Skip to content

/articlesConcepts

What is orchestration for company assistants?

AI orchestration in plain words: the software that lets an assistant use your systems under your rules and wait for approval. Examples, parts and what to check.

Updated
Published
min read
06
On this page
09
On this page09

What does orchestration mean here?

A language model on its own reads text and writes text. It cannot check your stock, create a quotation or know who is asking. AI orchestration is the name for everything around the model that makes those steps possible and safe.

From here on we use plain words. The assistant is what your customers and staff talk to. Rules say what it may do. Approvals are the moments it stops and waits for a person on your team. The log records what happened.

The word is also used for data and training pipelines, with tools such as Apache Airflow. That is a different job. This article is about a live assistant that answers people and works in your systems.

How is it different from a chatbot or a single model call?

Take one customer message on WhatsApp: “I need 20 boxes of A4 paper. What is the price, and can you send me a quotation?” Each option below handles it differently.

One model callChatbotAssistant with rules and approvals
What it knowsOnly the text you paste inA fixed list of answers or an FAQYour price file, your website and live data from your systems
What it can doWrite a replyFollow a scripted menuSearch products in Odoo and prepare the quotation
Who decidesNobodyWhoever wrote the scriptYour rules, and a person when you ask for approval
What you can check laterNothingThe chat transcriptEach step, each system call and each approval

What are the parts of an assistant you can trust with real work?

Vendors draw the picture differently, but the same parts come up every time. If one is missing, your team ends up building it later, usually after a mistake.

  • Knowledge: the files and pages it answers from, such as a price list in Excel or the clinic’s opening hours on the website.
  • Integrations: the systems it works in, such as Odoo for quotations or Gmail for replies, each with a list of allowed actions.
  • Rules: what runs on its own, what needs approval and what is never allowed. Reading an order can run on its own; sending an email waits for a person.
  • Approvals: the assistant pauses, someone on your team approves or rejects, and the work continues from the same point.
  • Hand over to a person: when the question is outside what it knows, the conversation moves to your staff.
  • The log: who asked, what the assistant looked up, what it changed and who approved it.
  • Channels: where people reach it, such as WhatsApp, Telegram, Slack or the chat on your website, all under the same rules.

What happens step by step?

Back to the paper order from a company in Riyadh. This is what happens between the customer’s message and the reply.

  • The message arrives on WhatsApp and the assistant for sales picks it up.
  • It searches the products in Odoo: A4 paper, SAR 145 a box, 64 in stock.
  • It reads your delivery terms from the knowledge you uploaded: free delivery in Riyadh above SAR 1,000.
  • It prepares quotation S00231 in Odoo. If you turned on approval for quotations, it waits for your sales team first.
  • It replies to the customer with the total, and every step is written to the log.

Where do companies use it?

The best fits share three things: the answer sits in more than one place, someone on your team copies information between systems today, and the last step needs care.

  • Stores and sales: answer price questions on WhatsApp and prepare quotations in Odoo.
  • Clinics: answer questions about services and hours, book appointments and send reminders in the same conversation.
  • Odoo companies: staff ask the assistant inside Odoo for an order status or a customer’s overdue invoices.
  • IT teams: gather what Dynatrace, Splunk and ServiceNow show about an incident and suggest a ticket update that an engineer approves.

How Botify does it

Botify is a platform for building these assistants. You connect your systems, teach the assistant from your files and website, and launch it on the channels your customers use.

Channels today are the chat on your website, WhatsApp, Telegram, Slack, inside Odoo, the API and MCP. WhatsApp and Telegram also take voice notes. A phone agent is on its way.

Integrations include Odoo, ERPNext, Gmail and Google Calendar, Outlook and Microsoft 365 tools, HubSpot, Salesforce, Zoho CRM, Freshdesk, WooCommerce, ServiceNow, Dynatrace and Splunk, plus any MCP server you add. Each action has a rule: it runs on its own, it follows your rules, or it always waits for a person. On WhatsApp and Telegram your team can approve with a button in the chat; other requests go to the Inbox. Everything is recorded in the audit log.

Frameworks or a platform?

Developer libraries such as LangGraph or LangChain help engineers build the steps of an assistant. They do not give you per-company data separation, stored credentials, roles, approvals or an audit log. Your team builds and runs those.

A platform comes with those parts so every new assistant gets them. Open standards help at the edges: the Model Context Protocol, released by Anthropic in November 2024, gives tools a common way to plug in. It does not decide whether a given person may use a given tool. That stays with the platform.

What to check before you choose a platform

Ask for a live demo on a task you care about, not slides. Each check below should take a few minutes.

  • A refused action: someone without the right role asks for something, and the assistant refuses before anything runs.
  • An approval: a quotation or an email waits, a manager approves it, and only the approved version goes out.
  • The log for that conversation: what was asked, what the assistant looked up, and who approved.
  • Data separation: how the vendor keeps your company’s data apart from other customers’ data.
  • Credentials: how the passwords and keys for your systems are stored, and whether the model can ever see them.
  • Arabic: the same task in Arabic and English, including a customer who writes in Gulf dialect.

Common questions

01

Is orchestration the same as an assistant?

No. The assistant is what people talk to. Orchestration is the software around it: the integrations it may use, your rules, approvals and the log. One platform can run many assistants under the same rules.

02

Do I need this if my staff only use ChatGPT or Copilot?

Not if people only ask questions and copy the answers themselves. Once an assistant reads your systems or acts for your company, you need rules on what it may do, approvals for the steps that matter and a record of what it did.

03

Where do approvals fit?

Before a step you marked as important, such as sending a quotation or an email, the assistant pauses and waits for someone on your team. If they approve, it continues from the same point. If they reject, nothing is sent. Either way the decision is in the log.

04

Is MCP an orchestration platform?

No. MCP, the Model Context Protocol, is a standard way to connect AI apps to tools. It makes a tool easier to plug in, but rules, approvals and the log still come from the platform that uses the tool.

Sources

  1. [1]Introducing the Model Context Protocol (Anthropic, November 2024)

Where you’ll find this in Botify

Start with one assistant on WhatsApp or your website.

Book a demo and we’ll show you an assistant answering questions like your customers’. We set up the first one with you and review the results together before you add more.