top of page
  • Writer's pictureDeandra Cutajar

For every AI action

There is a data reaction.

Design by Geoff Sence

AI brought about a lot of controversy. AI companies argue that #AIforGood will improve productivity and performance, and regulations may stifle advancement and experimentation. Conversely, ethical arguments advocate for guidelines and regulations to ensure that AI technology safeguards fundamental human rights such as privacy and protects against misuse and misinformation.

One thing that both parties agree on is that when AI is used ethically, it can benefit humankind. So why am I writing this article?

Data leaders have voiced their approach towards preparing for AI. These include Data & AI strategy, preparing the foundational platforms on which AI will run and upskilling their employees to use AI to its full potential. This article will discuss trusting an AI without human supervision or, rather, a use case for what I refer to as a post-AI strategy. For example, I will discuss how an AI-automated process can influence and impact your MARKET SHARE.

Every business has a product or service to sell, and before doing so, they do market research to understand what is referred to as PEST & SWOT analyses. In short, and it is much more complicated than I'm going to write here, the businesses need to understand the demand for such services, the environment they'd be marketing into, opportunities, and so much more. All companies have done this to understand their position in the market, expected demand, and what growth opportunities to chase.

Some companies may sell similar products and target the same customers. Due to a broad customer base, they could all thrive with almost equal shares in that market. However, a new player tapping into that customer base may need to outperform some players before owning a decent market share, let alone beat all companies and become the market leader. AI companies are doing the same thing today. There are a myriad of common problems that AI can solve. Most companies offer similar solutions to the same problems because the market is new. While some companies have an advantage and appear as the market leaders, I believe it is still volatile. Players will succeed depending on their agility and ability to respond quickly and ethically to the market demands.

Now that I have shared everything, what does "every AI action has a data reaction" do with the market?


Every AI action allowed without human oversight can shape and impact the company's market share. Whether for better or worse depends on the business. Companies risk monopolising their market share unless the AI is supervised not to take extreme actions to ensure the targets are reached.

I like examples because it puts things into perspective. So here is a company called "AB" selling seven products for the last five years. Year after year, the sales and marketing teams look at the revenue targets and business objectives and compare the products' performance. During this season, they discuss:

  • which product generates the most revenue

  • which product is selling more than the previous year

  • which campaign attracted more sales

  • what new marketing plans shall be considered to reach new customers and sales

  • how to reach the new sales targets for the new financial year

  • how to distribute the budget to ensure that the targets are met

  • and much more...

After learning about the potential of AI, the company decided that they now want to embed an AI tool to make automated decisions on these products. In their mind, the AI will be able to analyse the data quicker, make inferences faster, and analyse different ways to reach targets. Then, using some rules, the AI can provide marketing content and publish it to reach a target of some €/£/$ value. The company trusts that the AI will be able to do so by itself, and after some thought and successful tests on past data, they decided to replace the two teams with one AI tool. Why? Because in one or two iterations, the AI actions reached the targets sooner.

The first few iterations of AI will be impeccable because the AI was trained on recent past training data. But then, as it will inevitably happen, changes in people's behaviour and trends will result in data and/or concept drift. Data drift describes the occurrence where the data the AI was trained on is not representative of recent data, whilst concept drift means that the relationship between the training data and the AI output has changed. Both drifts can cause the AI to decline in performance, so monitoring models is essential to detect these changes early.

But what if AI's performance doesn't falter? What if the training data is still representative of the new data? What does that mean for the company? Is this an AI success story?

Sure, if you exclude the company's market share from the definition of success.

This is what I refer to as reaction drift. A reaction drift describes a shift in your data caused by a reaction to your actions, or in this case, AI actions.

To illustrate the point, I will use a Customer Support use case before moving back to Sales and, thus, market share.

The same company, AB, hired a data scientist to look at their customer complaints and understand which topics are the most common among these complaints.

Figure 1: A count of issues related to each Topic.

Figure 1 shows seven topics that customers complained about. Each issue was raised multiple times, with Topic 5 covering almost a quarter, ~25 %, of the overall issues, whilst Topic 2 has the least occurrence. Furthermore, Topics 5 and 6 cover almost half the issues raised.

