Skip to content
MakerGain
← Back to project

When a Visitor Clicks Handoff to a Human, Chat Data Sends a Colleague an Email with a Login-Free Link

Chat Data trains company documents, webpages, and files into an AI customer-service agent that can be deployed to websites and messaging channels, and uses a login-free email link to connect visitors who ask for a human. It charges by message credit and supports custom models and MCP hosted by Composio; the TrustMRR page lists MRR of 17,645 US dollars and 304 active subscriptions, with a team of 2–5 people.

The most awkward moment for a customer-service bot is not giving a wrong answer; it is getting stuck and not responding. Chat Data makes a very small, very concrete arrangement at this step: when a visitor clicks handoff to a human in the chat box, a colleague's inbox receives an email with a unique link. No need to log in to the dashboard; open it and you can pick up the conversation in real time. If the visitor has already left but left contact details, new messages can still be followed up by email Chat Data Cookbook - Docs[1].

This detail basically explains what Chat Data is: it trains a company's own documents, webpages, and files into an AI agent, deploys it to its website, WhatsApp, Shopify, and other channels, lets visitors ask questions, leave leads, and click handoff to a human on the site or in messaging apps, and uses MCP to execute actions in more than 800 apps Chat Data - $18,682 last 30 days | TrustMRR[2]. What it sells is not 'AI can do everything,' but 'when it can't, someone can pick up.'

Image: Source Chat Data official image[3], retrieved 2026-11-11, original interface preserved.

Training Usually Takes About Five Minutes; When There Is a Conflict, Follow the Q&A

When creating an agent, the default selection is Train on your data. After adding files, webpages, text, or Q&A and training, it usually takes about five minutes. The documentation contains a sentence that is easy to skip: structured Q&A carries more weight than unstructured text, and when the two conflict, structured data is followed Chat Data Cookbook - Docs[1].

This sentence points to a product decision: when the two clash, it does not let the model decide on its own which material is newer or which passage is more authoritative; instead, it hands 'who decides' back to the person who uploaded the material. For example, if a PDF says 30-day returns and the Q&A says 7 days, the system answers according to the Q&A—this is a ranking rule for conflicts, not a guarantee that the model will not make mistakes. For companies handing customer service to a bot, at least there is a checkable rule for whom to listen to when there is a conflict.

Plug In Your Own Model; What You Pay For Is Remembering the Conversation

Besides the platform's built-in models, customers have another integration path: custom-model. Connect your own backend; the backend receives requests and returns plain text. This mode does not consume the platform's OpenAI tokens. Each chat costs 0.1 message credit, and the official explanation says this fee corresponds to storage of conversations and daily metrics Integrate Endpoints With Chatbots - Docs[4]. Ordinary conversations deduct message credits based on model cost and content volume: lightweight models start at 0.5 credit, and Claude 3 Opus is 20 credits Chat Data Pricing | AI Chatbot Cost & Free Plan[5].

This 0.1 credit corresponds to storage of conversations and daily metrics; it does not include model calls. Among comparable products, DocsBot supports BYOK in its available plans: customers pay model fees directly to the model provider, while DocsBot still handles training sources, the chat experience, deployment, access control, and workflows AI Credits Are Here So DocsBot Can Grow With You[6]. Both are called 'connecting your own model,' but the delivery boundaries are different and cannot be compared as the same thing.

MCP is another layer of action: it lets the bot call external tools during a chat. The official documentation says it supports 834 built-in apps and custom remote servers. These integrations are hosted by Composio; customers click connect and complete OAuth authorization; credentials are not stored directly by Chat Data and can be revoked at any time. Each MCP tool enabled takes up one AI action quota by default; if it is placed in an AI Agent node, it is counted against that node's own tool limit and does not take up the plan quota MCP Integration | Chat Data - Docs[7]. It is not the whole of 'actions'—web search, buttons, lead collection, sending files and images, and calling your own API are all AI Actions; MCP is only the path for bringing in third-party tools.

The Math Is Clear, but Not Easy

message credit is deducted based on 'model base cost × content units': the number of input characters divided by 8,000 and the number of files divided by 3, each rounded up, then the larger value is taken, with add-ons such as reranking, medical knowledge base, and extra-long user information added as needed. According to the official formula, for 8,000 characters of input, no attachments, and no add-ons, using a lightweight model with a base cost of 0.5 credit, this call deducts 0.5 credit; with Claude 3 Opus it is 20 credits Chat Data Pricing | AI Chatbot Cost & Free Plan[5]. The math is transparent, but for people who do not understand character counts it is still a page of rules they cannot understand—this is my most direct criticism of it.

