Copilot Studio for voice: build it, or buy it?
In short. Copilot Studio can answer phone calls. But Microsoft’s own documentation routes voice through Dynamics 365 Contact Center, you supply a number from Azure Communication Services, and you build and maintain the agent yourself. Enquiry 360 is the other option: a finished application from AppSource that installs into your own Microsoft Dataverse, needs no Dynamics 365 Contact Center licence, and has a standard process answering the phone the same day. Which is right depends entirely on whether you want to build or to buy.
We sell one of these two things, so we have set out both cases and left the choice where it belongs.
What Copilot Studio actually does for voice
Two kinds of voice agent. Basic voice agents use classic orchestration, with natural language understanding for intent recognition and entity extraction. Real-time voice agents use a speech-to-speech model for lower latency and a more natural back and forth. Both handle speech and keypad input, context variables and call transfer.
It is a capable platform and it is going to keep getting better, because it is where Microsoft is putting its money.
What the phone line actually requires
This is the part that surprises people. Microsoft’s voice documentation is titled around integrating with Dynamics 365 Contact Center, and states that before you can create or edit agents for voice you need a phone number, obtained through Azure Communication Services.
So the shopping list is Copilot Studio, a number, and a Dynamics 365 Contact Center licence. For an organisation already running a contact centre on Dynamics that is no obstacle at all. For a business with fifteen staff and a phone that goes unanswered at six o’clock, it is a different proposition entirely.
And then you build the agent
The licence gets you the ability to start. What you are signing up to build is the process itself: the topics, the questions, the order they are asked in, the branching, the validation of what the caller says, the escalation rules, and the data model that the answers are written into. Then you own it, and you maintain it every time the process changes.
That is entirely reasonable work if building it is what you want. It is a poor use of a month if what you wanted was the phone answered.
And then you build the next one
An agent is built for a process. A second process means a second agent, with its own topics, its own branching and its own maintenance. That is fine for one thing. It compounds badly when a council runs forty services, or an outsourcer answers for fifty clients and each one asks a different set of questions.
Enquiry 360 works the other way round. One application reads a form definition and runs it, and the definition is data held in your own Dataverse. So a new service is a new record rather than a new build, and the fortieth costs about what the first one did. The same application answers a bulky waste booking, a housing repair and a dental enquiry, because the only difference between them is configuration.
The hard part is not the conversation
Sounding natural is the part everyone demonstrates. It is not the part that goes wrong. What goes wrong is the value.
The values worth getting exactly right are the ones with a shape: a postcode, a ZIP code, a National Insurance number, a tax reference, a case number. Getting those right, every time, is what this is built around, because the cost of not doing turns up later as a callback, a missed appointment or an engineer sent to the wrong address.
So the engineering that matters sits between hearing and storing, and it is the part of this that took longest. Every value is checked against the rule the form itself declares, on every turn, so a malformed answer cannot be written. A spelled value keeps the caller’s own letters and digits. An answer that matches none of the offered options is asked again rather than guessed at. And where the form asks for it, the value is read back and confirmed before it is stored.
None of that came from a specification. Every one of those rules is there because a real call went wrong and we went and fixed it, which is the part that cannot be shortcut.
Do not take our word for it. Ring 01924 926000, spell something awkward at it, and ask what it stored.
Where this one is the better choice
- Nothing to build, so nothing to wait for. You describe your process rather than construct one.
- You have no Dynamics 365 Contact Center licence and no plans for one. It runs on Power Apps and Microsoft 365 plans.
- The process matters more than the conversation. It holds a natural conversation, and it is built so that what it stores at the end is right: every answer validated as it is given, and read back where the form asks for it.
- You want the same process on screen too. Your staff work it in a browser, the voice agent works it on the phone, and both write the same record.
- You want to change it yourself. Change a form, and the phone behaviour changes. No release, no developer.
Where Copilot Studio is the better choice
- You want to build something bespoke and own it. That is what it is for.
- You need generative answers over your own knowledge. Enquiry 360 answers callers’ questions from the information you configure on the form, and says so plainly when a question falls outside it. Pointing an agent at SharePoint, your website or a whole document library and having it answer from those is Copilot Studio’s territory.
- You want one agent across Teams, web chat and voice. We answer the phone and we run the same process on screen. Teams and web chat are channels Copilot Studio covers and we do not.
- You already run Dynamics 365 Contact Center. The licence question disappears and the integration is native.
- Latency is the thing you care about most. The real-time speech-to-speech agents use a lower-latency architecture than ours does today.