Looking at this graph, the company decides to dive deeper into the causes of Topics 5 and 6 to understand where the problems arise and aims to reduce the %. They put a solution in place, hoping to have five issues instead of seven, because they trust that the solution removes the problem altogether. Realistically, solving an issue may give space or expose another issue, and it may not always be feasible to expect that a solution will reduce the overall customer complaints. Moreover, every new solution provides room for technical or UI/UX-related issues. In other words, addressing two topics may lead to the rise of other issues as a reaction to the solution.

An AI may as well be capable of doing what customer support did. They identified which issues comprise the majority of issues and addressed them. The issue's complexity may require different departments to come together, but an AI could create what work needs to be done and prompt each department with the task. I think that one could also argue that if AI has access, then it can fix it. I think we all agree that it is a possibility.

The above case is a common problem in customer success departments. But what happens when it is a Sales use case?

Instead of Topics, we will have Products as shown below.

Figure 2: A bar chart showing the sales driven by each product.

Yes, it is the same graph in a different context. Products 5, 6 and 7 are the most successful; almost half of the sales are driven by two products. Every quarter, the sales team looks at a chart similar to this and builds a strategy with the marketing team on how to drive more sales. I invite you, the reader, to think about potential conversations before I share mine.

The two teams will determine that Products 5 and 6 are getting a good share in the market, potentially even Product 7. The sales team will ask the marketing team to devise a campaign to drive more sales and reach the current customer base to promote the product again. It is easier to sell to an existing customer than a new one. Moreover, they may launch a "refer to a friend" discount, a very popular and, I assume, successful campaign.

Using the success of those two products, they can build a recommendation engine to cross-sell less successful products, such as Product 3 or 4, to loyal customers. The teams may even debate whether they'd keep pushing Product 2 or whether they would remove it from their range of products. However, after brainstorming and market analysis, they decided to keep it because they forecast their sales to increase in the long run. Their conclusion was based on the fact that it is better to have a product in the market when the demand arises instead of trying to catch up with competitors after the demand booms.

What if instead of sales and marketing teams, we prompt an AI to assess the sales chart quarterly, make decisions, and even run campaigns to increase sales and revenue? The AI can be given a target and asked to "make the necessary decisions to ensure that the number of sales increases by 10% and thus reach an added revenue of €/£/$". The company trusts AI so much that it believes that the AI will know what to do on its own. Do you think the AI will reach the same conclusion as the teams?

The AI may leverage Products 5 and 6 to increase sales via Products 7 and 4. It may also consider keeping Product 3, but since Products 1 and 2 comprise 10 % of the total sale, AI may ignore these altogether. (I write may because it depends on the AI prompt and the AI algorithm's development.)

If an AI is given the prompt "make the necessary decisions to ensure that the number of sales increases by 10% and thus reach an added revenue of £/€/$", then it's not a "maybe". The AI will process the information and calculate how to reach those targets with all means necessary. In its limited ability to think outside the box, the AI will focus on already successful products and attempt to upsell and cross-sell to other products from that customer base. AI will design marketing campaigns if you like. From a company perspective, once all the targets are reached and perhaps even exceeded, they can celebrate"Get the champagne. We cracked the code."

That is until the market changes, which it does, and if the company is lucky, the market demand for the Products that AI focuses on remains stable. But what happens if the forecasted trends by the sales and marketing teams come to fruition? In the case of having human oversight, they will probably cash a massive bonus because they saw the trend picking up, and patience paid off. However, the AI will fail because it has ignored low-performing products and thus monopolised the company's market share or limited it to some products. The AI didn't factor in market trends because it was not prompted to do so, and unless told, the AI will not consider the case. The company shifted from being present or even shaping market demand to responding. The latter comes as a reaction to AI actions.

AI is an enabler and a great tool to use. The Sales and Marketing team can leverage its power to identify trends in their customer base, market share, and forecasts and tweak campaigns to target different audiences. However, letting AI run automatically without human oversight on automated decisions can lead to missed market opportunities, reduced market share and limited market growth.

The solution would be to write a better prompt telling the AI to factor in market trends. I agree. And continuing from that, who would be writing and formulating better prompts to guide AI into making more sensible and "out-of-the-box" decisions? Humans.

AI and Data experts will ensure that any AI-human action will lead to a desired data reaction.


Recent Posts

See All


bottom of page