The Assistant and the Employee: Where Each Role Begins and When a Human Steps In

The Assistant and the Employee: Where Each Role Begins and When a Human Steps In

Part four set out six controls that stop an assistant inventing prices and facts, and one of them was escalation: when it does not know, it hands over to a human. But escalation is a technical control, and what a business owner needs is bigger than that — a clear map of who answers what.

Start with the sentence that occupies every sales or service team the moment they hear the words artificial intelligence: this is not a replacement for your staff. The simpler truth is that what consumes most of your employee's day today is neither selling nor persuading, it is answering the same ten questions over and over. Moving those ten to the assistant does not reduce your employee's value; it frees them for the work nobody else can do.

The rule: what is repetitive and known goes to the assistant, what needs judgment goes to the human

You do not need a complicated matrix to divide the roles. One question is enough: does this conversation have a single correct answer that your company already knows? If yes, it belongs to the assistant. If the answer depends on reading the situation, on a concession, or on a decision that commits the company, it belongs to a human.

On that basis, repetitive conversations fall naturally to the assistant: opening hours and location, do you offer this service, what does this package include, the frequently asked questions, and collecting the first details of an enquiry before an employee takes it over. These are not marginal conversations — they are the bulk of what reaches you, and they are exactly what you lose after hours and on weekends.

Four conversations you never automate, however good the assistant is

• Complaints. An angry customer does not only want a correct answer; they want to feel that someone responsible heard them. A polite automated reply to a genuine complaint multiplies the anger instead of absorbing it.
• Negotiation. Discounts, concessions and exceptional terms are commercial decisions that commit your company financially. These specifically belong on the refusal list we built in part four.
• Large or custom deals. A non-standard request needs someone who reads between the lines and asks questions that never occurred to you when you built the knowledge. The assistant opens the door and gathers the details; the employee closes the deal.
• Anything that binds you legally or contractually. A firm delivery date, a warranty, a written commitment. Those are contractual statements, not chat replies.

Notice that these four are also your highest-value conversations. What you keep for your employee is not the leftovers — it is the most valuable part of the day.

A clean handoff: the customer never repeats themselves

The worst possible experience is a customer explaining their problem to the assistant in detail, and then an employee arriving to ask: hello, how can I help you? At that moment the customer loses confidence in the whole setup, and the assistant becomes an obstacle in their eyes rather than a service.

A clean handoff means the entire conversation moves with the employee: what the customer asked, what the assistant answered, and any details it collected. Require three things from your vendor by name:

• The employee sees the full conversation log before typing a single character.
• The assistant stops replying the moment the employee joins, so two voices are never in the same conversation.
• The transfer is made clear to the customer in one short line: I am passing you to my colleague, who has read everything you have written.

Define when escalation happens, too: when the customer explicitly asks for a person, when the same question is repeated twice without a convincing answer, when a complaint or an angry tone appears, and on any topic in the refusal list. Make these written rules, not individual judgment calls.

Your employee's day after launch

Before launch, your employee opens WhatsApp to dozens of conversations, most of them identical questions, answers them in arrival order, and discovers at the end of the day that the serious buyer was lost in the crowd. After launch, they open WhatsApp to conversations that reached them because they deserve them: a customer whose details are already gathered and who is ready for a proposal, a complaint that needs handling, a deal that needs negotiating.

This changes how you evaluate them as well. The measure is no longer how many messages an employee replied to, but how many conversations they converted into customers and how well they resolved complaints. Say this to the team out loud, because it is the difference between seeing the assistant as a threat and seeing it as something that serves their own numbers.

How do you introduce it to a team afraid of being replaced?

The fear is natural and is not cured by a reassuring sentence in a meeting. Handle it in four steps:

• Explain the intent openly on day one: why you are launching it, what will change in their day, and what will not.
• Let the team build the knowledge and approve the answers — an employee who wrote the assistant's reply is not afraid of it.
• Start in draft mode as in part four: the assistant writes and the employee approves. That inverts the relationship, making the employee the supervisor rather than the replaced.
• Change the performance metric publicly from reply count to conversation outcome, or the team will resist anything that lowers their numbers.

The Origami view

In Mahir, Origami's AI assistant for WhatsApp, this division runs on your company's own existing number with no new number and no extra app on the employee's phone, so the conversation stays in one place your team can see and continue from rather than moving to a separate system. Training starts by giving it your website link to read and fill in your details, then you add the services, prices and frequently asked questions with answers you approve — which are precisely the repetitive questions you want lifted off your staff. It also has two tiers you choose between: Malik for fast, economical handling of repetitive direct questions, and Mahir for long conversations, compound requests and objections while holding your company's voice, and you can switch at any time with no re-linking and no loss of training. The trial is free, so try it on the ten questions you get most.

Conclusion

The dividing line is simple: what has a single known answer goes to the assistant, and what involves judgment, concession and decision goes to a human. Write the escalation rules, insist on a handoff that carries the whole conversation, and involve your team in building it before you launch. Which leaves a question that precedes all of this: your customer conversations are personal data, so what does the Personal Data Protection Law actually require of you? That is part six.

Sources

• Saudi Data and AI Authority (SDAIA) — AI ethics principles, including human oversight and accountability.
• Saudi Vision 2030 — the National Transformation Program, on raising business efficiency and developing human capability.
• Official WhatsApp Business documentation from Meta — handing a conversation between an automated reply and a human agent.