WhatsApp Business Provider or Custom AI Worker for Customer Replies

The short verdict

Choose a WhatsApp Business Solution Provider (BSP) if your team wants to build and own the replies, your enquiry volume is modest, and a person will read most conversations. Choose a custom AI worker if most messages are predictable, the answers live in your order or billing system, and the bottleneck is staff hours spent copying status updates into WhatsApp. Neither option removes the need for a human owner. Someone must approve templates, review escalations and decide what the bot may promise.

How the two options differ

The table below sets out the practical differences. Prices are shown only where the vendor publishes them, and the year is given with each figure.

Who each option suits

A BSP suits a team that already has someone who runs the inbox, sends campaigns and answers questions by hand, and wants a shared screen for that work. It also suits a business that changes its offers often and wants marketing staff to edit flows without a developer. The weakness is that the bot knows only what is typed into its flows. If an order status depends on a warehouse record, someone must copy it across.

A custom AI worker suits a business where the same five or six questions make up most of the traffic, such as where is my order, when does my invoice fall due, or can I change my delivery address. It reads the source record, answers, and hands the rest to a person with the conversation attached. It is a poor fit for a team that wants to test new campaigns every week, because each new rule is a change someone has to make and check.

What Meta's October 2026 change does to your bill

Until the end of September 2026, replies inside the 24-hour customer service window were not charged. From 1 October 2026, replies in that window cost the utility rate after a free allowance of 1,000 service messages per business phone number each month, according to a 2026 pricing guide from Getkanal. Another 2026 guide from Zenvia says service messages move to the same unit value as utility and authentication messages in each market, with no volume tiers. Both point the same way: a bot that answers every inbound question now carries a per-message cost, whether it runs through a BSP or a custom build.

The rate itself depends on category and country. Getkanal's 2026 list puts a United States marketing message at $0.0250 and a United States utility message at $0.0034. In Germany and Brazil the same categories cost more. Marketing messages cost far more than utility messages, so a business that sends promotional text inside an order update can see its bill rise sharply. Keep order messages strictly transactional.

Template approval: the step both options share

Any message sent after the 24-hour window must use a template that Meta has approved. Infobip describes templates as the required format for business-initiated messages outside that window, including order confirmations and delivery updates. Meta's review is commonly described as taking up to 24 hours, although some guides say up to 48 hours in rare cases. Utility templates typically clear within minutes. Marketing templates take longer and are more often rejected.

Wording matters. Courier's 2026 guide says anything mixed or unclear counts as marketing, which means a shipping notice with a discount line can be re-classified and cost more. Rejected templates must be edited and resubmitted. A BSP usually submits templates through its dashboard, and a custom worker uses the same Meta submission route. In both cases the business owner must approve the final wording.

Failed deliveries: who sees them and who acts

A message that Meta accepts can still fail later. Meta reports the failure asynchronously through a status webhook, and the failure is not known at the moment you press send. Common causes include a recipient who is not registered on WhatsApp, a message sent outside the 24-hour window without a template, and a number that has hit a throughput limit. Some providers say they do not charge for failed messages, and that should be confirmed in the contract before you sign.

The risk is silent loss. If the webhook endpoint is down or rejects the payload, failure notices disappear. A BSP handles the Meta connection and usually shows statuses in its dashboard, but someone still has to decide what happens next. A custom worker has to be built to act on each status: retry, fall back to a template, or alert a named person. Ask any vendor which team owns the webhook and who gets the alert when it breaks.

Total cost of ownership beyond the sticker price

• Setup time. A BSP takes days to configure in the simple case. A custom worker takes longer and needs access to your systems, which is often the slowest step.
• Maintenance. Templates, flows and rules need updates whenever a product, price or policy changes. Count the hours your staff will spend on this each month.
• Breakage. Webhook outages, rate limits and expired access tokens stop replies quietly. Budget for monitoring either way.
• Message fees. Meta's rates, the provider's fee and any markup on top of Meta's rates all grow with volume. A 2026 review of one BSP reports markups between 42% and 80% on Meta's rates, while the vendor's own pricing page should be checked for the current figure.
• Vendor lock-in. Flows built in one BSP's builder may not move to another. Conversation history and tags may export only in part. A custom build is yours, but you depend on whoever maintains it.

Note that third-party guides disagree on Wati's 2026 entry price. Some list a Growth tier at $29 a month, others at $49 or $59 a month on annual billing. Take the current figure from Wati's own site before you compare.

The custom-built alternative and when it wins

A custom build wins when your order or billing data is complex, when the same answers repeat thousands of times a month, or when you need the bot to act on records rather than read a script. It also wins when you want one place to control consent, logging and escalation. Meta's own AI infrastructure is another option to check. Zenvia reports that Meta launched it on 1 July 2026, with per-token billing starting 1 August 2026. Ask any builder how it plans to use that route before committing.

A custom build loses when your volume is low, when your rules change weekly, or when no one in the business can review what the worker says. Budget for a person to read escalations every day in the first month.

Limits of both approaches

Neither option replaces judgment. Courier's 2026 guide notes that a bot is acceptable only if customers can reach a human. Meta also restricts general-purpose AI assistants offered as the product itself, so a worker must serve your business, not be sold as a generic chat tool. Automation cannot resolve a refund dispute, a complaint about damage, or an answer your policy does not cover. Those belong with a person from the first message.

Migration notes

1. Export contacts and consent records before changing anything. Do not import contacts who never opted in.
2. Map every template you use today, note its category, and check whether any order message carries promotional wording.
3. Decide who owns the Meta Business account and the phone number, and keep that access in the business's name.
4. Run the old and new setups side by side for one order cycle. Watch the failed-status reports before you switch the number.
5. Keep the full conversation history in a file you control, so a later move does not lose it.