Steps You Can Take Today To Improve Your Promotional Modeling


Promotions have never been more important in the CPG industry. It is widely held knowledge that trade spend is the second largest expense on the profit & loss statement, next to cost of goods sold. Furthermore, the efficiency is not close to what it could or should be.  

In fact, an August 2017 Nielsen post stated that their latest research found 72% of promotions fail to break even.

Why is it so hard?


To understand lift gained from promotions, there must first be a true baseline to work from. The problem is that this rarely exists. Departments within a company may be using different baselines. Sales and marketing may be maintaining on based on syndicated data from a provider like Nielsen and calculating what they believe promotional lift will be with a given customer.

They give that lift figure to demand planning, who are probably maintaining a baseline off of orders or shipments, in their ERP system. The lift figure has been calculated from different starting points and that’s a big problem. 

When there is one single baseline in use, it is often poorly maintained. That is, the impact of promotions is not being separated out from the regular demand, or it is being separated out in a blunt manner. This could mean a crude outlier correction where all demand above a specified number of standard deviations is cleansed out. 

Without a sense of the true baseline demand, forecasting moving forward will involve adding promotions on top of volume that already has some degree of promotional lift from last year. Or, it can go the opposite way, and seasonality can be mistakenly cleansed out along with promotional lift. The negative repercussions that come from a lack of true baseline can then be exacerbated by other forecasting practices like the choice of hierarchy and disaggregation logic. 


In an increasingly competitive environment, where retailers don’t want competitors to match promotions, they don't like to give more notice than they have to about a promotion being run. Manufacturers are being left with shorter lead times around promotions, making it is increasingly difficult to keep up, especially if there isn't a solid sense of what lift promotions bring.  


What gets measured gets managed. Unfortunately, when it comes to promotions, often times ‘measuring’ consists of dumping information into a spreadsheet. Or the information lives in a trade promotion management tool...then there is a spreadsheet somewhere. And come to think of it, there might be some more information in another spread sheet somewhere else. Sound familiar?  

Sometimes seemingly obvious promotional history details are overlooked, or left out entirely. Promotions often aren't tracked in terms of actualization, or if they ran when they were supposed to. All of that incorrect information makes it difficult to establish a good promotional forecasting process, particularly if this period’s plan is loosely copied from last period’s.  

The net effect is that the tracking of promotions, the detail around them, and if they even ran or not, is often spotty at best. Garbage in, garbage out.  


Every Stakeholder of a given promotion has their own motivations, including how they are being incentivized and compensated. Without having an objective and quantitative means of forecasting how well a promotion will do, room is left for individuals’ motivations to creep into the calculation.  

The less objective information available to quantify a promotion, the more the process is reliant on individual participants to contribute their input. That is not to say to heuristics and experience are not important. But the more qualitative the discussion is, the more likely it's going to be biased or subjective, and driven by personal motives rather than facts about past promotional performance and current market conditions.  


Often times what will happen is that a promotion is planned at the start of the year, with the expected performance based on how a similar promotion has done in the past. The trouble with that is, retail environments are very dynamic.  

If a category is heavily promoted, with little product differentiation and minimal brand equity, then what is going on in the market overall has to be taken into account. You may have a certain type of promotion that typically yields a consistent lift, but if a competitor ran a similar promotion the week before, then your promotion may not perform as well this time around. Consumers have already stocked their shelves with what they see as a commoditized product.  

Without the market context that a competitor just ran a given type of promotion, a holiday just passed, or one of numerous other factors occurred, it is impossible to see the whole picture. If the historical performance of a promotion is the only factor considered in estimating future performance, and market factors are not taken into account, problems will arise.  


Steps You Can Start Taking Right Away


This is the minimum needed to start. A new system cannot be put in place without first taking stock of all your available data on promotions, how it's being tracked, and how good the quality is.  

Start with these three questions:
1) Are we separating out promotional demand from baseline demand?
2) What methodology are we using to track the types of promotions, when they run, and the lift that we believe is associated with them?
3) Are we making use of data from syndicated data providers like AC Nielsen? 

Moving forward, make sure that you start to close any gaps identified. 


Building off of question two above, there's value in tracking promotional metrics separately from other forecast metrics. In doing so, an understanding of the impact that they are having on forecast performance specifically, as well as supply chain performance in general, will be derived. We recommend tracking metrics such as forecast value add, bias, fill rates, and accuracy.

Separating out the two periods and then comparing provides an understanding of what proportion of forecasting issues are related to promotions and if more time needs to be spent on quantifying them. If forecast accuracy is 70% overall, but its 80% for non-promoted periods and 60% for promoted periods, then it is easy to see where improvements need to be made.  


Two bigger, More Challenging Steps 


Once the two steps above have been taken, it is time to start putting a plan in place to establish a true baseline. Only from here can proper promotion planning be built off.   Without getting deep into the math, promotional lift can be modelled similarly to how regular demand signals are modelled. If a promotion is known to have happened at time t, the difference between the regularly forecasted value that doesn't take into account promotions, and the observed value, is roughly equal to the lift. However, the difference between these two values cannot be 100% attributed to promotions. That's why it is important to smooth these values out across multiple observations of the same promotion to understand what the typical promotional lift looks like.


Next, start moving from qualitative to quantitative promotional modeling, or causal modeling. That is, moving from basing expected promotion yield on historical performance, to understanding quantitatively the attributes of promotion that actually lead to lift, and then modelling those. Leverage innovative tools that apply advanced analytics to syndicated data, such as our Promotion Effectiveness Benchmark. Developed through our Nielsen Connect Partnership, Promo Guardian™ takes advantage of our clients' Nielsen custom databases to understand and quantify the underlying drivers causing variance in promotional yield, and give recommendations on how to improve promotion effectiveness.  


What is important in modeling promotions is to objectively ascertain both a broader, and deeper, picture of lift being generated, as well as the underlying performance drivers. At a strategic, category level, this means understanding how promotional activity is driving trends, such as overall category growth. Use this information to manage product portfolios, product renovations and innovations, and longer-term trade spend plans.  

At a tactical level, it means understanding the attributes of promotions that are most likely to drive a desired lift, and quantifying them. Use this data not only to determine how well promotions will perform, but also to identify and modify the attributes that are pulling performance down.  

Ultimately, successful promotion modelling will require a balance of both tactics and strategy, but employing the appropriate analytics will eliminate the guess work.  

Author: Marcus Rogers

Comments are closed.