Is your company data valuable in the AI era?
Hello again! Today we continue the series of publications dedicated to the launch of the course “Big Data for Managers” . So, let's begin.
“AI is close.” This is what we hear from 2017 and, most likely, we will continue to hear further. For established companies that are not Google or Facebook, a natural question arises: what do we have, what will allow us to survive this transition?
In our experience, the answer is “data.” This view is held by the business press. Hundreds of articles have been written, stating that “data is a new oil,” implying that it is the fuel that will stimulate the economy of AI.
')
If so, then we can assume that your company was lucky. You collected all this data and when the AI finally appeared, it turned out that you were sitting on oil reserves. But if you’re really so lucky, you may want to ask yourself: “Are we really so lucky?”
In the analogy “data is oil” there is a grain of truth. As a fuel for an internal combustion engine, data is required for AI operation. The AI takes the raw data and turns it into something useful for decision making. Want to know the weather for tomorrow? Let's use the weather data for the previous period. Want to know yogurt sales next week? Let's use the past yogurt sales data. An AI is a data-driven prediction engine.
But does AI need your data? Today it is assumed that all data can potentially be useful for AI, but in fact it is not. Yes, data is needed for the daily operation of your forecasting machine. But most likely it is not the data that you have now. Instead, your company is accumulating data that will be used to build a forecasting machine , and not to operate it.
You now have training data. They can be used as a learning material for the algorithm. And already this algorithm is used to generate forecasts for committing actions.
That is, yes, it means that your data is valuable. But this does not mean that your business will survive the storm. Once the data is used to train the prediction machine, it depreciates and becomes useless for this kind of prediction. Continuing the analogy with oil, the data may burn. They are lost after use. Scientists know this. They spend years collecting data, but as soon as they give results, they start collecting dust on a shelf or a forgotten flash drive. Your business may be sitting in an oil well, but its reserves are limited. This does not guarantee you anything more in an AI economy than just a more profitable residual value.
No matter how valuable your data may be, the ability to benefit may be limited. How many sources of comparative data are there? If you are one of the many suppliers of yogurt, then your databases that contain information on the sale of yoghurts over the past 10 years and related data (price, temperature, sales of related products, such as ice cream) will have a lower market value than if you are the sole owner of this data. In other words, as in the case of oil, the more suppliers that have data similar to yours, the lower the value of your training data. The value of your training data is further influenced by the value obtained due to the increased accuracy of forecasts. Your training data will be more valuable if improved prediction accuracy increases yogurt sales by $ 100 million, not just 10.
Moreover, the current value of the data usually depends on the actions performed in everyday business - new data received every day that allow you to use your machine for forecasting after training. It also helps to improve it through training. 10 years of yogurt sales data - useful for teaching an AI model to predict future yogurt sales, but real predictions used for supply chain management require operational data on an ongoing basis. And this is an important point for today's companies.
An AI startup that acquires past yogurt sales data can train the AI model to predict future sales. He will not be able to use the model for decision making if he does not receive current operational data for training. Unlike startups, large corporations generate operational data every day. This is valuable. The more operations, the more data. In addition, the owner of the transaction may actually use the prediction to further improve future operations.
In the economy of AI, the value of your accumulated data is limited by the one-time benefit from learning the AI model. And the value of training data, like oil, depends on the total number - the more people own them, the less valuable they become. In contrast, the value of your current operating data is not limited to a one-time gain, but rather provides a constant benefit in the operation and subsequent improvement of the predictive machine. Therefore, despite all the talk that the data is new oil, your old accumulated data is not the main thing. However, they can lead to the point. Their value to your prospects is low, but if you can find ways to generate a new, steady stream of data that provides a functional advantage in terms of the predictive ability of your AI, it will give a stable advantage when it appears.
Ask questions, write your comments, and do not forget that tomorrow, April 10, there will be an open door , which will be held by the CEO, CleverDATA - Denis Afanasyev .
“AI is close.” This is what we hear from 2017 and, most likely, we will continue to hear further. For established companies that are not Google or Facebook, a natural question arises: what do we have, what will allow us to survive this transition?
In our experience, the answer is “data.” This view is held by the business press. Hundreds of articles have been written, stating that “data is a new oil,” implying that it is the fuel that will stimulate the economy of AI.
')
If so, then we can assume that your company was lucky. You collected all this data and when the AI finally appeared, it turned out that you were sitting on oil reserves. But if you’re really so lucky, you may want to ask yourself: “Are we really so lucky?”
In the analogy “data is oil” there is a grain of truth. As a fuel for an internal combustion engine, data is required for AI operation. The AI takes the raw data and turns it into something useful for decision making. Want to know the weather for tomorrow? Let's use the weather data for the previous period. Want to know yogurt sales next week? Let's use the past yogurt sales data. An AI is a data-driven prediction engine.
But does AI need your data? Today it is assumed that all data can potentially be useful for AI, but in fact it is not. Yes, data is needed for the daily operation of your forecasting machine. But most likely it is not the data that you have now. Instead, your company is accumulating data that will be used to build a forecasting machine , and not to operate it.
You now have training data. They can be used as a learning material for the algorithm. And already this algorithm is used to generate forecasts for committing actions.
That is, yes, it means that your data is valuable. But this does not mean that your business will survive the storm. Once the data is used to train the prediction machine, it depreciates and becomes useless for this kind of prediction. Continuing the analogy with oil, the data may burn. They are lost after use. Scientists know this. They spend years collecting data, but as soon as they give results, they start collecting dust on a shelf or a forgotten flash drive. Your business may be sitting in an oil well, but its reserves are limited. This does not guarantee you anything more in an AI economy than just a more profitable residual value.
No matter how valuable your data may be, the ability to benefit may be limited. How many sources of comparative data are there? If you are one of the many suppliers of yogurt, then your databases that contain information on the sale of yoghurts over the past 10 years and related data (price, temperature, sales of related products, such as ice cream) will have a lower market value than if you are the sole owner of this data. In other words, as in the case of oil, the more suppliers that have data similar to yours, the lower the value of your training data. The value of your training data is further influenced by the value obtained due to the increased accuracy of forecasts. Your training data will be more valuable if improved prediction accuracy increases yogurt sales by $ 100 million, not just 10.
Moreover, the current value of the data usually depends on the actions performed in everyday business - new data received every day that allow you to use your machine for forecasting after training. It also helps to improve it through training. 10 years of yogurt sales data - useful for teaching an AI model to predict future yogurt sales, but real predictions used for supply chain management require operational data on an ongoing basis. And this is an important point for today's companies.
An AI startup that acquires past yogurt sales data can train the AI model to predict future sales. He will not be able to use the model for decision making if he does not receive current operational data for training. Unlike startups, large corporations generate operational data every day. This is valuable. The more operations, the more data. In addition, the owner of the transaction may actually use the prediction to further improve future operations.
In the economy of AI, the value of your accumulated data is limited by the one-time benefit from learning the AI model. And the value of training data, like oil, depends on the total number - the more people own them, the less valuable they become. In contrast, the value of your current operating data is not limited to a one-time gain, but rather provides a constant benefit in the operation and subsequent improvement of the predictive machine. Therefore, despite all the talk that the data is new oil, your old accumulated data is not the main thing. However, they can lead to the point. Their value to your prospects is low, but if you can find ways to generate a new, steady stream of data that provides a functional advantage in terms of the predictive ability of your AI, it will give a stable advantage when it appears.
Ask questions, write your comments, and do not forget that tomorrow, April 10, there will be an open door , which will be held by the CEO, CleverDATA - Denis Afanasyev .
Source: https://habr.com/ru/post/447396/
All Articles