**Making most of every dollar spent!**

When we enter any store to buy our groceries, we find ourselves surrounded by too many products being flashed at us from different locations. As if asking us to put them in our basket. And when we pick that soap from that rack, somewhere a promotional planner is happy. Because, he won the battle of correctly identifying that IF, he places this soap at

- The correct location inside the store (Gondola End)
- With appropriate discount (40% OFF)
- In the best store (Walmart)
- During the most profitable month (Feb)
- For an apt number of weeks (2 Weeks)
- Starting at (2nd Week of Feb)

Then there are bright chances, that you might purchase this soap.

The promotional planner has to answer all the above questions while planning a promotion event for every product which he is managing. This can get overwhelming if he has to create a promotional calendar for multiple brands including hundreds of products in them.

Typically, this is where promotional planning software comes into the picture, which helps the planner to chalk out a good calendar. However, most of these tools lack the forward-looking capability to predict the results of a planned promotion accurately.

This is something the promotional planner approximates using his experience and often it contains bias. As a result, it is an industry trend that eight out of ten promotions does not perform well because the Trade Spend(investment) in that promotion was either too low or too high, hence yielding poor returns.

Due to this very problem, the CPG Trade Promotions Industry has very low returns in fact as low as 16 percent! Advanced Market Research has published a paper that states, in the US alone approximately $105 billion were spent on trade promotions and the Incremental Turnover it generated was merely $16 billion.

**Profitability of CPG Trade Promotions is very low**

In simple terms, if somehow I know, how much Incremental Turnover will my promotion generate and what will be the Return on Investment right when I am planning the promotions. I can do wonders.

Let us discuss the approach on how this problem can be tackled. It can be broken down to below smaller problems and solved step by step.

Step-1: How to distribute a given amount of trade spend among retailers?

Step-2: For each retailer, what exact promotions to run?

Let us take a simple example to facilitate understanding.

We have 5 retailers and 3 months for which we need to publish a promotional calendar with a budget of $1,000,000.

Retailer | Jan | Feb | March |
---|---|---|---|

Tesco | ? | ? | ? |

Asda | ? | ? | ? |

Walmart | ? | ? | ? |

Co-op | ? | ? | ? |

Sainsbury | ? | ? | ? |

How much to spend with each retailer in every given month for Trade Promotions?

So if you look at the problem closely, all we need to do is fill 15 buckets with the optimum amount of trade spends, which will sum up to $1,000,000.

But, how will we decide whether each of those amounts is optimum or not?

The answer is the metric Spend Ratio= Incremental Turnover/Trade Spend.

We know the Trade Spend amount and we can “Predict” the value of Incremental turnover.

Using a Machine Learning predictive model which is built on at least the last 2 years of historical data for promotions, we can predict the Incremental Turnover generated by any given promotion.

The most crucial section of any data science project is the find out those critical factors which affects the outcome. Here I have illustrated some of the internal and external factors which drive the amount of Incremental Turnover generated by the promotion.

This data is used to create a Predictive Model which helps us in predicting the value of Incremental Turnover for any future promotions.

**Internal and external factors which drive the value of Incremental Turnover**

Further, we need to take into account the volume handling capabilities of a given retailer. If a retailer is small but very profitable, still we can’t assign a huge amount of Trade Spend to it, because it cannot handle such volume of investment.

This can be done by using the total turnover of a retailer for last year and distribute the Trade Spend proportionately across all retailer-month combinations.

Retailer | Jan | Feb | March |
---|---|---|---|

Tesco | $72,000 | $92,000 | $40,000 |

Asda | $94,000 | $86,000 | $77,000 |

Walmart | $77,000 | $81,000 | $73,000 |

Co-op | $88,000 | $38,000 | $75,000 |

Sainsbury | $32,000 | $53,000 | $22,000 |

**Recommended Trade Spend amounts**

*Once we know **how much** to spend and **where,** the next step is to find out **how**? this is achieved using **prescriptive analytics*

**For achieving this we follow the below steps.**

- For each retailer-month combinations, we find out all possible combinations of Trade Promotions based on historic data.
- We then simulate all the scenarios to measure the outcome of them IF the given amount of Trade Spend was invested in those ways.
- Compare the results based on any Key Performance Indicators required by business such as spend ratio or ROI etc.
- Choose that promotion which is producing maximum revenue.

**Starting from a single point budget and getting back end to end the promotional calendar**

Further, this framework is domain agnostic and can be applied to any industry where we have an investment and related outcome which needs to be optimized.

So where do we go from here? In recent times we have seen a lot of enthusiasm about Deep Learning algorithms, one gem under that umbrella is known as Reinforcement Learning, which can also be utilized to solve this problem and take the solution to next level where the algorithm learns on its own based on the previous outcomes. A huge amount of research is going on in this area currently and I hope to see matured results in coming future.