How to Start Effectively your AI Project

Many teams try to initiate an Artificial Intelligence project by applying algorithms and data before determining the desired results and objectives.

Share Give it a Spin!
Follow by Email

Many teams try to initiate an Artificial Intelligence project by applying algorithms and data before determining the desired results and objectives. Unfortunately, that’s like raising a puppy in an apartment in New York City for a few years and then being surprised that he can not hunt rabbits for you.

You can not expect to get something useful by asking some attendees to produce machine learning magic in your business without an effort on your part first.

How to start

Instead, the first step is for the owner, get a clear vision of what he wants from his ML / AI system and how he will know that he has trained it successfully.

There are very extensive reference guides in this regard, so we try to compact the content to find a truly useful one.

Find out who is in charge of the project is paramount. The tasks that we are going to address are the responsibility of the project manager. If a doctoral researcher is selected for this role, it must be due to that person’s decision-making abilities and to the deep understanding of the business. It could be a person or a committee. Choosing wisely to this entity is fundamental.


Identifying the use case is another factor to take into account. The key is that learning machines or deep (ML, in English) and Artificial Intelligence (AI, in English) is not magic and does not solve all problems. It is a tagger of things and it is up to you as the owner to find out what needs labeling.


Labeling things does not mean classification, it refers to output. It could be a category, a number, a sentence, a waveform, a sequence of actions, etc.

Do not hire that Ph.D. guru before confirming that you need it. Focus on the outputs first.


Third, you must do some reality checks. Once you can clearly articulate what labels you are looking for, it’s time for a quick reality check. No access to the data means that there is no point in the procedure. However, you may be able to get what you need online: there is a growing tendency to make the data available for free.

Performance metrics

Then elaborate a performance metric wisely. The next bit can be a bit complicated if you are new to it. You are in charge of deciding how much each type of result is worth. Grab someone who likes numbers and ask them to help you exchange ideas.

Qualitative experts are specially trained for this, but your standard calculator-slinger will work if necessary. If you want the best helper, pronounce the formal jargon for it by causing indifference curves aloud in a pentagram to summon an economist.

Economists make advisors surprisingly useful for artificial intelligence projects.

Test criteria

Finally, establish test criteria. Indicate your population of interest. Talking about “running” your system does not make sense until you specify in which instances you want to work. That means specifying the statistical population of interest. Now that you have your performance metric and your population at hand, you have more work to do before you can put your feet up.

Decide which is the minimum performance you are willing to sign, because you will not let that system take charge of labeling things for you unless it is good enough.

Setting criteria in advance is part of how you keep yourself safe from the difficult machine learning and artificial intelligence concepts.