Growing from a thousand users to a hundred thousand does not just multiply your support volume. It changes what customers ask, who is answering, and how fast your documentation goes out of date. An AI agent helps most with one part of that shift, the repeated questions that have written answers, and much less with the rest. This post walks through what tends to change at each stage, what to automate first, how to size a LayBuild plan, and which work still needs people no matter how big you get.
Why support does not scale in a straight line
Two things change as you grow, and they pull in opposite directions.
Volume goes up, and repetition goes up with it. With a small user base, most questions are close to unique: an early adopter found an edge case, or wants a feature explained. With a large one, the same questions arrive many times a day. Repetition is what makes automation worthwhile, because one good answer covers many conversations.
The long tail also gets longer. More users means more integrations, more plans, more countries and more unusual setups. The number of distinct questions keeps growing even as the top questions dominate the count. An AI agent that only answers from your content will handle the top of that curve well and hit gaps in the tail, which is where your people will spend most of their time.
Early users: founders answer, and every question is a product signal
At the small end, the people building the product usually answer support themselves. That is worth keeping for as long as you can, because the questions are the clearest feedback you will get about what is confusing.
What to do at this stage:
- Write down every answer you give twice. That becomes your first knowledge base.
- Keep a short list of topics that must always go to a person: billing disputes, data deletion requests, anything legal.
- If you add an AI agent now, treat it as a way to handle the questions you have already answered in writing, and read its transcripts. Its "I do not have specific information about that" replies will show you which answers you have not written yet.
Growth: the same questions, many times a day
This is where an AI agent pays for itself in time, and where most teams make their first dedicated support hire.
The mix shifts towards how-to and policy questions: setup steps, plan differences, refund rules, shipping times, password resets. These have stable answers and are exactly what retrieval over your documentation handles well.
Two problems appear at the same time. Documentation starts to lag behind the product, because features ship faster than help articles get updated. And the first support hire inherits a queue that is mostly repetition, which is the fastest way to wear someone out.
What to automate first, in order:
- Questions that are asked often, have one correct answer, and are written down. Start here. LayBuild's Knowledge Q&A page ranks the questions customers ask most often, with a button to turn one into a Q&A pair.
- Policy questions where the policy is stable: refund windows, supported countries, plan limits. Pin the documents that state these so they are always in the prompt, rather than depending on retrieval to find them.
- Setup and how-to questions, once the relevant help articles are current.
What not to automate yet: anything that needs to look at a specific customer's account. LayBuild's API tools call an HTTP endpoint you configure, with the default values you set, after an answer is generated. They do not extract an order number from the customer's message and pass it to your API, so "where is my order #1234" is not something the AI can answer from live data. Route those questions to a person, or give customers a self-serve order status page and document it.
Larger scale: incidents, languages and many products
Past a certain size, the shape of the work changes again.
Incidents cause spikes. When something breaks, many customers ask the same question within minutes. The fastest response is a short status note in a pinned document ("Payments are delayed; we are working on it; no action needed from you"), which goes into every prompt, so the AI can answer from it without depending on retrieval, until you remove it. Pinned documents are always injected into the prompt, capped at 4,000 characters per document and 12,000 characters in total, so keep incident notes short and remove them when the incident ends.
Languages start to matter. LayBuild tells the model to reply in the customer's language, but the default embedding model and the full-text search are English. A question in Spanish against English documentation retrieves less reliably than the same question in English. If a large share of your users write in another language, test retrieval in that language before you rely on it.
You may run several products or audiences. Different AI agents with different knowledge and prompts keep a billing assistant from answering onboarding questions with the wrong context. Plan limits cap how many agents you can run (see below).
Channels multiply. LayBuild supports the web widget, WhatsApp through the Meta Cloud API, and the REST API. It does not connect to email, SMS, Slack or voice. If your customers expect email support, that stays in your helpdesk.
Sizing a LayBuild plan as you grow
Our plans have a flat monthly price and a monthly allowance of conversations. A conversation is one chat thread started in the billing period, however many messages it contains. There is no per-resolution fee.
| Plan | Price (monthly billing) | Conversations / month | AI agents | Team seats | Documents | URLs |
|---|---|---|---|---|---|---|
| Starter | ₹1,499 | 1,500 | 5 | 5 | 10 | 5 |
| Pro | ₹3,999 | 8,000 | 50 | 15 | 25 | 25 |
| Premium | ₹9,999 | 20,000 | 125 | 100 | 50 | 50 |
Annual billing lowers the monthly price to ₹1,199, ₹3,199 and ₹7,999. Current details are on the pricing page.
To pick a plan, you need one number of your own: how many conversations your users start in a month. Here is an illustration with made-up inputs. Suppose 30,000 people use your product in a month and 4% of them open a chat. That is 1,200 conversations, which fits Starter. At 100,000 monthly users with the same rate, you would expect 4,000 conversations, which needs Pro. Your chat rate is the input that matters, and it varies with your product, your audience and how visible the widget is, so measure it rather than borrowing someone else's.
Size for your busiest month, not your average one. A launch, a pricing change or an outage can push a month well above normal, and the usage view in your account shows conversations used against your allowance. If you expect to go beyond Premium's allowance, talk to us before it happens.
Two other limits bite before conversations do for some teams. The document and URL counts cap how many knowledge sources you can add, so consolidate small help articles into fewer, well-structured documents. And seats cap how many people can answer handoffs in the dashboard.
Where you still need people
However well the AI performs, some work stays human at every size:
- Actions on accounts: refunds, exceptions, ownership changes, disputes. The AI can explain the policy; a person should apply it.
- Questions your content does not answer yet, and the work of writing those answers down.
- Incidents, where customers need judgment and honest updates, not only a pinned note.
- Customers who ask for a person. Honor that every time.
- Reviewing what the AI says. LayBuild's learning loop publishes some AI answers back into the knowledge base automatically, without approval, and retracts them if the customer rates the conversation 2 or lower. At scale, someone should read a sample of what it learned each week.
A reasonable goal is not a support team that stops growing. It is a team whose size tracks the hard, account-specific and new work, while the repeated questions stop consuming their day.
Next steps
- For the method behind "automate the top questions first", read a playbook for raising the share of questions your AI answers correctly.
- If your spikes come from sales events, preparing an AI support agent for a sale covers pinned sale terms and staffing handoffs.
- To estimate what support costs you today, see how to calculate AI customer support ROI.