AI phone assistant:
build it yourself
or have it managed?

Building it yourself pays off if your calls are simple and predictable and someone on your team has time to build, test and maintain the assistant. Having it managed pays off if the assistant needs to connect to a calendar, ticket system or industry software, must recognise emergencies reliably, or nobody on your team can take on maintenance.

Both routes can lead to a good result. What matters is less the technology than who does the work around it: preparing knowledge, setting rules, testing, clarifying data protection and keeping the assistant up to date.

Build it yourself

You use a platform with an editor in which you create the greeting, knowledge and workflows yourself. Or you assemble the assistant from separate services for telephony, speech recognition, language model and speech output. The platform provides the technology; building, testing and maintenance are up to you.

Have it managed

A service provider sets the assistant up with you, connects your systems, tests it with typical calls and adjusts it in day-to-day operation. You provide knowledge and decisions; the provider does the implementation.

There are hybrids in between: some platforms refer agencies that set it up for you, and many service providers build on such platforms. The tasks are the same on every route. What changes is who does them, when you can first try it out and how the costs are spread over time.

The costs are made up of a few components. Which of them apply depends on the provider and the route.

If you build it yourself, add your own time: for building, testing and maintenance whenever opening hours, services or responsibilities change. Managed offers usually only quote prices after a conversation, because scope, connections and support differ from business to business. That also applies to Humaify.

With every offer, check whether prices are net or gross, the minimum term, the cost of the phone number and what happens when you receive more calls. Also compare the costs over the whole term, including your own time, rather than just the entry price.

Whether you build it yourself or have it managed, these steps belong to every good start.

1. 01

   Collect calls

   For two weeks, note who calls, with which request and at what time. That shows what the assistant needs to be able to do.

2. 02

   Prepare knowledge

   Opening hours, services, responsibilities and frequent questions as short, unambiguous answers. The assistant should only answer from approved knowledge.

3. 03

   Set rules

   What is an emergency, who is informed when, what may the assistant promise and what never?

4. 04

   Greeting and AI notice

   The assistant says right at the start that it is an AI and names your company.

5. 05

   Test

   Run through typical and difficult calls, including dialects, background noise and impatient callers.

6. 06

   Set up call forwarding

   Usually you forward your existing number, for example when nobody answers after a few rings or outside office hours. Your customers generally do not need a new number.

7. 07

   Launch and maintain

   Read the call summaries during the first weeks and adjust knowledge and rules. When opening hours or services change, the assistant needs to know too.

A phone assistant processes personal data: names, phone numbers, addresses and the request itself. For use in Germany, clarify these points before launch:

• Data processing agreement: You need a contract under Art. 28 GDPR with the platform or provider.
• Data locations and third countries: Where are calls processed, and which subprocessors are based outside the EU?
• Recording or summary: A written summary is often enough. Recording a call generally requires the caller's consent.
• Data minimisation: The assistant should only ask for what the request needs and delete data after fixed periods.
• Privacy notice: Add the phone assistant to your privacy information.

AI disclosure under the AI Act

Article 50 of the EU AI Act has applied since 2 August 2026. AI systems that talk directly to people must be designed so that callers learn they are talking to an AI, clearly and at the latest at the start of the conversation. The obligation is addressed to the providers of such systems. As a business, make sure your assistant introduces itself as an AI in its greeting, for example: “Hello, you are speaking with the AI assistant of Example Ltd.” The only exception is when this is obvious to callers anyway; on the phone you should not rely on it.

This section is an overview and not legal advice.

Building it yourself suits you if

• your calls cover a few clearly defined requests
• someone on your team has the time and interest to learn prompting and to build and maintain the assistant
• no connections or only common ones are needed
• you want to try it first on a small budget

Having it managed suits you if

• the assistant must recognise and pass on emergencies reliably
• a calendar, ticket system or industry software should be connected
• nobody on your team can take on maintenance for the long term
• you want one contact for changes and data protection questions
• you want experts to build and structure your assistant, which usually means better quality in exchange for a higher one-time setup cost

Questions for every provider

• Where are calls processed and stored, and for how long?
• Is there a data processing agreement and a list of subprocessors?
• How does the assistant introduce itself as an AI?
• What happens when the assistant does not understand a request?
• How is it tested before launch, and who adjusts it later?
• What does it cost when calls increase, and how long does the contract run?
• Can you take your data and settings with you if you switch?

To be clear: Humaify offers the managed route. We have still written this guide so that you may conclude that building it yourself is right for you. If you would rather have the assistant managed, this is how we work:

1. 01

   Discovery

   We look at your calls, workflows and systems.

2. 02

   Design

   Knowledge, greeting, emergency rules and handovers are created together with you.

3. 03

   Prototype

   You hear the assistant early and give feedback.

4. 04

   Pilot

   It answers its first real calls; we review them with you and refine.

5. 05

   Go-live

   From then on we support it in day-to-day operation and keep knowledge and rules up to date.

We plan around six weeks until go-live. The initial call is free, and you will hear back within one working day. After that, your effort is mainly approvals and feedback on calls. For knowledge, opening hours and similar details, we set up interfaces through which your changes take effect immediately. We make changes to the assistant's behaviour together with you.

• Regulation (EU) 2024/1689 (AI Act), Article 50: transparency obligations for AI systems that interact with people; applies from 2 August 2026 under Article 113
• Regulation (EU) 2016/679 (GDPR), Article 28: processors