Journal · AI agents

Where an AI agent fits into a business process

An AI agent does not replace a whole process. It slots into specific steps, while CRM automation and people keep handling the rest. Here is how to map a process so each step goes to whoever handles it most reliably and at the lowest cost.

An AI agent belongs on the steps where someone has to understand free text, speech or a document and decide by set rules: the first reply to a client, qualifying a lead from a chat, scoring a call, a summary for the manager. Steps with a clear trigger go to CRM automation, while money, exceptions and disputes stay with a person. At MakeBiz Technologies one agent costs AED 5,000, and CRM automation is set up at AED 200 an hour.

MakeBiz Technologies · 11 November 2026 · 7 min read

Every process has three kinds of steps

Write down every step from first enquiry to payment. Nearly all fall into three groups: an action that follows a rule, an action that requires understanding, and a decision someone is accountable for. The group decides who should do the step.

Process stepWho does itWhy
A website form becomes a lead in the CRMCRM automationA fixed rule: create the lead and record the source
A client messages on WhatsApp in the eveningAI agentIt has to understand the question and answer from the knowledge base
Assigning the account managerCRM automationRouting by queue, language or team
Qualification: need, budget, timingAI agentThe answers are buried in free-text chat
A discount above the limitPersonMoney and responsibility for margin
Scoring a call against criteriaAI agentIt means analysing speech, not reading a field
An invoice from a template once terms are agreedCRM automationAll the data is already in the deal record
A summary of stalled dealsAI agentIt has to pick what matters and explain why a deal is stuck
A disputed client complaintPersonThe relationship and a decision on behalf of the company

MakeBiz Technologies mapping for a typical sales process. In your process the steps and their owners may differ.

CRM automation takes every if-then step

If a step can be written as "when the deal moves to stage X, do Y", it does not need AI. Creating a lead from a form, setting a task, sending a template email, moving the stage after payment: that is the job of Bitrix24 automation rules and workflows. A rule fires the same way every time, costs less and does not slip on simple things.

An agent on such a step is paying for understanding where there is nothing to understand. We explained the difference between the two tools in our news piece on AI agents versus CRM automation. When standard rules are not enough, custom scenarios and links to outside systems are built through Bitrix24 development at AED 200 an hour.

An agent steps in where meaning matters

You need an agent on the steps where the input does not fit into a field: a client message, a conversation, a supplier email, a document. A rule cannot tell that "would it be cheaper if we signed for a year?" signals interest in an annual contract. An agent can, answers from the knowledge base and notes in the deal record what the client is after.

Typical places for an agent in a sales process:

  • the first reply to a client when the rep goes quiet, plus pipeline control: that is the AI sales manager
  • scoring every conversation against criteria instead of a sample: that is the call agent
  • sorting incoming enquiries in a messaging app: we described how it works in our news on the AI agent in WhatsApp

Some of these tasks are covered by the AI built into Bitrix24, which transcribes calls and fills in fields. Where that is enough and where you need a separate agent is explained on our AI-powered CRM page.

Let us map your process together

We will show where you need an agent and where CRM automation will do. Free 30 to 40 minute demo, one agent from AED 5,000

Book a demo

A person stays where mistakes are expensive

Some steps belong to neither automation nor an agent: a discount above the limit, non-standard payment terms, a disputed complaint. A person decides here, because a person answers for the outcome.

The agent's job on these steps is different: spot that the situation has gone outside the rules and bring in the right colleague with a short summary. The manager gets the essence rather than the raw chat: what the client wants, what has already been promised, how much is at stake. The decision stays with them, it just happens faster.

The agent's limits are set before launch

Before launch every agent gets three lists: what it can read, what it can change and what it must never do. For example, the agent may reply to clients and update qualification fields, but cannot change the deal amount and never promises discounts. Everything it does is visible in the deal record.

The agent runs under its own user account with the right permissions. Access and data stay with your company, and if the agent replies on a messaging channel, that channel is connected the official way: we connect WhatsApp to Bitrix24 for AED 500, details on our messaging page.

One working session is enough to map the process

Mapping does not need a long project. You can do it in a single meeting with the process owner:

  1. List the steps from first contact to payment as they happen today, not as they were designed
  2. Tag each step: rule, meaning or decision
  3. Mark where things fall through most often: unanswered enquiries, calls nobody reviews, deals that do not move
  4. Put an agent on one narrow step that needs understanding, and hand the rest to automation
  5. Decide which number will show the result: response time, share of calls reviewed

With that map the budget is easier to plan. What it is made of is covered in our guide to AI agent scope and pricing, and how to run the first quarter in our 90-day AI adoption plan.

A worked example: a service company taking WhatsApp enquiries

A hypothetical air-conditioning maintenance company in Sharjah gets most of its enquiries on WhatsApp. Evening enquiries wait until morning, and CRM automation only creates the lead.

After mapping, the process looks like this. Automation creates the lead and assigns a rep by area. The agent replies in the evening, asks for the address, the unit model and the problem, and records them in the deal. Automation books the visit in the calendar once the client confirms a time. A person handles only the exceptions: warranty disputes and discounts. Budget on our lines: AED 5,000 for the agent under Start, AED 500 a month for support and a few hours of automation work at AED 200 an hour. All agents are listed on the AI agents page.

Questions and answers

How is an AI agent different from CRM automation?

Automation runs an action on a clear trigger: the stage changed, so create a task. An agent understands free text, speech and documents and makes decisions within the rules you give it.

Can an agent run the whole process at once?

Technically yes, but it is safer to start with one narrow step that needs understanding. Rule-based steps are cheaper and more stable in CRM automation.

Which decisions should never go to an AI agent?

Decisions about money and exceptions: discounts above the limit, non-standard payment terms, disputed complaints. The agent spots these cases and hands them to a person with a summary.

How do I check what the agent is doing?

Every action the agent takes is visible in the deal record, and its permissions are set in advance: what it reads, what it changes and what is off-limits. A manager reviews its work the same way as an employee's.

How much does it cost to add an agent to a process?

One agent under the Start package costs AED 5,000 one-off, with support at AED 500 a month. CRM automation around the agent is set up at AED 200 an hour, prices exclude VAT.

Let us find the right spot for an agent in your process

Describe the client journey from enquiry to payment and where it gets stuck most often. We will suggest a step map and an estimate in AED

By clicking the button you agree to the privacy policy

CallWhatsAppRequest