The Curious Case of FMCG Sales Beats

Indian Retail is a Buyer’s game today. In E-Commerce as well as offline modern trade, consumers are truly the king, with every tick of the clock bringing new choices. Hence it has become both, vital and challenging for enterprises to be close to the end consumer. An efficient distribution system, which reduces order to delivery & replenishment time becomes the key arrow in the arsenal of an FMCG company. The key element to this is “Sales Beat Planning”.

“Sales Beat Plan” also called “Permanent Journey Plan” is a day level route plan made for field sales & marketing executives to visit several stores at a pre-defined frequency. These visits are necessary not only to handle order collection but also for visual merchandising and most importantly competitor analysis. Stock outages are periodically reported from retail stores as well as distributors to continuously replenish the orders. Stock displays at all stores are regularly audited and changed to gain a competitive advantage.

When one of the world’s largest FMCG enterprise approached us to optimise their sales beat planning with high benchmarked figures in mind, We at Locus decided to solve this for them. Without much ado, we jumped right in!

The goal of optimizing the sales beats was to have the right service levels at each retail outlet. The intent was to ensure that every outlet is serviced by the right person on the right day at the right time. While doing so, it was also intended to come up with the most ideal beat size (the most optimal number of outlets within a beat) keeping the best possible mix of outlet types. Additionally, what also needs to be suggested, was the sequence in which, the outlets in every beat would be visited. This indicated the need of a cluster with minimal back and forth, travel time and distance.

The problem is indeed as complex as it is beautiful. Just to add to the complexity, any implementable solution envisioned for an NP-hard problem like this, requires beats to be intelligently designed. The design algorithm should be open to accommodate dynamic and fuzzy constraints that may occur on ground.

With gigabytes of sales data, location trails and timestamps at our disposal and the best data scientists to derive insights out of it, we started to mimic the existing modus operandi to find potential areas to change. And thus, the first thing that we started with was to cleanse few hundred thousand bad addresses. The addresses were so malformed that they couldn’t be sensibly tagged to a set of latitude and longitude. Now, this is a problem that exists in most developing economies. Lack of infrastructure around capturing addresses results in poor data collection.

Courtesy Locus’ proprietary geocoding algorithms, we were good to go after a few iterations through efficient Tokenising, Machine Learning and Natural Language Processing of these addresses.

Inferences:

We moved on to replicate the client’s existing beat plans and ended up drawing several ground breaking conclusions.

1. Under-utilised salesmen/marketing executives — They could be much more productive within the same working hours
2. Long and cumbersome Beats — A Salesman walks for a significant time in his beat hence its unfair to expect high productivity at all times
3. Overlap in salesmen routes — With overlapped routes, salesmen were walking more than they had to. The time spent in transacting at every outlet was taking a hit resulting in potential revenue loss.
4. In parallel, we also analysed behavioural factors that impact B2B sales.

Following were the factors that we narrowed on:

1. Experience of the salesman in selling the product.
2. The tendency of the store to re-order increases due to familiarity with the salesman

This was a breakthrough. Putting the pieces together, we had substantial data to conclusively derive the most optimal transaction time at each outlet. This was a function that we were trying to maximize per salesman. Beats were to be designed such that majority of the time is spent servicing the outlet rather than travelling. Our routing engine smartly handles constraints like geographical proximity, variable walking speeds, traffic conditions, outlet availability etc. and ensures that the same outlet is not visited by any two salesmen on the same day (known as “mirror beats” in the FMCG world). We also ensured that the volume sold on a given day is within the range that can be delivered during the distribution cycle and is fair across the week

Here’s a snapshot of Existing vs Locus generated beats for a sample data set

Source : Locus

Results and Insights

Increase in Serviceability Ratio by 10% — The portion of the sales beat spent in transacting at a store was increased to 70% from 60%

Reduction in beat length by 20% — The Salesmen now spend a higher fraction of time in transacting at a store than walking

Reduction in total number of beats by 8% — More outlets covered per salesmen per day thereby reducing operational costs

With no easy algorithmic tool/product to map a beat to a salesman, we jumped at the opportunity to build one.

A few brainstorming sessions and failed attempts later, we had a tool to take the salesman’s historical performance, experience and familiarity with outlets and associate him with the best beat. Basis time range of the input data, the tool tweaks the weights of each input parameter.

Source : Locus

In logistics, a successful on-ground deployment is the crucible. The beats have been successfully deployed and the salesman performance is being tracked and monitored for over a quarter.

Locus’ salesman-beat mapping has proven to be about 30 % more productive.

So, next time you think Sales Beats, think Locus :)