Playbook

White label chatbot for agencies: how to resell AI chat well

Thinking about a white label chatbot for your agency's clients? Learn how to package AI chat, price it, handle client data and keep the answers accurate.

Key takeaways
  • Price AI chat as an ongoing service with a monthly content review, because upkeep is what keeps client answers accurate and it is the part clients pay for.
  • Keep each client's content, conversations and data separate, and confirm in writing what your vendor allows before you promise it to anyone.
  • Start with one client site and a 14-day trial, then turn the onboarding steps into a checklist you can repeat for the next three or four clients.

An agency that adds AI chat to client websites tends to make the same first mistake. It sells the demo. The client sees the assistant answer their own pricing question in a warm voice, signs off on a monthly fee, and then discovers in month three that the bot is still quoting spring prices. A white label chatbot is a useful business to build, but the business is not the widget. It’s keeping a client’s knowledge correct, accountable and well governed, and that’s the part most agency pitches leave out.

This is a playbook for agencies that want to offer AI chat under their own name. It covers where the margin really sits, how to price the work, what onboarding involves, how client data and roles should be handled, and how to avoid promising more than you can maintain. It deliberately doesn’t assume a particular vendor’s partner programme, because the terms vary and you should confirm them yourself.

Where the margin actually sits

The assistant software is largely the same for every client. What differs from one client to the next is the content the assistant answers from: their services, their prices, their policies, their opening hours, their staff routing. That content is the product you’re really selling. Everything else is plumbing.

This matters because it tells you where the recurring value is. A one-time setup fee covers the initial content work, which is real but finite. The monthly value comes from keeping the content correct as the client’s business changes. A plumber changes his call-out fee. A hair salon adds a service and drops another. A clinic opens an extra Saturday and closes for a refurbishment. Each change, handled late, creates a wrong answer that the client will blame on you.

So the clients worth having are the ones whose content changes, because those are the ones who need someone to manage it. Clients with a static brochure site will not value an ongoing service, and you’ll struggle to justify a retainer for them. Pitch the retainer as content management, not as software access.

Pricing a white label chatbot service

There are four common ways agencies package AI chat. Each has a different risk profile, and most agencies end up combining two of them.

Model How it works Works best when Main risk
Setup fee plus monthly retainer Onboarding is billed once; a fixed monthly fee covers upkeep and reporting Clients with a steady but occasionally changing business Underpricing the monthly upkeep
Resold subscription with margin You buy plans at volume and resell with a markup Many similar small sites with similar needs Vendor price changes squeeze your margin
Results-linked share A fee tied partly to bookings or leads Clients with clean, trackable outcomes Disputes over what counts as a result
Bundled into a care plan Included in an existing website maintenance retainer Agencies that already run site care plans Chat becomes an unlimited support cost

A workable starting point for many agencies is a setup fee plus a monthly retainer, with the retainer scoped in writing: a fixed number of content updates a month, a monthly report, and a defined window for urgent corrections. Results-linked fees look attractive, but they require the measurement discipline covered in how to measure chatbot ROI, and many clients won’t accept the attribution you can honestly prove. Use them as an upside, not the base.

Onboarding is the content week

Most of the work in a white-label project happens before the widget goes live. You’re collecting and checking the client’s content. Clients tend to underestimate how long this takes, and agencies tend to underestimate how much of it is missing.

A practical onboarding checklist covers the following, in roughly this order:

  1. Service pages and descriptions. What the client does, for whom, and what is excluded.
  2. Prices and what they include. Including any fees that appear late in the process, such as delivery, call-out charges or booking deposits.
  3. Policies in plain sentences. Cancellations, refunds, returns, warranties and guarantees, each with its conditions.
  4. Hours and exceptions. Seasonal changes, bank holidays, and the days the owner is out of the office.
  5. Staff and routing. Who receives which kind of enquiry, and how they want to be contacted.
  6. Calendar and booking details. Appointment lengths, buffers, lead times, and which staff member each service needs.
  7. Languages. Which languages the assistant should answer in, and whether the client’s content exists in those languages.
  8. Brand voice and forbidden phrases. How formal the client is, and anything the assistant should never say.
  9. Privacy text. The client’s existing privacy notice, so the chat’s data handling matches what they’ve already promised.

Assign a named person on the client’s side to each item. This is the step that agencies most often skip, and it’s the one that determines whether the project will still be accurate a year later. If nobody on the client side owns the prices, you will end up owning the wrong ones.

Build the checklist on your first client, not after: the first onboarding will take longer than you expect and will surface gaps you didn't know about. Write each one down as you hit it. The second and third clients should take roughly half the time.

Shared setup and templates Client A site own content and data Client B site own content and data Client C site own content and data Client D site own content and data Same setup for every client; separate content, conversations and data
One shared setup feeds each client site, but every client keeps its own content, conversations and data.

