We know that A.I. has been kind of a buzz word for the last couple of years. But truth is although several projects have failed, many others thrived. We identified some characteristics within the success cases on data extracting applications, and share here 4 steps they have in common:

1) Assess current processes

This might sound obvious for some people, but it is amazing how we forget about assessing our current processes. Any technology, including A.I., can help a lot when it comes to automation, but they are not a magic spell. This step has 2 main purposes: first, of course, to identify where things don't go so smoothly and second to identify root-causes of current pains. Applying new technologies in processes that already have room for improvements or require adjustments can cause sub-optimal results, hide the real problems (current operation inefficiencies) and make user claim that the new technology simply doesn't work.

2) Choose which pain points to tackle

Once you know your pain points and what causes them, it's time for choosing. It is very important here to have focus and pick a small number of problems. Planning is key to not only understand your capacity but also choosing what to work on and what not to. In general, it is a matter of classifying, prioritizing and "time lining". There are many tools available to carry on these tasks. A special mention on prioritization: I like to use the bidimensional outcome x effort matrix, where we can classify all initiatives simultaneously by their potential outcomes and by the workload required.

3) Redesign the processes in the light of A.I. applications

This is the less obvious of the 4 steps. When I say redesign the process, is not just to adjust the process to accommodate the A.I. piece. It is literally redesign the process taking in account that an A.I. application will be part of this workflow. And why this holistic view is so important? Because the logics involved in many A.I. applications (such as N.L.P., computational vision, among others) might be very different to our human empirical logic, and therefore a task that is difficult for humans can be easy to our application and the other way round is also truth. Therefore if you want to achieve great results you must build your process taking in account the characteristics of data and technology to be used.

4) Test, test, test ... and deliver

Finally, can't stress more the importance of testing the model, hoe the A.I. piece interacts with the other pieces, talk to users and deliver the new process. An important point here is to deliver something that is aligned to users will. It is very important, beforehand to demystify the technology (remember it is not a magic spell). To agree on some basic premises such as SLA, accuracy rate among others, is key to be able to go live with the new process, and enhancing the results as the process is being used.

I hope these 4 steps are helpful for you. If you have any ideas or experiences implementing A.I. applications in the data extraction space please share them in the comments.

All the best,

Stephan