top of page



Applying Artificial Intelligence to a business process or operation appears like a daunting task to most businesses or even developers. The uncertainty of the success rate of applying Artificial Intelligence can be a fundamental part of the fear and hence most business avoid going into AI. This is to some extent understandable. Up to a few months ago, the process of applying AI was full of investigations to identify whether AI can really be of help in such a situation. Hence a lengthy analysis would be required, with the process comprising of

  • collecting the data

  • cleaning the data

  • researching the correct AI models to use

  • train the model selected

  • compare the models and evaluate whether the model outputs satisfactory results

The steps would be required before even knowing whether the implementation of AI would be successful, hence it could be a wasted effort. In this regard, the two major cloud providers have identified a gap and are building an arsenal of tools which bootstrap the above process. Microsoft Azure cognitive services and Google AI Platform are the two platforms which provide API integration for generic trained AI models to be consumed by third parties. These can be easily integrated into applications and workflows to harness AI power.

Generic use cases have been identified and AI models built for them to be consumed, such as:

  • identifying anomalies in the data

  • language translations

  • text analysis for key phrases or sentiment

  • translation of speech to text and vice versa

  • content recognition in images/ videos and other services related to vision

  • speech and basic decision analysis

These providers have done well in identifying a set of generic cases where AI can be used to extract more valuable information and which can then be integrated into more specific workflows with one case being text recognition from images which can be used in digitisation of documents, cheques or labels.

The more niche the use case gets, the less these generic models provide successful results.

The models used provide good results in cases where your dataset follows similar patterns as the training data. Any deviations from such norm would results in lower accuracy because the training data is not representative of your dataset, or in more advanced cases the model might not even be appropriate. In such cases, some services even offer the facility to train the model on your dataset which would achieve better results.

As discussed briefly, AI is becoming more accessible with providers reducing the entry barrier for smaller businesses and enabling a faster return on investment. These models are good for prototyping a model and reducing time to market considerably with time being invested to analyse results and plan for longer term solutions while building your AI platform with knowledge gained during this journey.

14 views0 comments


bottom of page