/articlesComparisons
Build or buy a company assistant: how to decide
Build your own AI assistant or buy a platform? What a live assistant needs beyond the demo, a side-by-side comparison, when each wins, and what to ask a vendor.
- Updated
- Published
- min read
- 06
- On this page
- 08
01
Why does the demo make building look easy?
One engineer can build a working demo in a few days: call a model, give it two or three functions, put it in a chat window. It answers questions, finds an order, drafts a reply. It is convincing, and it is a small part of what you will run for real customers.
The demo leaves out what makes it safe on real systems: who is asking, what they may do, which steps need a manager’s approval, where passwords for your systems are kept, and what happened last Tuesday at 14:05 and why. None of it is hard to understand. All of it takes time, and none of it can be skipped once the assistant sends messages or changes records.
02
What does a live assistant need?
Use this list to size a build honestly. Each line is something your team designs, builds, tests and keeps running after launch.
- Access to AI models, with limits on cost and a fallback when a provider is down.
- Integrations with each system, such as Odoo, a CRM or Gmail, including sign-in, errors and rate limits.
- Safe storage for the passwords and keys of those systems, separate for each company.
- Roles: who may use which assistant and who may approve its actions.
- Rules per action: runs on its own, follows your rules, or always waits for a person.
- Approvals that can wait hours for a manager and then continue from the same point.
- A log of each request, each system call, each approval and the result.
- Keeping each company’s data apart, checked on every query.
- Channels: the chat on your website, WhatsApp, Telegram, Slack, an API, each with its own setup and failure cases.
- Testing and monitoring: test questions before each change, and a look at failed conversations.
- For the Gulf: Arabic and English throughout, including customers who write in dialect.
03
Build vs buy, side by side
The table is written for a company whose business is not software. If the assistant is your product, several rows weigh differently.
| Build your own | Buy a platform | |
|---|---|---|
| First demo | Days | Hours to days |
| Live with customers | Months: security, rules, channels and operations | Weeks: setup, integrations and review |
| Control over behaviour | Total | Limited to what the platform offers |
| Your in-house systems | Anything you can code against | Built-in integrations plus extension points such as MCP |
| Security and rules | Designed and checked by your team | Included; you check how they work |
| Maintenance | Yours, including model and API changes | Mostly the vendor’s |
| Where data sits | Wherever you decide | Depends on the vendor’s hosting |
| Cost | Engineering salaries, up front and ongoing | Subscription or usage fees |
| Lock-in | To your own code and the people who wrote it | To the vendor, unless you can export |
| People you need | Backend, security, testing, operations | Someone who knows the process and its risks |
04
When should you build your own?
Building is right when the assistant is how you compete, or when your limits rule out every platform you checked. The question is not whether your team can build it. Most can. It is whether they can carry it for years.
- The assistant is your product, or a feature your customers pay for.
- Regulation requires data and keys to stay on servers you run, and no vendor offers that.
- Your process needs behaviour no platform offers.
- You have a team that will own maintenance, security reviews and on-call after launch.
05
When should you buy a platform?
Buying wins when the assistant is a means to an end: faster replies to customers on WhatsApp, quotations ready in Odoo, appointments booked inside the chat. The value is in the work it takes off your team, and every month spent rebuilding approvals and logs is a month without it.
- Your engineering team is small or busy with other work.
- You need rules, approvals and a log from day one.
- You want to see whether a use case pays off before you fund a build.
- You need several channels or several assistants working under the same rules.
06
The middle path: buy the platform, build the edges
The choice is rarely all or nothing. Many companies buy the core, meaning the assistants, rules, approvals, log and channels, and build only what is theirs: a connection to an in-house system, their instructions and their test questions.
Open standards make this practical. With the Model Context Protocol (MCP), your team wraps an in-house system once as an MCP server and plugs it into any platform that supports it. In Botify you add the server by its address. Until someone on your team reviews each of its tools, every call waits for approval, and disabling the server stops all its tools.
It works the other way too. Botify runs its own MCP server, so your developers can create, edit, test and publish assistants from tools such as Claude or Cursor with a Botify API key.
Your MCP servers, instructions and test questions stay yours if you ever change platforms.
07
Questions to ask any vendor
Ask how it works, not whether it is “secure” or “enterprise-grade”. Those words are not answers.
- How are the passwords and keys for our systems stored and encrypted, and can the AI model ever see them?
- Can we make one action, such as sending a quotation, always wait for approval?
- Where can our team approve: inside WhatsApp, in an inbox, or both?
- How is our data kept apart from other customers’ data, and what exactly is checked?
- What does the log record, and can we export it?
- How do we connect a system you do not support: an SDK, a webhook or MCP?
- How does it handle Arabic, including customers who write in dialect or send voice notes?
- If we leave, what can we take with us?
08
Common questions
01How long does it take to build your own?
A demo takes days. A live system with roles, rules per action, approvals, a log, safe credential storage and monitoring usually takes months, depending on how many systems and channels it touches. Plan as much effort for running it as for building it.
02What does building really cost?
Engineering time for the platform parts, model usage, hosting, security reviews and ongoing maintenance as models and APIs change. Teams usually underestimate the platform parts, not the model bills.
03Can we start on a platform and build later?
Yes, and it lowers the risk: you learn what your customers and staff really need before you fund a build. Keep what moves with you, such as instructions, test questions, documents and any MCP servers you write.
04Do we need a machine learning team?
Not for most business assistants, which use hosted models instead of training their own. You need someone who owns the process, someone who knows the systems being connected, and someone accountable for rules and approvals.
09