Artificial Intelligence (AI) is rapidly transforming the way businesses operate and compete in today's digital age. AI technologies can be utilised to automate business processes, optimise operations, enhance customer experiences, and create new products and services. As the volume of data being generated by businesses continues to grow, AI can help organisations extract valuable insights from this data to improve decision-making and drive growth. From predictive analytics to chatbots and virtual assistants, the applications of AI in business are vast and diverse. As a result, businesses across various industries are increasingly adopting AI technologies to gain a competitive edge in the marketplace. However, as with any transformative technology, there are also potential risks and challenges that need to be carefully considered and managed. In this context, it is essential for business leaders to understand the potential benefits and limitations of AI and develop effective strategies for integrating it into their organisations.

What is Artificial Intelligence and what is Machine Learning?

Artificial Intelligence (AI) is a broad discipline that seeks to create intelligent machines capable of emulating the natural intelligence displayed by humans and animals. Machine Learning (ML) is a subset of AI that utilizes statistical techniques to enable machines to learn from data without explicit instruction. This is achieved through the process of "training" a "model" using a learning "algorithm" that progressively improves the model's performance on a specific task.

While classical programming can be effective for applying rules in business, it becomes increasingly challenging to manage a system with too many rules and answers. In such cases, AI can be a valuable tool, particularly for solving complex problems that have no known algorithm. Speech recognition is an excellent example of such a problem.

The general workflow for applying AI in business involves identifying the problem that needs to be solved, ensuring that the necessary data is available, selecting the appropriate algorithm or combination of algorithms based on the problem and data, building and fine-tuning the model, and validating the output of the model. Depending on the problem, AI algorithms may require large amounts of data to perform effectively. As such, businesses need to carefully consider the potential benefits and limitations of AI and develop effective strategies for integrating it into their operations.

AI or not AI? When should businesses use it?

Artificial Intelligence is the right way to go when:

• Business problems handle very complex logic. Search is the perfect example of an AI application. It cannot be handled with rules.
• It scales up fast. It impacts customers on a grand basis.
• Requires specialised personalisation. If you need to link a specific item with a particular customer.
• The problem adapts in real time. Financial fraud is a good example of an ever-adapting challenge.

When AI is not the right way to go

Data-related issues:

• If you don’t have data available. Or you cannot access it.
• Is data under privacy protection? Is it stored securely? Don’t use AI in these situations
• If you have a lot of data but it is not relevant. Or it is stale or unrepresentative.
• If your data is biased.

Business problem nature:

• When a problem can be handled with simple rules.
• The issue is not adapting to new data
• If it requires 100% accuracy.
• It requires full interpretability - if you are not comfortable with not knowing 100% how something happened.

What are the ways in which businesses can benefit from AI?

The magic of using AI happens when you have enough data. If you have a lot of relevant data you can make good predictions. If you can predict well, you uplift your customer experience and that generates more traffic. And more traffic means more customer data to feed back into AI models.

Common Pitfalls:

1. Data leakage is an umbrella term covering all cases where data that shouldn’t be available to a model in fact is. The most common example is when test data is included in the training set. But the leakage can be more dangerous: when the model uses features that are a proxy of the outcome variable or when test data come from a distribution which is different from the one about which the scientific claim is made.
2. Cost of productionising the model. One well-known case is the Kaggle competition where Netflix announced a prize of $1M for the best predictive AI model. However, the winner model has never been deployed into production [source]
3. Data is irrelevant / noisy / dirty / patchy
   A lot of business leaders claim that they collect a lot of data. The reality is that quite often that data is stored in a format that cannot be used in AI algorithms. Or requires so much preprocessing, that it is easier to collect it from scratch. Or it is patchy and misses important representative patterns.

Let’s do Machine Learning (ML): What is the ML lifecycle?

Developing and implementing an AI product requires three main pillars in place.
First and foremost is the team. Depending on the business problem, a range of skills might be required. In the beginning, a lot of collaboration happens with database specialists (database administrators, modelers, analysts and cloud engineers). Their main role is to provide access to the data. Once the access is sorted, research scientists and data scientists work closely with domain knowledge experts and business analysts to build meaningful features for the future model. They are the ones who also frame the data science problem and they are the best equipped for identifying relationships and assessing the statistical significance of variables. It is critical to separate the problem definition in data science and business problems. Essentially, what happens here is that data scientists transfer business metrics into quantitative problem space. They do it by framing the problem space. After the problem is defined in mathematical terms, they are selecting a range of algorithms that are the best fit for the problem. Depending on the company, once the algorithms are selected, machine learning engineers or data scientists build features. Then they split the data into train and test data sets (and validation). This is an iterative process that takes place until threshold metrics are achieved - accuracy, recall, etc., depending on the problem definition.

After the model is built, machine learning engineers deploy it into production. Quite often this might mean changing a lot of model components. The crucial difference here is that the initial model is trained on “offline data”, while deployment into production assumes it works with real-time data. In addition, monitoring tools should be in place to detect any deviations in the model’s performance.

What does a typical AI project look like?

1. Formulate the problem:
   - What is the problem to solve? - What is the measurable goal? - What do you want to predict?
2. Select and preprocess data
   - What data is available? - What data is missing? - Discarding irrelevant data - Formatting data - Cleaning data. This is what takes 70% - 80% of the time for an AI project - Sampling data. It is required in cases there are not enough observations for a model to generalise accurately.
3. Feature Engineering
   - Scaling - all the data is unified - Decomposition. For example, a DateTime variable should be parsed into a day of the week, an hour of the day, a weekday/weekend flag - Aggregation. A typical example is body weight and height measurements. These two should be combined into a ratio as a single variable, otherwise multicollinearity issues may arise when training an AI model.
4. Modelling
   - Dividing data into Training and Testing Data sets. General practice is to divide data into 70/30 proportions where 70% is for training and the remaining 30% is for testing purposes. - Training model. The model gets trained on the training data set, which is generally 70-80% of the original data. - Testing model. To understand how well the model is performing, the model gets tested on the testing dataset, which was “hidden” from the model during the training phase.
5. Productionising model
   - Deployment environment - Data Storage - Security and Privacy - Monitoring and maintenance

So what can we learn about AI in business?

Clearly, Artificial Intelligence (AI) and Machine Learning (ML) are rapidly transforming the way businesses operate and compete in today's digital age. While AI and ML offer immense potential for businesses, they are not a one-size-fits-all solution, and careful consideration needs to be given to when and how they should be applied.

Businesses can benefit from AI in various ways, such as optimising operations, improving customer experiences, and creating new products and services. However, there are also common pitfalls to avoid, such as ethical and legal considerations, data quality issues, and a lack of transparency.

To leverage the benefits of AI, businesses need to understand the Machine Learning lifecycle, including identifying the problem, collecting and preparing data, selecting the right algorithm, building and fine-tuning the model, and validating the output. A typical AI project involves various stages, including data acquisition, preprocessing, modelling, deployment, and monitoring.

In summary, AI and ML have immense potential for businesses seeking a competitive edge in the marketplace. However, it is essential to approach their adoption with a clear understanding of their potential benefits and limitations and a robust strategy for their integration into business operations. By doing so, businesses can harness the power of AI and ML to drive growth, improve efficiency, and enhance the customer experience.