It is clear and well-known that the price a particular customer is willing to pay for some good depends on a great many variables. Some of these variables are known to the retailer and some are not. Examples of known variables are day of the week, upcoming holidays, upcoming local events, weather, fashion trends and so on. It is unrealistic that a human being would be capable of taking all this into account (and in the right way) to determine the best price at any time. This is where you need a formula (price = function of all these variables) to tell you what price to quote. The best way to obtain this formula is to collect empirical data from customers who accept and reject prices under various scenarios and then to learn this formula from the data. That is what machine learning is for. This approach saves you a great deal of time and gets you a very good price at any time. That is the aid that is provided and it is substantial. We have done this in the airline domain and achieved a revenue boost of 7% without doing anything other than learn this formula and so this is pure profit.
The data used to make this model does not need to be "big." In any case, the terms big and small are too imprecise to really mean anything. You need the right amount of data in your case so that enough statistically relevant and significant data is available. In many situations, a design of experiment scenario planning will allow you to get away with a very small dataset and come up with an accurate model. I advise looking at each case through the lens of "what data do we need" instead of "what data do we have" and it usually turns out that you need less than you have.

I would like to echo Patrick's commet but add additional insights I have observed. AI pricing technologies are leveraged to provide targeted data on customer expectations and behaviors. Not only has it become criticially important to identify factors that encourage and discourage purchasing behaviors and qualify them. The human brain cannot decide simply based on qualitative interviews. Instead, it has become imperative to develop algorithms to deeply understand customer behavior.

Sophisticated pricing was done using multivariable regression in the past. AI machine learning algorithms may add some marginal value over traditional regression, but only if the data sets are good. This is far more of a data set problem, or a data science problem than it is an AI problem.
You can only use deep learning if you have a very big and dense data set. Deep learning is very good for clean, dense data sets, but it is not so good at outliers.
There are other ML method. Clustering, probabilistic reasoning, even regression in a sretch is considered AI. There is no need to invent new AI algorithms here. You can just use a free one from OpenAI. The value is in the data engineering.
However, the buyer should be concerned about propoer pricing and not really worry about whether its AI. If AI gives a better answer, use it and if it doesn't then don't. The product is the determination of pricing, not using AI algorithms. We get lost in this sometimes. Keep your eyes on the prize.

I agree with most points made above. However, I would disagree with Ed on the AI part.
I do think that AI methods are going to be key for pricing strategies. Yes, all ML methods should, are and will be used with the data sets availble to the store. However, AI can further this by contextualizing the data, finding new relationships between nodes and even help determine, what other data might be useful in order to increase precision or clusterization. AI can help us go beyond just the data and relate the concepts behind the data, reducing the reliance on rather static model building and offer more flexible exploration.

First, AI is a huge area. Different methods have different challenges. Multi-variate regression is one form of AI, by the way. The basic challenge is getting enough good quality data. The secondary challenge is getting a person who really knows the algorithms well. Get the domain expert and the data scientist to talk to each other. If you have these three, you're sorted.
The key data source is the customer himself/herself. You need to get data regarding the price they are willing to pay for what. That means that you need to vary the price proactively so that you can gather this data. Ideally you want the customers to identify themselves. Knowing their address, you can buy databases that will tell you their educational and income levels and so on (statistically speaking, of course).
You also want to gather data regarding weather, major events, holidays. Then some data around the product such as its age or any relevant competing product's features.

Thank you All ! You highlighted few excellent points.
Can you touch on any challenges and practical limitations on AI-based pricing engines?
Specifically, I am interested to know the key data sources (not types of data) for AI-based price engines? Which ecosystem partners aids in more precision and accuracy of AI-engines?

AI behaves on the data-set you feed in. To create a pricing engine you need various factors -

1. Product Attribute
2. Market Situation
3. Competitive Advantage
4. Customers Pattern and attribute
5. Profit Margin

And this data-set should change from country to country. Such data sets are complicated. This is one key challenge in my opinion.