With the various options companies have for hosting their AI platforms, they must ensure the method they select actually meets their needs. Today we look at one of these options in particular, on-premise, and outline three times businesses might benefit from this method and when they may want to select another.
You have and will continue to receive tons of data.
To learn and improve, machine learning and other AI models need a lot of data. Cloud APIs are beneficial when you don’t have a lot of data to start with, as the cloud “consolidates knowledge” from all the existing models in the API. That is, as each model learns they add to the knowledge base of the platform. Your new model will, therefore, have that knowledge pool to draw from.
If you have a lot of data to start with and build your model, that's great, but think of it as a library. While the New York Public Library has some 50 million items in its inventory, every year, a seasoned group of selectors adds and removes thousands. This is to ensure they “create a well-rounded collection with a wide range of voices and viewpoints.” Your data pool can be thought of similarly. If you choose on-premise, stagnant data pools will leave you with weaker models, so it’s best to have a consistent stream of new data.
The data you use is very sensitive.
One reason businesses like retailers or online marketplaces can benefit from using a cloud API is that their data is not particularly personal. For instance, while a clothing retailer may carry unique patterns or designs, depending on the use case, the data concepts they’ll need (like shirts, shoes, bags) may not that different from those of a different company in the same industry. Two clothing retailers who want to add visual search to their platform could, therefore, benefit more from combining the knowledge each of their models collects in a cloud.
Where an institution deals with much more personal data, on-premise may be the better option. While many Cloud API providers adhere to stringent data protection policies, those regulations may be more general than you need. Institutions like banks and insurance companies, for instance, often have rules that are specific to their industry. By keeping their data in-house, they can better ensure that they are meeting industry standards.
Your data or use-case is very distinct and you have adequate hosting facilities in place.
Some companies have data or use-cases that are so unique, they likely won’t benefit from the consolidated knowledge of the Cloud API. For instance, Ars Italica, one of the world’s foremost caviar producers, performs thousands of scans every year on the multiple breeds of sturgeon they rear, as they cannot otherwise accurately identify the males and females in the species.
Critical as this activity is to their business, however, they are one of only a few prominent sturgeon caviar producers. If they decided to use computer vision to help them with this, this would a very rare use-case. As such, if you have a similarly uncommon use-case and already had adequate hosting facilities in place, on-premise is a viable option for you.
That said, cloud APIs offer other benefits, e.g., your AI provider will be responsible for hosting and so responsible for the time & costs associated with hosting, maintenance, and improvement. If you don’t already have your own hosting facilities in place, then you should consider cloud APIs.
With the level of control that on-premise offers, many businesses might think the on-premise option is best for them. However, as shown above, there are other points to consider before making that decision. Working with the right team of AI experts can help businesses to determine what will work best for their needs.