Data, roles and the GDPR question

Once a chat is live on a client’s site, visitors are sending personal data through it. Names, phone numbers, booking details and sometimes health or family information all pass through the conversation. That makes data roles a practical matter, not a theoretical one.

In most arrangements, the client is the controller of its visitors’ data: it decides why the data is collected and what happens to it. The agency acts on the client’s instructions, which makes it a processor in many setups. The vendor running the assistant is typically a further processor or sub-processor. The GDPR expects a written contract with each processor that sets out the subject matter, the duration, the purpose, the types of data, and the obligations on the processor, which is what a data processing agreement is for.

Practically, that means checking three contracts before you sell anything. The client should have a relationship with the assistant’s vendor or with you that includes an appropriate data processing agreement. You should know which sub-processors are involved, such as the hosting or language-model providers behind the assistant. And the client’s privacy notice should describe the chat accurately, including what is kept and for how long. The fuller picture for a single site is covered in what you actually have to do under GDPR for a chatbot. This is practical orientation, not legal advice. Involve a privacy professional in each market you serve.

Accountability: you will own the wrong answers

A client who sees a wrong answer on their site will blame the agency that set it up, whether or not the client supplied the wrong price. That’s the nature of the arrangement. You are accountable for what the assistant says on your client’s behalf, which means your contract needs to define the service precisely enough that accountability is fair.

Put these in writing:

  • Scope. What the assistant covers, and what it doesn’t: which pages, which languages, which bookings, and which questions it routes to a person.
  • Update commitments. For example, a price change supplied by the client is live within two working days, and an urgent correction within one.
  • Review cadence. A monthly content check and a short report on hand-offs and unanswered questions.
  • Client responsibilities. Supplying changes promptly, naming a contact, and approving the content before go-live.
  • Limits. The assistant is not a legal adviser, does not give professional advice outside the client’s content, and does not make commitments the client hasn’t authorised.

These clauses protect you from the most common dispute: the client who expects a perfect assistant without supplying the content to keep it perfect. They also make the retainer easier to justify, because the upkeep is now visible work with a defined scope.

Don't sell "AI that knows your business" without a maintenance clause: you are responsible for what the assistant says on the client's site, even when the error came from a price the client forgot to update. Without a defined upkeep scope, that responsibility is unlimited.

Keep each client separate, and say so

Multi-client work raises the question of separation. A plumber’s enquiries should never appear in an electrician’s dashboard, and a clinic’s patient conversations should never be visible to a salon. Before you promise anything about separation, ask the vendor, in writing, how client content, conversations and data are kept apart, who can access them, and how they are deleted when a client leaves.

Don’t assume that because a dashboard shows separate sites, the underlying data is separated the same way. Ask specifically about storage, access controls, and whether any client’s content is used to improve shared models. The answers should be clear and documented. If they aren’t, that’s information too, and it may mean the vendor isn’t the right fit for an agency.

Pitching without overpromising

A good pitch is short and specific. It names the client’s problem, describes what the assistant will do, states what it won’t do, and explains the upkeep. Three sentences often do the job:

“Most of your enquiries are the same ten questions, and they reach your inbox after hours. We’d set up an assistant that answers those from your own prices and policies, books the simple appointments into your calendar, and passes anything complicated to your team. We’d keep the content current each month, so the answers stay right as your business changes.”

Notice what’s missing: no claim of a specific percentage uplift, no promise that the assistant will handle every conversation, and no suggestion that it replaces the team. Those claims are hard to back up and they set up disappointment. The honest version wins more durable clients.

Don’t promise results you can’t measure for the client. If you offer a results-linked element, agree the measurement method before launch, using a split test where possible, so that both sides can accept the outcome later.

A realistic first six months

Agencies that succeed with this tend to follow a similar path. In the first month, pilot with one client who has a clear, changing business and an owner who will engage. Keep the scope tight: one page, a short set of questions, and a defined hand-off route. Read every conversation in the first fortnight and fix every gap you find.

In the second and third months, turn what you learned into a template and a checklist. Document the onboarding steps, the content sources, the review routine and the contract clauses. Then onboard two or three more clients, in sequence rather than all at once, because each new client surfaces new edge cases and you need the capacity to fix them.

By month six, you should know which kinds of client you serve well, which content types take the most upkeep, and what your retainer needs to cover to be profitable. That’s the point to decide whether to expand, not before. An agency that adds ten clients in a month usually ends up with ten sets of out-of-date prices.

Where to start

The commercial logic of a white-label chat service is sound when it’s built around content management and clear accountability. It’s shaky when it’s built around the widget. Start with one client, define the upkeep in writing, separate the data properly, and measure what happens. The rest of the playbook follows from those four things.

If you want to test the approach before building a service around it, Voxy offers a free 14-day trial with no card, and you can try it on a single client site first. Start with the live demo to see how the assistant handles a pricing question, a booking and a hand-off, then try it on one real site. Before you put any client’s visitor data through it, read up on GDPR for chatbots and make sure your contracts are in place.