Providing your client with insights, briefly defined as short texts of analytically processed information, is a valuable addition to the services provided by virtually any company. Unfortunately, as engineers, or technicians in general, our training does not address in detail the techniques for writing insights. This short text seeks to serve as a basic guide for future analysts. I introduce the concept of insight and provide advice for the creation of concise and short intelligence pieces.
As a senior data analyst, I must do precisely what Ray Dalio, finance magnate, mentions in his December 2019 conversation with Lex Fridman in his podcast “Artificial Intelligence” when asked what role machine learning will play in making decisions and in the analysis:
“AI is particularly good at processing data, but the notion of what you want to do is most of the time better determined by the human mind. What are the principles? ” (Approximate quote) - Ray Dalio in an interview with Lex Fridman
TSC.ai (where I work as a Senior Data Analyst) is a technology company
that uses artificial intelligence to provide, precisely, intelligence to our
customers. In part, this intelligence is provided through insights.
Be careful, an insight is not the same as a report or a thorough investigation. An insight is a short text, less than 2,000 words, that provide intelligence to a client who does not expect to read a 10,000 paper or read five news to arrive to a conclusion on a topic. Examples of what are perhaps the most famous insights are in the Bloomberg QuickTake series.
The insights provided to the client require information processing for their creation. For the information processing the analyst can use tools such as articial intelligence, for example Atium of TSC.ai, however, this is not enough. The AI has serves as an invaluable tool for the creation of intelligence, it is formidable in the work of processing data for the prediction or identication of patterns. However, we are at a point where AI is not yet excellent in decision making that involves considering a large number of variables that evolve over time (playing Go or Poker does not apply): As Dalio says in the same interview, you will not allow an AI to dene how you are going to educate your children.
Once data is processed by the AI, the analyst has the decision of how to use the information, if he uses it, to make it useful for the client. Think, for example, of an NLP algorithm that identies names of persons in news and creates a summary of the story; in this case, when carrying out the analysis, it must be decided whether, given the fundamentals of what is important for our client, the news is of sufcient relevance or not, and whether the individuals identied are relevant to the story we want to provide to our client

Ajay Agrawal, Joshua Gans and Avi Goldfarbo, make it clear in “Prediction Machines: The Simple Economics of Artificial Intelligence”: although prediction (provided through AI) is a key component of intelligence, it is not one in the same.
To provide intelligence, artificial intelligence is not enough. Insights are precisily that and can only be made by humans.
Next, I will present some relevant points that my team and I have found for performing quality insights. But, first of all, remember the main rule:
Put crudely, the objective is to develop quality insights that will be sent to very, very busy people (customers) who want to see the note in two minutes.
The work of the analysts will be based on three fundamental pillars which I adressed here: 1) search for information, 2) writing, and 3) content structuring.

1) Search for information

2) About the writing

The quality of an insight is not in its extension but in its content: clear ideas, short and successful writing.

3) Content structuring

Since the content of the insight depends on the addressed topic, there is no recipe for the structure of the insight. Therefore, the creativity of the analyst and his own initiative is appreciated at the time of bashing the ideas in a coherent and orderly speech.
Here is the general structure I recomend. For the beginning of the insight:
For the main body:
Avoid writing a lot and saying little. That is, go to the point.
  • It can contain as many parts as the analyst considers, as long as they are directly linked to the subject under investigation and offer accurate and important information.
  • A clear wording is essential to understand ideas perfectly. Write as if your insight was directed at someone who doesn't know anything about the topic. Force yourself to explain thoroughly in the least possible words; I know, it requires work.
  • While it is true that information sources will be used, please do not copy and paste!
  • The argument must be taken care of; that is, maintain the logical (or chronological, if necessary) order of ideas as they develop.
  • Try to organize your ideas in bullets and if they branch out another level of bullets can be used to show that the idea has "subideas". In my opinion, there is no limit to the use of levels, you just need to maintain the relationship and order.
  • Our clients are generally interested in knowing who are proponents and who are critics of certain topics, for example, regulations, so we provide a list of interested parties and their positions on the topics in question. This can be beautifully summarized in a relationship map as provided by TSC.ai. As an example, look at this map about who is who in the 2020 regulation of fintechs in Argentina:
  • Conclusion: Depending on the insight requested, it may close with a brief conclusion, usually a paragraph, which will indicate the most relevant findings observed during the investigation. It may also contain an analysis of its potential future consequences (it will depend on the topic).
  • Recommendations: Suggestions will be established if there are any or if they are required. In this phase the analytical capacity of the one who makes the insight is crucial.

    • Finally, remember...

    Artificial intelligence is an excellent tool to provide intelligence to your customers. However, most of the time it will not be enough; It will take the human component that processes the information. In this short text, the concept of insight is introduced as a piece of short text that provides concise information that, ideally, can be read in two minutes.