Where they work alongside each other
Framing this as a choice is the wrong shape for a lot of organisations, because the two do not compete for the same slot. Copilot Studio builds agents across Teams, web chat and your own knowledge sources. Enquiry 360 answers the phone and turns the call into a structured record. Because that record lands in Microsoft Dataverse in your own tenant, a Copilot Studio agent can read it, report on it and act on it: the phone becomes the front door, and whatever you have built picks it up from there.
Both are Microsoft Power Platform citizens and both live in the same tenant, so running the two together is a division of labour rather than a compromise. Plenty of the better setups look exactly like that.
The honest summary
Copilot Studio is a platform. This is a product. Platforms are the right answer when your requirement is unusual and you have people to build with. Products are the right answer when your requirement is ordinary and your constraint is time.
Most businesses asking us this question have an ordinary requirement and no spare capacity, which is why they end up here. If yours is genuinely unusual, use Copilot Studio.
Frequently asked
Can Microsoft Copilot Studio answer phone calls?
Yes. Copilot Studio supports interactive voice response through two kinds of voice agent: basic voice agents using classic orchestration, and real-time voice agents using a speech-to-speech model. Both support speech and DTMF input, context variables and call transfer.
What do you need to put a Copilot Studio agent on a phone line?
Microsoft’s own documentation routes voice through Dynamics 365 Contact Center, and you need a phone number, which you get from Azure Communication Services. So the realistic shopping list is Copilot Studio, a number, and a Dynamics 365 Contact Center licence, plus the work of building and maintaining the agent itself.
Is Enquiry 360 an alternative to Copilot Studio?
It can be, though the more useful answer is that they sit alongside each other. Copilot Studio is a toolkit for building agents of many kinds across many channels. Enquiry 360 is a finished application on Microsoft AppSource that installs into your own Microsoft Dataverse and answers the phone. If you want a configured process taking calls without building an agent, this does that job instead. If you are already building in Copilot Studio, this takes the phone and leaves the rest to you.
Can we run Enquiry 360 alongside Copilot Studio?
Yes, and it is often the sensible shape. They are not competing for the same slot. Copilot Studio builds agents across Teams, web chat and your own knowledge sources. Enquiry 360 answers the phone and turns the call into a structured record. Because that record lands in Microsoft Dataverse in your own tenant, a Copilot Studio agent can read it, report on it and act on it. Running both is a division of labour rather than a compromise.
When is Copilot Studio the better choice?
When you want to build something bespoke and own it. When you need generative answers over your own knowledge sources. When you want one agent across Teams, web chat and voice. When you already run Dynamics 365 Contact Center. And when latency matters more than anything, because the real-time speech-to-speech agents are a lower-latency architecture than ours.
How long does it take to get a working phone process?
There is nothing to build, so there is nothing to wait for. It installs into your Microsoft tenant and you describe your process rather than construct one. Building the equivalent in Copilot Studio, with the questions, the branching, the validation and the data model behind it, is a project.
Can one AI voice agent handle many different services?
Here, yes. One application reads a form definition and runs it, and the definition is data held in your own Dataverse, so a new service is a new record rather than a new build. Building agents means one agent per process, each with its own topics, branching and maintenance, which compounds when a council runs forty services or an outsourcer answers for fifty clients.
How do you stop an AI voice agent storing the wrong value?
Accuracy on values that have a shape, like a postcode, a ZIP code, a National Insurance number or a case reference, is where this is won or lost. Every value is checked against the rule the form itself declares. A spelled value keeps the caller’s own letters and digits. An answer that matches none of the offered options is asked again rather than guessed at. And where the form asks for it, the value is read back and confirmed before it is stored. The quickest way to judge any of it is to ring the line, spell something awkward at it, and ask what it stored.
Do we need Dynamics 365 to use Enquiry 360?
Enquiry 360 installs into the same Microsoft Dataverse that Dynamics 365 runs on, so it sits alongside Dynamics 365 where you already have it, and stands on its own where you do not. It runs on Power Apps and Microsoft 365 plans, without a full Dynamics 365 licence and without Dynamics 365 Contact Center.
What are the alternatives to Copilot Studio for voice?
For building, Dynamics 365 Contact Center is Microsoft’s own route, and the independent conversational AI platforms are the rest of the field. For buying, the alternative is a finished conversational AI voice agent that already runs your process on the phone, which is what this is. It is a voicebot, a virtual agent and an IVR in one, sold as an application rather than as a toolkit. The split is the one set out above: platforms suit an unusual requirement when you have people to build with, products suit an ordinary requirement when your constraint is time.
Hear it for yourself. Ring 01924 926000 and hear a configured process answer. It is a live line, not a recording.
From £99 + VAT a month with phone answering, or take the app on its own from Microsoft AppSource and add the voice later. Or read about replacing Dynamics 365 Dialogs.