The free plan at $0 provides 100 message credits per month, 1 agent, and a capacity of 400,000 characters per chatbot; conversation history is retained for half a year, and chatbots that have been idle for 60 days are deleted. The 60-day rule is written in the pricing table, not hidden later; it targets chatbots that no one has attended to for a long time, not the deletion of all free accounts once they hit 60 days Chat Data Pricing | AI Chatbot Cost & Free Plan[5].

Prices come in two sets of terms that must be kept separate: the official website's annual-billing display is Entry at $15.83/month equivalent, Standard at $82.50/month, and Professional at $332.50/month (annual billing includes two months free, retrieved 2026-11-14); TrustMRR lists monthly list prices of Entry at $19/month, Standard at $99/month, and Professional at $399/month, with Enterprise custom Chat Data Pricing | AI Chatbot Cost & Free Plan[5]Chat Data - $18,682 last 30 days | TrustMRR[2]. The annual-billing monthly average is a display price, not the amount charged each month.

Standard and above include a free HIPAA BAA, but HIPAA mode only supports OpenAI and Anthropic models; it requires enabling permission gates, member 2FA, logging PHI access, and blocking channels without a BAA, and data is not deleted and will be retained Chat Data Pricing | AI Chatbot Cost & Free Plan[5]. Compliance here is not a slogan; it is a string of switches that change daily operations.

Revenue, Team, and Hosted Integrations

The TrustMRR page (updated through October 31, 2026, and stating that revenue is verified via the Stripe API) has the headline 'last 30 days $18,682'; the body's MRR is $17,645, historical total revenue is $243,196, and there are 304 active subscriptions Chat Data - $18,682 last 30 days | TrustMRR[2]. The two figures cover different time windows and cannot be added together or exchanged; MRR is a subscription metric, not profit—none of the costs such as model calls, hosting, communications, or labor have been disclosed, so whether this business actually makes money cannot be seen from the outside. 304 subscriptions likewise do not equal 304 customers.

About the people making the product, the public page leaves only this: founder Yongrui Su, team of 2–5 people, VC-backed, started in 2023, self-serve SaaS, listed in the Shopify App Store Chat Data - $18,682 last 30 days | TrustMRR[2]. The choices left in the product say more about why it turned out this way than the résumé does.

A company of 2 to 5 people can put out 834 MCP apps because it delegates integration hosting to Composio; customers each complete OAuth authorization, and Chat Data does not directly store credentials—this is leveraging outside help. Hosting reduces the work of building integrations in-house, but service maintenance does not fall to zero because of it. Yet when something actually goes wrong, customers have to find the person who should click 'revoke'; PII anonymization only handles the segment before it is sent to the model, and stored conversation records are not anonymized Chat Data Cookbook - Docs[1]. These boundaries are written by the company itself, but writing them clearly does not mean it can backstop them. Among peers, Chatbase's external narrative emphasizes a unified entry point for chat, email, and voice; its website shows order lookup, ticket creation, and human handoff at the same time, and it says it serves 10,000+ brands Customer-facing AI Agent for Support and Sales | Chatbase[8]. Both companies list actions and human takeover as part of the delivery; they just emphasize different things in their external narratives.

Chat Data has turned 'someone picks up when it cannot answer' into a clear product path; but after the handoff, authorization revocation, how stored conversations are handled, and the deletion of free-plan chatbots after 60 days of idleness remain in customers' hands. What it sells is the moment of connection; whether it can keep holding on is written clearly in the terms, but it does not backstop customers.

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 search
When AI does not mention your company, he publishes articles on the customer's own site

MyZia helps brands get mentioned in AI answers: it scans visibility, finds questions where competitors are cited but the brand is absent, then writes articles for the customer’s own website. Founder Dhairya Patel also runs a development agency. He did not stop at a dashboard that offers advice. He chose end-to-end automation. Keeping articles on customers’ domains also requires connecting each CMS, making publishing the first part to hit a scaling barrier.

AI commerce
He reported a 15% close rate, but customers left after three months. He rebuilt his AI sales agent to screen out buyers who are a poor fit

Sohaib Ahmad’s DM Champ is an AI sales agent working across WhatsApp, Instagram, and other channels. It helps agencies and businesses handle incoming leads: restart conversations with old lists, assess potential, arrange the next step, and hand back to a person when stuck. He reported a 15% close rate, yet customers left after about three months. He attributed this to AI persuading people who did not truly need the product, then shifted optimization from conversion to customer fit and rebuilt the system to proactively screen out unsuitable buyers. Retention data after the rebuild have not been published.