Azure Cognitive Services – What are they?

I’ve recently started working for a client that required me to dive into Azures Cognitive Services. This post is all about what those services are, who they’re aimed at and how much it costs!

What are Azure Cognitive Services?

Azure Cognitive Services are Microsofts offering in the assisted AI space. Specifically, they offer API’s that allow you to get smart insights into an image, videos or sounds on a pay per call basis. The API’s that have been most useful to me so far are –

Who should use Azure Cognitive Services?

The audience for Azure Cognitive Services is very clearly developers who have had some exposure to Machine Learning concepts. Microsoft have worked hard to abstract away as much of the detail as possible meaning that a high level understanding of the ‘process’ is enough to train a labelled model (for Form Recognizer) as well as providing some APIs with nothing more than an endpoint (like Face).

Developers don’t need to know the science behind Machine Learning to use Azure Cognitive Services and to be honest if you are coming to these services from that angle, I think you will be disappointed. The services do not offer anywhere near the level of control that a dedicated ML service would and any Data Scientists looking to work in Azure should take a look at the Machine Learning service where you will be much more at home.

How much does Azure Cognitive Services cost?

As with all Azure services, the pricing model is pretty messy and each service has its own version. A couple of key points in the pricing area though. Ultimately though, the audience of this blog is going to be more interested in the free tiers for development and learning!

The good news is that there are Free Tiers available, so check out the Microsoft pricing pages to see what the latest freebie level is. Right now is pretty good, for example Face API is 20 transactions per minute and 30,000 free per month. Form Recognizers is up to 500 free pages per month.

That said, be mindful that a document might be more than 1 page, if you uploaded a 500 page pdf your quota would be gone in 1 hit. Similar story with Face API, a request for multiple features like detection, verification and identity all count as 1 transaction each. If you’re chaining requests you get eat up your quota much quicker than you realise.

If you go to production you will need to be very careful about your anticipated use case. For example the Computer Vision API charges £750 for a million face operations. If you write an Azure Function to identify, tag and verify every image that comes in, this means you are being charged £750 per 333k images. This sounds like alot, but you’d be amazed how quickly you can burn through this at scale and this is just 1 API. If you adopt several, your bill can easily end up a grand per API for large scale enterprise adoption, so make sure you have your business benefit quantified to justified an ongoing OPEX to your boss!

Summary

For Developers with high level ML knowledge, this is an amazing set of APIs. I will link some additional ‘how-to’ blogs in future but I can attest that I have been able to quickly train models using Form Recogniser and integrate them seamlessly into Power Automate to almost entirely remove some manual processes (a clear ROI is easily sellable!). I’ve not had the pleasure of working too closely with the text sentiment or Face APIs yet, but pretty excited about the possibilities around being able to intertwine our digital and physical worlds abit closer!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.