Search for AI all-in-one business software and you get two kinds of answer that look alike and are not. One is a suite with AI features built into its products: a button that drafts the email, summarises the call, scores the CV. The other is a suite that lets the assistant you already use, Claude or ChatGPT, reach in and do the work. Both are called AI. They are bought differently, they fail differently, and the question that decides which one helps you is not about the model at all. It is about where your data lives and what a piece of software is allowed to do with it.
This piece is written by a company that sells an all-in-one suite with a connector for assistants, so the bias is declared. It is also why we have kept to what vendors say on their own pages, and to what our own product does today.
Why AI in a suite is mostly a question of where the data is
An assistant is only as useful as two things: what it can reach, and what it is allowed to do once it gets there. Ask it how many hours a person logged on a client this month and it needs the timesheets. Ask it to turn meeting notes into tasks and it needs the project, the sprint and the right to create issues. Ask it what someone was paid and it needs payroll, and a reason to be allowed near it. None of that is intelligence. It is access.
That is why the suite question and the AI question have become the same question. When projects, time, leave, hiring and pay sit in five products, an assistant needs five connections, five sign-ins, five ideas of who you are and five logs, and the data still does not line up: the person spelled three ways in three systems is spelled three ways to the assistant too. When they sit in one product on one record per person, there is one connection, one set of permissions and one answer to “did the assistant do that?”. The case for a suite used to be about invoices and offboarding; all-in-one versus best-of-breed sets that out. Assistants add a new line to it.
The two honest models
There are two ways a suite can offer AI, and most vendors now offer both. Neither is wrong. They answer different needs, and it helps to name them.
| AI features built into each product | Your assistant, connected to the suite | |
|---|---|---|
| What it is | The vendor runs a model behind a button in its own screens | Claude, ChatGPT or another client works in the suite through one connector |
| Who chooses the model | The vendor | You, when you choose the assistant |
| What it reaches | The one screen the feature was built for | Everything the connector exposes, across every tool it covers |
| Whose permissions | Usually the feature’s own, set by the vendor | Yours, if the connector is built that way: check it |
| Where it is strong | A repeated, narrow job with the data already on screen | Work in your own words across tools, and questions nobody built a report for |
| Where it is weak | Anything the vendor did not anticipate | Anything the connector does not expose, and judgement calls |
The built-in model suits a job that is the same every time and whose data is already on the screen: reading a hundred CVs against one job description, transcribing a call. The connected model suits the work that spans tools and changes daily: the Friday timesheet, the client notes that become issues, the leave request you want to approve without opening a browser. A suite that is honest about which of its AI is which is easier to buy than one that calls everything intelligent.
How three suites describe their AI, in their own words
We read three widely used suites on their own pages, and wrote down what each one says rather than what we think of it. Each has both models.
- Zoho One. Its pricing page mentions AI once, under the Standard plan, as analytics and business intelligence. The connected model is Zoho MCP, a platform Zoho describes as connecting LLMs such as GPT or Claude to its APIs and actions, with OAuth-based authorisation for access control. Zoho also says it is “made for developers” and “autonomous by design”: agents can act without a human prompting them.
- Bitrix24. Its prices page lists CoPilot in CRM, Tasks, Chat, Sites, Feed and video calls, and a top tier with unlimited AI, voice-to-task and a chat assistant. The connected model is the Bitrix24 MCP server, which its help article says lets an external AI system such as Manus, Cursor or n8n manage tasks and CRM, and performs actions “only when an employee gives a command and only within that employee’s access permissions”. An MCP Hub for connecting outside services is listed as coming soon.
- Odoo. Its pricing page lists an Artificial Intelligence app, and puts Odoo Studio and agentic AI in the Custom plan alongside the API and custom development. The connected model is an MCP server in recent versions: the user creates an API key with an MCP scope and a validity period, and the tools exposed to the client are server actions an administrator chooses in the technical settings. The documentation says the key authenticates the user’s identity and permissions in the database.
Three things stand out. Every one of these suites offers a connector, so the question is no longer whether a suite has one but how it is built. Two of the three say in plain words that the assistant is held to the person’s own permissions, which is the right standard. And the setup differs: a developer-oriented platform, a server addressed to outside AI systems, and an API key a user generates. Which one suits you depends on who will connect it, and whether that person is an administrator or an account manager with a Friday timesheet. The comparison pages for Zoho One, Bitrix24 and Odoo carry the prices and the rest of the picture.
What to check before you connect an assistant to anything
Whichever suite you look at, the checks are the same, and each one is something you can see on a screen rather than take on trust. We hold our own connector to them, so they are written against what Orbiz shows; ask any vendor to show you the same.
- It acts as you, in one organisation. A connection belongs to one member in one organisation. The consent page names both, and says the assistant can only see and do what you can on the web, in the areas you allow. A shared service account with broad access is the thing to avoid.
