help center lists OpenAI's GPT models and Amazon Bedrock models, including Anthropic's Claude and Amazon Nova, but says it "does not publish a fixed mapping between individual AI models and specific product features." You cannot know which model is answering your customers, or notice when it changes. On training, the AI compliance page says customer data is not used to train global models, except where a capability is tailored to one customer, in which case it may refine models for that customer alone. OpenAI runs under zero data retention, and AWS offers an opt-out.
With no public seat price, any total is an estimate. This one is our arithmetic, using the last published Enterprise price of $89 per seat and the October 2026 add-on rates, for a 10-rep team whose Concierge engages 3,000 conversations a month.
| Line | Unit price | Per year |
|---|---|---|
| Enterprise seats x 10 | $89/seat/month (March 2026 list price) | $10,680 |
| Envoy copilot x 10 | $50/user/month | $6,000 |
| Concierge, 3,000 engaged conversations a month | $0.75 each | $27,000 |
| Total | about $43,680 |
The AI agent is the biggest line, bigger than seats and copilot together. At the old $0.60 rate the same Concierge volume cost $21,600, so October's change adds $5,400 a year to this team's bill with no new line on the invoice. Ten seats is not far above the floor, either. At the 8-seat minimum the seats alone came to $8,544 a year before any AI. Implementation, Premier Success and extra API capacity appear on the pricing page as a "$" with no number.
The most interesting move of 2026 is not a price. On March 10 Kustomer started selling its AI as a standalone platform: a team on Zendesk can switch on Kustomer AI as "a fully integrated intelligence layer" without moving a single ticket. The announcement says Salesforce and other integrations are "launching soon", and the overlay is sold to enterprise customers only.
This is the cheapest way to learn whether Kustomer's AI beats Zendesk's own, because a failed pilot leaves your help desk untouched. The March blog post offered Zendesk customers a free 30-day trial; that page now redirects to the demo form, and so does /free-trial/. There is no self-serve trial of anything. Every route into Kustomer goes through a sales call.
Leaving is where Kustomer bites. Its own help article on exports says CSV exports "include conversation-level attributes and message previews only. They do not include full message content." The full transcripts, the whole point of a timeline, come out only through the Messages API. Report exports run 30 days at a time with at most 25 attributes. The Export Buddy app covers users, teams, snippets, shortcuts, tags and knowledge base articles, but not your conversation history.
The API caps how fast that history leaves. Enterprise gets 1,000 requests per minute and Ultimate 2,000, while machine users doing search have their own limit of 100. The standard Search API only returns records updated in the past two years; anything older needs the archive endpoint. A team with five years of history should budget engineering days, not an afternoon, to get out.
Reporting data is kept for two years on current plans. Support runs 8 AM to 8 PM Eastern, Monday to Friday, with 24/7 on-call only for critical issues. The help center's response targets are 2 hours for critical, 1 day for high and 3 days for medium, and Ultimate's SLA tightens high priority to 8 hours. Uptime is promised at 99.9% a month, with service credits capped at 50% of the fee.
Kustomer rates about 4.4 out of 5 on G2, and the complaints repeat across G2, Capterra and Software Advice. The app is slow to load. It takes a while to learn. Search is weak, the built-in reports fall short, and customising workflows past a point needs someone who knows JavaScript. Reviewers on small teams name the 8-seat minimum as the reason it felt expensive. Slow loading hurts more than usual in a product that asks reps to live inside one timeline all day.
Buy Kustomer if you are a B2C brand with at least eight reps, customers who come back again and again across several channels, and a support lead who thinks in customer histories rather than ticket counts. The timeline is still the best reason to choose it. If you already run Zendesk and only want better AI, the overlay lets you test that without a migration.
Skip it if you have fewer than eight seats: Help Scout and Freshdesk publish their prices and start far lower. A small Shopify store will get more from Gorgias. If you want AI billed only when it actually resolves something, look at Gladly, its closest customer-centric rival. And if you need a number before a sales call, Zendesk still prints one.
Everyone else should walk into the demo with three questions written down: the seat price at your headcount, what counts as an "engaged" conversation, and how full transcripts get out if you leave. Kustomer no longer answers the first one on its website.