This is not a ‘stack’ it is a demo app for a bunch of Azure features
> This project is a proof of concept. It is not intended to be used in production. This demonstrates how can be combined Azure Communication Services, Azure Cognitive Services and Azure OpenAI to build an automated call center solution.
Literally the only reason in want to ring a call centre is because the online forms got me nowhere, the “press 1 for…” robot got me nowhere, and I’m desperate to speak to someone sentient, even if they are just reading from a script.
For people unfamiliar with this area: you can do all this with AWS, and have been able to for a while now. I left the industry a few years ago, but at the time people were still waiting for Microsoft's response because Amazon were clearing up in the cloud contact centre space.
The problem with modern LLMs and AI expectations in this area is that 99% of call centre calls can typically be handled with a basic yes/no/"get your answer from here instead" type response and deflections that doesn't need really fancy AI.
"Well great, LLMs can deal with the really hard 1%" - nope, they're the high value, big impact issues that you want your best remaining humans to handle and learn from to feedback into the business.
This will be an absolute game changer in the phone call scam space. AI scams were already a thing, but now Microsoft made a proof of concept to replace their custom, terribly constructed software.
Automate outgoing call. Get a record of your voice. AI synthesize your voice. Call your family members in your voice, in distress. Scams of biblical proportions will happen in 2026.
Like anyone else I am also annoyed by customer service experience over the phone. However if you're on the other side of CS calls you'll notice so many calls that are for simple things that the user could easily do online. Those trivial calls takes the time from CS representatives to address more important queries. If AI CS is done right, it would be mostly a filter for helping human CS folks only deal with real issues rather than "how do I log into my account? I'm on your blog page..."
One annoying part of the new AI world is these worthless AI-generated diagrams. I don't mean AI generated the pixels, I mean it was asked to generate docs, maybe even a "system diagram", and the result is just noise.
I wonder how many risks there are with prompt injections and "social engineering" with these. If you were persistent enough how long would it take to trick it to give you some information it should not. And would company even monitor for that?
While so far automated customer service has been pretty bad, I can see why nobody wants to actually take the calls. There are some real rage-beast assholes out there. Blame them for ruining it for everybody.
Yeah I definitely want to speak to an AI bot when I've just been in a car crash. /s. There will be a place for this though, for people who either can't afford or don't want to pay for better.
On one hand this is lovely code and seems to work well, on the other hand there are currently estimated to be 2.86 million call center staff in the US alone.
Did we really let customer service get so bad that people think this will be an improvement? My gut feeling says this will not transform companies who are bad at customer service to improve, rather the ones that are already bad are now just bad with a lower headcount. Curious to hear any positive expectations for this.
Microsoft Releases AI Call Center Stack with Voice, SMS, and Memory
(github.com)42 points by bakigul 11 hours ago | 60 comments
Comments
> This project is a proof of concept. It is not intended to be used in production. This demonstrates how can be combined Azure Communication Services, Azure Cognitive Services and Azure OpenAI to build an automated call center solution.
The problem with modern LLMs and AI expectations in this area is that 99% of call centre calls can typically be handled with a basic yes/no/"get your answer from here instead" type response and deflections that doesn't need really fancy AI.
"Well great, LLMs can deal with the really hard 1%" - nope, they're the high value, big impact issues that you want your best remaining humans to handle and learn from to feedback into the business.
This may not be the thing that kills them, but it’s only a matter of time.
This is also entirely predictable. We’ve known this was coming, for at least two years.
One annoying part of the new AI world is these worthless AI-generated diagrams. I don't mean AI generated the pixels, I mean it was asked to generate docs, maybe even a "system diagram", and the result is just noise.
Curious about the future where customers are also AI.