Skip to content
MakerGain
← Back to project

The database crashed in his first week: Zipchat wants every store's support ready out of the box

Edited by MakerGain

Zipchat is an AI support agent for ecommerce. When Ruslan took over in January 2024, the team struggled to account for its MRR and customer sources, and the database started crashing in the first week. He first fixed the books, using Mantle to make MRR visible and collecting feedback on replies. His blog reports a 25% MRR increase about two months later, reaching US$20,000 for the first time. The product claims 30-minute setup, but brand rules, helpdesk connections and answer corrections still require store-by-store configuration.

In a blog post published on March 19, 2024, Ruslan wrote that when he took over Zipchat in January, the team did not know its MRR or where customers came from. Communication was scattered across WhatsApp, Slack and email. There was no shared knowledge base or automated testing. In his first week, the database started crashing blog[1].

It is not a typical AI startup story. But it answers a more practical question: when an online store hands support to AI, what does it actually buy, and whom does it pay?

It takes the question a customer asks in the middle of the night

The company describes Zipchat as a single ecommerce agent handling on-site search, product-page questions, website chat, WhatsApp, Instagram, Messenger and email, retaining the same memory of a customer across channels official description[2]. It answers from the brand's catalogue, website, documents and rules, prioritising live product data such as prices and stock over cached text.

Here is an illustration assembled from that description, not an actual customer record. Someone asks on a product page, “Can this arrive tomorrow?” The agent consults live stock and the brand's shipping rules. If those rules require a handoff at that point, it hands over. It can also take actions in the brand's tools, such as checking orders or sending discount codes. Agentic Skills connect those actions to any tool offering an API, webhook or MCP. The brand chooses which actions run automatically and which require human approval same source[2].

Image: official product website[3], captured on 2026-11-18, retaining the official interface illustration; not an independent test or a record of actual customer results.

“Hand off to a human” sounds simple, but the connections are uneven. In the official helpdesk list, Zendesk is native; Freshdesk is a ready-made Agentic Skill that opens a ticket with a summary and full transcript during handoff; Gorgias remains beta and available on request; Intercom and Salesforce Service Cloud require custom skills using Zipchat's own API. These connections are available on every plan same source[2]. The company also says self-service setup takes under 30 minutes and works after installation from the Shopify App Store. The box can be opened: the catalogue and website can be read automatically from day one. Brand rules, helpdesk connections and corrections to wrong answers still need the brand's configuration. The deliverable is not simply “a box that chats,” but a workflow in which the brand sets rules, connects support and corrects responses.

The person taking over first had to fix the accounts

Ruslan had built ecommerce tools before. In 2017, he built Checkout X, an alternative Shopify checkout, and bootstrapped it to $8M ARR. Shopify later changed its platform and locked down checkout, killing the product Indie Hackers interview[4]. That history does not explain everything about Zipchat today. But he had run a business where someone else controlled the door.

Public accounts of his relationship with Zipchat differ. The founders page lists Luca Borreani, Carlo Bellati, Ruslan Leteyski and Michael Monin as four cofounders who started the company in May 2023 About page[5]. In his blog, he calls the January transaction an “acquisition,” says he “inherited” Luca and Carlo, and reports that the business had operated for about 10 months and was serving 200+ business customers when he took over blog[1]. In the same Indie Hackers interview, he says he cofounded Zipchat in 2024 interview[4]. The blog was published on March 19, 2024 and dates the acquisition to January. What can be established is that a group of business customers was already being served when he took over, while the books were a mess.

His least obvious “growth move” was fixing those books: using Mantle to make MRR visible and starting to collect feedback on chat responses same source[1]. Revenue came next. He reports raising MRR by 25% around two months after the acquisition, reaching $20k MRR for the first time. That claim comes from the March 19, 2024 blog; $20k MRR roughly two months after acquisition is not a current figure.

He included the ugly parts too. The database started crashing in week one and still had performance problems when he wrote. Server spending was higher than it should have been, and he acknowledged that he was not a senior developer. Features were missing, chat answers were sometimes awful, and there were many bugs same source[1].

The bill is tied to how many replies the AI sends

The pricing reveals the trade-off. Zipchat's standard metered plans charge by monthly AI replies, rather than seats or store traffic. Starter costs an annual-billing equivalent of $64 a month ($768 a year), or $80 billed monthly, with 750 AI replies per month. Pro costs an annual-billing equivalent of $240 a month or $300 billed monthly, with 3,000 replies. Annual billing saves 20%. Starter and Pro can buy overage packs of 250 replies for $49 pricing[6]. Every plan has a 7-day free trial. A full refund to the original payment method may be requested within 30 days of the first payment, except in cases of suspected fraud or abuse. Cancellation is available at any time without a cancellation fee, but unused months of an annual plan are not refunded after that 30-day window same source[6].

