Bullwhip Effect in AI?

Many of us already know what is bullwhip effect, but just to warmup and set the context let’s understand what and why of it.

What?

Bullwhip effect in supply-chain means increase in variability(inventory) upstream of the supply-chain. Trying to keep it simple to communicate the point of view.

Why?

Bullwhip effect occurs when we become too reactive to demand (requirements) and amplify expectations at each step of upstream of the supply chain, causing overall variability or inventory in the system to go up.

Now let’s see if we can draw its analogy in AI.

Let’s start with the most important attribute of any business supply chain, i.e. customer requirement. The importance of it remains same; be it manufacturing or be it IT or be it AI. Now let’s talk about the other attribute or stakeholder which takes this requirement and communicate it to upstream of the supply-chain, Salesperson. Now, salesperson and his/her reactions to the demand or requirements also remains the same be it Manufacturing, Retail, Banking etc. In real life those communications are never static, they are always dynamic and keeps on changing. No offence to salesperson, but it’s the reality of the world.

Let’s define the use-case

Now, let’s say we have an hypothetical requirement to build a recommendation engine for an online food ordering application. Let’s understand the other end of the supply chain (the most upstream), which is raw-material to make a required product. In AI world that raw material is “DATA”, and we AI engineers love it 😍 in front of the world. Bullwhip effect is generally related to quantity of demand, but how does the requirement of a recommendation engine be interpreted as quantity. Hmm, here is the trap.

Let’s dissect it to relate it to bullwhip effect

Actually the variability in this case, is at both ends of the supply chain. Let me explain how? First question: what will be the load on the Engine i.e., how many customers are we talking about, who will be using the APP. Here comes the input from salesperson, consider the 50% population of XYZ city (First assumption with amplified variability). What are the varieties of food items, we are talking about: Consider North Indian, South Indian, Chinese thats it :) This covers almost 80% of the varieties, adding one more amplified variability. Idea is to put across the point, so though the use-case or requirement MAY seem hypothetical, but surely resonates.

Now let’s move to the other end of the supply chain : raw material or DATA. For the above requirements, we tend to amplify the data requirement and infrastructure requirement. Security people will amplify the security requirement to cover different modes of using the engine (IOS, Android, Web ..). Remember this was not stated in the requirements. Hence the bullwhip effect in AI.

BUT, remember the app is one and recommendation engine is one but deployed on multiple devices. Unlike in case of physical products, wherein each product (though same in specifications) will be deployed or installed independently for each customer. There is no CI/CD for physical products or one common pipeline on cloud to multi-deploy…

So the question is: Are the problems similar in case of bullwhip effect?

May not be similar, but still the variability in data and infrastructure definitely increases the overhead in the overall value chain.

We will talk about the remedies in the Comments :)

“You cannot control the events or circumstances, but you can control your reaction to them”

— Anonymous