Chatbot vs supervised AI: what automation suits a brokerage's WhatsApp conversations?

"Automating WhatsApp" covers two very different approaches. The classic chatbot is a decision tree: a button menu ("1. Policy schedule, 2. Claim, 3. Speak to an adviser"), pre-written answers, perhaps a form. It answers alone, predictably, and stops dead as soon as the request leaves the script. Supervised AI, as ORIS practises it, does not take over the conversation: it reads each incoming reply, classifies it (opportunity, churn risk, neutral, escalate), measures sentiment, proposes a draft to the adviser, and only auto-replies inside a narrow set of rules — a few sentences, never a price or a commitment on cover, and always a human as soon as the tone turns negative or the request is sensitive.

For a broker the question is not technological but regulatory: advice under FAIS (given by a licensed representative), the FCA's Consumer Duty and the IDD cannot be delegated to a machine. This comparison describes what each approach really does, its costs, its risks, and the cases where a simple chatbot remains the right choice.

Scripted chatbot (menus, pre-written answers)

Pros

• Predictable and auditable: every answer was written and approved by the firm; nothing leaves the script.
• Effective for repetitive, closed requests: office hours, address, FICA document list for a new policy, link to the client portal.
• Inexpensive and available in many WhatsApp Business Platform tools, using quick-reply buttons.
• No regulatory ambiguity: the chatbot gives no advice, it routes.

Cons

• Frequent dead ends: the client who writes "my debit order bounced and I'm driving to Durban tomorrow" fits no button and gets annoyed.
• No understanding of meaning: no detection of an unhappy client, an intention to cancel or a cross-sell opportunity.
• Degraded service image if the chatbot is the systematic first contact; in insurance, the client wants a person.
• Maintenance: every new case adds a branch; the tree soon becomes unreadable.

Supervised AI (classification, drafts, rule-based auto-reply)

Pros

• Reads natural language and classifies each reply: opportunity, re-engagement, high churn risk, neutral, escalate; the adviser sees what matters first.
• Suggested drafts from the latest messages: the adviser reviews, edits, sends; the advice stays human.
• Framed auto-reply: enabled by rule, capped per client and over time, reserved for positive or neutral exchanges; negative sentiment or a failed send means a draft for a human.
• Feeds the client record: churn risk, opportunities, detected life events, with no manual capture.

Cons

• Does not replace an adviser: the AI gives no price, no cover, no advice, and does not run a conversation end to end.
• Needs the Business Platform and a tool: subscription cost, and classification is not perfect — human review remains essential.
• Can look opaque: you must be able to explain why a reply was sent automatically (audit log, visible rules).
• Little value without volume: a firm receiving ten messages a day reads them faster than it configures rules.

Criterion-by-criterion comparison

Criterion	Scripted chatbot (menus, pre-written answers)	Supervised AI (classification, drafts, rule-based auto-reply)
Who answers the client	An approved script.	The adviser, helped by AI; the automaton only under rules, on short messages.
Understanding of the message	Keywords and buttons.	Natural language: intent, sentiment, urgency.
Advice duty (FAIS, Consumer Duty)	Out of scope (routing only).	Stays human; the AI gives no price or cover.
Unhappy client	Not detected; loops through menus.	Negative sentiment detected → draft and notification for a human.
Opportunities and risks	Not detected.	Classified and visible in Opportunities & Risks, with a risk score.
Traceability	Chatbot log.	Audit log, "Auto-replied" notification, reasoning recorded.
Indicative cost	Low; often included in Business Platform tools.	Subscription to a WhatsApp CRM with AI; no extra Meta charge for replies inside the 24-hour window.
Suitable volume	Any size for closed requests.	From a few dozen client replies a day, or after each reminder campaign.
UK / EU vs Africa	Useful everywhere for documents and hours; in Africa, plan for plain English and local languages.	Same logic; sentiment detection depends on language and model — test on your real conversations.

Our verdict

The scripted chatbot is the right choice for closed, repetitive requests, whatever the size of the firm: "which documents do I need to insure my car?", "where do I send the claim form?", "what are your hours?". It is also the only sensible option if you have no team available to review drafts: a predictable automaton beats an AI nobody supervises. And if your volume is low, neither is needed: an away message and a human answer the same day will do.

Supervised AI earns its place after a renewal or debit-order campaign, when hundreds of replies arrive at once: it sorts them, surfaces at-risk clients and opportunities, and proposes drafts the advisers approve. That is how ORIS works: reply classification, drafts, auto-reply only under rules (cooldown, maximum per client, positive or neutral tone) and always a human as soon as sentiment is negative or a request falls outside the frame. It does not advise, quote or bind cover — which keeps the representative, not the software, accountable under FAIS and under Consumer Duty.

The two can coexist: quick-reply buttons on templates to frame simple answers, supervised AI for the rest. What to refuse is the automaton that "answers everything" in insurance: the advice duty, the evidence of what was said and the client's trust cannot bear it. Our guides on AI for brokers set out the rules to put in place before switching on any auto-reply.

Frequently asked questions

Does an automatic WhatsApp reply bind the brokerage?

Yes, like any message sent from the brokerage's number. That is why ORIS's rules restrict auto-replies to short sentences with no price or promise of cover, and why every reply is logged; when in doubt, the AI produces a draft and a person decides.

Can we switch auto-reply off completely and keep only drafts?

Yes. Auto-reply is a rule the firm chooses to enable or not; without it, the AI simply classifies messages, measures sentiment and proposes drafts the adviser sends or rewrites. Many brokerages start that way.

Does the AI understand isiZulu, Swahili or mixed-language messages?

Understanding depends on the model and the language; everyday English is handled well, mixed-language messages unevenly. Test on a sample of your real conversations before enabling an auto-reply rule, and keep the human draft as the default on those segments.

Do chatbots and supervised AI comply with the WhatsApp Business policy?

Both must: no unsolicited messages outside approved templates, the 24-hour window respected, opt-outs honoured. An automatic reply is only possible inside a conversation the client opened; outside it, only a template can be sent.