Charging by reply ties the bill to work done by AI rather than the number of support staff hired. Gorgias also says it does not price by seat, but its unit is the ticket. Helpdesk plans scale from 50 to 5,000 monthly tickets. A ticket incurs a ticket charge once at least one message is sent from Gorgias helpdesk—whether by a person, the AI Agent or an automated rule. If the AI Agent resolves it without handoff and no human involvement is needed within 72 hours, an outcome-based automation charge is added. A fully AI-resolved ticket can incur both charges; a handed-off ticket incurs only the ticket charge Gorgias pricing[7] billing explanation[8]. Both charge against volume, but use different rulers. Zipchat counts each AI reply sent; Gorgias counts tickets with a message sent from its helpdesk and adds an automation charge for independent AI resolution. Store owners therefore have different calculations to make.

Another fee does not belong to Zipchat. For WhatsApp marketing, Meta charges the first message by country. In the US it is $0.02 per conversation initiation; the official example is $20 for 1,000 people, and applies only to marketing campaigns pricing[6]. Installations from the Shopify App Store are billed by Shopify and appear on the Shopify invoice; plan functionality does not depend on the subscription channel same source[6]. There is also a mismatch: the Shopify listing[9], displayed in renminbi, gives Starter 500 monthly replies, while the dollar-priced website gives it 750. The listing says its renminbi price is an estimate at the current exchange rate; actual charges use the rate on the billing date and the store's billing currency. The reply allowances do not match, and the materials do not specify which governs.

For results and revenue alike, check who is making the claim

The website reports outcomes such as “less than 3.2% escalated to humans” and revenue generated through conversations. These are company claims without independent testing or customer-level data to verify them. 2,500+ ecommerce brands, an 18-person team with 6 in engineering, 4 in marketing, 3 in customer success and 5 in sales, 40+ countries, and a 4.9/5 Shopify App Store rating are also official claims About page[5] official description[2].

Revenue has two definitions, both from the founder rather than audited accounts. In the visible portion of his Indie Hackers interview, he says Zipchat was handling 700k conversations per month, growing roughly 10% week over week and about to pass $2M ARR; the same page's revenue field says $167K a month interview[4]. The blog's $20k MRR two months after acquisition is a claim from another point in time. MRR is not profit. Server costs are described only as “higher than they should be,” without an amount, so profit remains unknown.

His own harshest observation was that he had thought Zipchat was merely a ChatGPT wrapper. Building AI chat that worked out of the box for any store was much harder, and that core job was still imperfect blog[1]. Acquisition work included App Store optimisation, marketing, outbound sales and conferences. Outbound worked, but there was not enough data to expand the team. Marketing produced results, but nothing was ready to receive more investment. The App Store brought installations while competition intensified. He said there was no good growth flywheel yet, and his target for that year's end was $60k MRR.

“Out of the box for any store” is easily underestimated. The company says setup takes 30 minutes, with catalogue and website reading automated from day one. Brand rules, shipping policies, helpdesk connections and corrections still need store-by-store configuration. What owners are more likely paying for is the right answer in the middle of the night, not the use of AI. Making that answer work for thousands of different stores is hard because of their differences, rather than the model.

Continue reading

AI support
Human handoffs are free: Fini puts its AI-support bill on “actually resolved”

Fini's AI support agent handles chat tickets inside a customer's existing helpdesk. Behavior trees and a deterministic executor read plan and account state node by node, then call customer system interfaces to perform actions such as canceling subscriptions. Human handoffs are not billed; by default, only end-to-end resolutions without human takeover count as billable tickets. The team cleared its first product and rebuilt after four weeks embedded with a fintech support team, choosing not to replace existing helpdesks and requiring human approval before publishing knowledge corrections proposed by background AI.

AI support
Teach the bot to admit uncertainty first: a side project bringing in about \$500 a month

Andy and his two cofounders at London's HeyDeacon built Deacon for founders and small teams still answering support themselves, starting with their own use. When it cannot answer, it admits that and asks visitors for an email address. Only after an address is actually provided is the conversation marked unread and the owner emailed; human replies can enter the knowledge base. The team focuses on the workflow around the model. Knowledge needs manual recrawling or publishing-script triggers to update, while changing models does not require rebuilding the product.

AI productivity
Klipy, the sales AI that waits for your approval

Klipy takes over follow-up after sales meetings. It captures customer conversations from email, meetings, and instant messaging, drafts follow-ups and CRM updates, and prepares for the next meeting. Follow-up messages go out only after the salesperson approves them. Founder Jung Hong Kim is a Korean serial entrepreneur who took on debt when COVID sank his retail venture. He met his cofounder while consulting and launched Klipy.ai in three months. It went live in November 2024 as an AI CRM that logged communications automatically. In June 2025, it shifted toward carrying out next steps for sales teams and tied billing to specific actions users approve.

AI development
Every small revision came with another fee. He moved app design into a chat box

Paying extra for every revision to a mobile app design led Diego Roshardt to build AppAlchemy. Users describe an app in chat, get React Native code and scan a QR code to preview it live on their phone. A change requested in chat updates the screen. Users can export the code, while the public materials leave it open who packages and publishes the app. Roshardt built the first version in about two weeks. He chose to build a history on Reddit, find relevant communities and share useful posts containing links instead of buying ads.