Their brand, not ours
Colours, avatar, header copy, launcher position, size and animation are all per assistant. On Pro the watermark comes off, so what the client shows their board is theirs.
Customisation optionsCrawl the client site, put the widget in their brand, hand them a read-only seat and keep the account. The second client takes twenty minutes, because the process is the same every time.
Written out because the value of this as an agency offer is that it is repeatable, not that it is clever.
Paste the domain. CustomerBot fetches the reachable pages, shows you the list, and you tick the ones worth training on — services, pricing, FAQ, contact. Skip the blog archive if it will only add noise.
Bulk URL training with a link picker, so you are never training on 400 pages by accident.
Every client has answers that live in someone’s head: lead times, the areas they actually serve, what they will not quote for. Add those as Q&A pairs so the wording is exactly what the client would say.
Q&A pairs are stored verbatim and retrieved like any other passage.
Primary, secondary and background colours, light or dark, position, size, launch animation, avatar, header title and subtitle, placeholder text, welcome line and the proactive bubbles that fire after a delay you set.
On Pro the CustomerBot watermark comes off entirely.
A booking availability endpoint, a stock lookup, a CRM write. Describe it in a sentence and CustomerBot generates the tool schema; the assistant calls it mid-conversation when the question needs it.
GET, POST and PUT with custom headers and body templates.
Invite the client as a viewer so they can read transcripts and leads, or as an editor if they want to fix answers themselves. You keep owner, which means you keep billing and settings.
Roles are per assistant, so client A never sees client B.
Colours, avatar, header copy, launcher position, size and animation are all per assistant. On Pro the watermark comes off, so what the client shows their board is theirs.
Customisation optionsOwner, admin, editor, viewer — granted per assistant. Give the client the seat that matches how involved they actually want to be, and keep the account yourself.
How roles workAnalytics show conversations, resolution rate and the questions with no answer behind them. That list is next month’s content brief, written by the client’s own visitors.
See the metricsGathered from what agency owners actually say about doing this, rather than from what it would suit us to claim.
Agencies reselling this kind of software routinely make more than the vendors do, and it is not because of a markup on a subscription. It is because somebody has to decide which forty pages to train on, write the Q&A pairs in the client’s voice, test the refusals and read the transcripts in month one. That work is the product you are selling.
Nobody is briefing you for an AI assistant. They have heard the noise and have no idea what it would do for them specifically. Which means this is a proactive pitch: build one on their own site content before the meeting, show it answering their own returns policy, and the conversation is about scope rather than concept.
Crawl the prospect’s public site on the free plan and ask it three questions they get asked every week. It takes ten minutes and it converts far better than a deck, because it is unmistakably about them.
Four components. Only three of them should carry margin.
The crawl, the source curation, the Q&A pairs, the branding, the refusal testing. This is the afternoon of real work and it should be priced as such.
Reading transcripts, adding answers for the questions it missed, re-training pages the client changed. Small, recurring, and genuinely valuable — the assistant degrades without it.
You hold it and bill it through. There is no reseller console issuing licences, so treat this as a cost you pass on rather than a margin line.
The unanswered-questions list is the single most impressive thing you can put in a monthly client report. It reads as insight into their customers, because it is.
CustomerBot is a hosted product, so there is no self-hosted or on-premise edition to resell, and assistant counts are set by the plan rather than unlimited. If your model is dozens of low-value client bots under one subscription, the plan limits will bite. If it is a handful of clients who each want something that genuinely knows their business, the economics work.
No. There is no reseller dashboard, no client sub-accounts, no licence assignment and no billing-on-behalf. You hold the account, create an assistant per client, brand each widget as theirs, and invite them to view their own one. That covers a great many agencies. If your plan is to sell chatbot software under your own name with a console to manage it, you want a vendor with a proper reseller programme — several run them, and you will be happier there.
That is yours to decide, but price the work rather than the software. The setup — choosing sources, writing Q&A pairs in the client’s voice, testing what it refuses — is the afternoon that makes it feel bespoke, and it is worth a setup fee. A small monthly retainer for reading transcripts and filling the gaps is where the recurring value sits, because an assistant nobody maintains slowly gets worse. Pass the subscription through at cost; the margin is not there.
Yes. Each assistant has its own knowledge base, its own branding, its own team and its own analytics. Plans set how many assistants an account can hold — Free and Hobbyist include one, Pro includes two — so a roster of clients means either multiple accounts or a Pro plan per pair.
No. Team membership is granted per assistant. Someone invited to one assistant has no visibility of any other, and roles below owner cannot reach billing or account settings at all.
The watermark is removed on the Pro plan. Everything else about the widget — colours, avatar, copy, position, animation — is configurable on every plan including Free.
Re-crawl the changed URL from the training screen. The old passages are replaced and the assistant answers from the new copy immediately — there is no retraining step to schedule.
The crawl and embed is the slow part and it is measured in minutes, not hours. The honest answer is that a first client takes an afternoon because you are writing the Q&A pairs that make it feel bespoke. The second one takes twenty minutes.
The free plan trains a full assistant on a real domain. If it does not sound like the client after twenty minutes of Q&A pairs, you have lost an evening.
Start free