- You choose what it may do. Connecting sends you to an Orbiz consent page, signed in as yourself. Access is listed tool by tool, reading separate from changing, and you can untick anything before you approve.
- Pay changes are a separate tick. Changing salaries is its own permission with its own warning on the consent page: only tick it if you run payroll.
- Every action is on a list. An activity log shows each action with the assistant, the outcome — done, not allowed or failed — and the time. You see your own; admins see the organisation’s and can filter by member.
- One switch for the whole organisation. An admin can turn AI assistants off for everyone. That stops every assistant and blocks new connections; existing ones work again when the switch goes back on.
- Some things have no action at all. There is no action for the Vault, none for clocking in or out, and none that deletes anything. An assistant cannot be talked into what does not exist.
- Disconnect whenever you like. Any member can disconnect an assistant from Settings → Connected AI apps, and it loses access at once. Admins can disconnect anyone’s.
The eighth check is the one vendors are least keen to write down: what the assistant cannot do. A connector that reaches everything is a connector that can be talked into anything, so the list of exclusions is a feature. In Orbiz, an assistant connected to HRMS cannot reach employee records and documents; interviews, assessments and offer letters; organisation, department and role settings. In AMS it cannot reach clocking in or out, and the attendance timer; viewing or taking screenshots; attendance corrections and shift settings. In Ledgr it cannot reach payroll runs and salary sheets; submitting claims or requesting loans, and anyone else’s claims, loans or settlements; tax registers and certificates. In Cadence it cannot reach deleting issues, sprints or projects; board, workflow and custom-field configuration. We publish those lists next to what each tool does expose, because a buyer should be able to see the edge.
The assistants add checks of their own. ChatGPT’s developer mode, which is how it connects to a custom server, asks you to confirm a write action by default, and OpenAI’s guide describes the mode as powerful but dangerous and names the risks: prompt injection, mistakes on write actions, and malicious servers. Claude’s help pages say a custom connector connects Claude to a service Anthropic has not verified, and that on Team and Enterprise plans an Owner adds the connector before members can use it. Read both as a reminder that the judgement about what to connect stays with you.
What Orbiz has built in, and what it does not claim
Orbiz is an all-in-one suite of 10 tools on one login. Its built-in AI lives in one of them: HRMS AI screening reads each CV against the job it was sent for and puts a match score, a recommendation and the matched and missing skills on the candidate’s profile. It is off by default and switched on per job, the only setting is on or off, and it never rejects, advances or hires anyone. The CV text goes to OpenAI only when a job uses it, and OpenAI is on our sub-processors page with exactly that scope. HRMS has a few smaller pieces beside it, résumé parsing and drafted assessments a person approves, and that is the whole list. The other tools have no AI built in, and we do not describe the suite as intelligent, because it is not: it is a place where the data lines up.
The connected model is where the suite earns the word. Orbiz MCP is one connector for the whole organisation. An assistant connected through it can take 81 actions across 9 of the 10 tools: 18 in Cadence for projects, sprints and issues, 10 in AMS for timesheets and leave, 14 in HRMS for tasks, jobs and candidates, 7 in Ledgr for a person’s own pay and, for payroll managers, salary revisions, with files, notes, chat and recordings besides. Every member connects with their own sign-in, the assistant acts as that member, and the seven checks above are the consent page, the activity log and the admin switch you can open in Settings. It is included in every plan. The requests it carries out as written are in the prompt library, each marked as reading or changing something.
Who this suits, and who it does not
A 10- to 200-person software house, agency or professional firm that already uses Claude or ChatGPT, keeps its operations in too many places and has nobody whose job is the joins between them will get more from a suite with a permission-aware connector than from any single built-in feature. The clerical half of operations is exactly the half an assistant can carry, provided it can reach the records and is held to the person’s own permissions.
It is the wrong answer for a team whose work depends on one deep specialist tool and its marketplace, for a company that wants AI to make decisions rather than do the typing, and for an organisation that does not use an assistant yet: a connector changes nothing until someone connects one. Orbiz in particular is not a CRM or an accounting system, and if the AI you need is a sales copilot or an agent that runs overnight without a person asking, the suites above describe those and we do not. For what an assistant means for a small team in general, and the order to switch it on in, see MCP for small business.
Sources: Zoho One pricing (zoho.com, read 2 October 2026) and the Zoho MCP page (read 4 October 2026); Bitrix24 prices (read 2 October 2026) and its “MCP in Bitrix24” help article (read 8 October 2026, last updated by the vendor 5 October 2026); Odoo pricing (read 2 October 2026) and the “AI MCP server” page of the Odoo saas-19.4 documentation (read 4 October 2026); Claude’s Help Center article on custom connectors and OpenAI’s developer mode guide (read 4 October 2026). Orbiz counts come from the Orbiz MCP catalogue on the day of publication.
Built for your industry
