By Arvind Ganesan and William Dowling
For a consumer brand that is sold offline, the number of doors it is sold in matters a lot. The more doors, the more exposure a brand has and the greater opportunity it has to sell its products to customers. If a brand is gaining doors, it’s generally a good sign that the brand is resonating with consumers, and that it is growing revenue. In fact, we have found that growth in distribution has the largest correlation to revenue growth for a brand – more than velocity, for example. Conversely, if a brand is losing doors, it’s generally a sign that the brand might be cooling down. Investors and retailers know this and will often look at the number of doors a brand is sold in and how that has changed over time when deciding whether to make an investment or stock a product. While looking at current and past door growth is important, it doesn’t paint a complete picture. What investors and retailers really want to know is how many stores it will be sold at in the future.
Historically, this has been very challenging to predict even with access to all the right data. If a new yogurt brand was sold in 100 stores a year ago and is sold in 200 stores today, how many stores will it be sold in a year from now? 300? 400? More? Less? Exacerbating this problem is the fact that different products grow in different ways (e.g. a yogurt will have a different growth pattern than an energy drink) and products grow faster in some retailers than others (e.g. a dozen doors in a major retailer may ultimately be more impactful than a hundred doors in a regional retailer).
This is a challenge we’ve spent a lot of time thinking about at CircleUp. Being able to accurately predict door growth could help overcome some of the uncertainty and heuristics that typically accompany early stage consumer investments. That’s why we’re thrilled to announce the launch of a new distribution model in our machine learning platform Helio. Simply put, our Future Distribution Model predicts door growth for brands. This isn’t our first distribution model- in the past we’ve discussed our model that scores brands based on their historical distribution and we’ve talked about our Consumer Reach Index which measures the offline and online reach of a brand. But this is our first distribution model that looks into the future, and we are very excited to share some early outputs with the world.
How the model works
The model answers the question, given all that we know about a product’s current distribution, what do we expect as the likely range of outcomes for that product’s distribution in a year? To do this, Helio ingests all the distribution data we have on a product across time. Among other things, it looks at the retailers a product is sold at, the growth or lack thereof within that retailer and across retailers, the quality and access to distribution that the retailer can provide to a brand, and the importance of that retailer to the category of the product. Based on all this historical information, the model predicts how much distribution growth we would expect a product to have in the future. The model can predict the growth of one product, or all products that belong to a particular brand. It’s already helped us surface some interesting insights as they relate to product growth.
For example, one interesting trend we found while examining historical data is that close to 60% of products across all categories exhibited negative to no growth, meaning only around 40% of products exhibited any positive growth (we are looking on a product level here but can roll up to brand). This points to just how competitive the consumer market can be with the introduction of new products. There is a relatively fixed amount of shelf space in physical retailers, creating a ‘zero-sum’ game situation for brands looking to grow. Being able to identify what products will grow is critical. The below chart tracks the actual yearly growth (from Q3 2016 to Q3 2017) of 7 different product categories, with the X axis representing the percent of companies that fit in that growth bucket. We are showcasing 7 categories here, but we have the ability to run the same analysis on every category in consumer.
As can be seen, growth patterns differ across categories, although the biggest differences are in the percentiles within the same category. For example, yogurts in the 0.1 percentile for distribution, lost 65% of doors over the past year while those in the top percentile gained 61% in doors. Sports and energy drinks in the bottom decile lost 92% of doors while drinks in the top decile gained 115% in doors-or more than doubled their footprints. These massive differences in growth underscore how important it is to correctly pick the right brands and products within a given category.
After examining past performance of products, we ran our new model on the same products to predict growth a year from now. The results are below.
Note that the growth scale is different for this chart, and that predicted growth tops out at 80% in the prediction vs 200% for the actual. This is not because we think that all brands will have slower growth in the future but rather due to the nature of our model. Without going into too much detail, one reason for this is that we predict door growth for the products that exist today- not products a brand may or may not add 3 or 4 months from now. In 2017, 13% of SKUs were new SKUs. With that in mind, for currently existing products today, our model predicts that yogurts in the bottom decile will lose 28% of doors over the course of a year while those in the top decile will gain 12% in doors and sports and energy drinks in the bottom decile will lose 66% of doors while those in the top decile will gain 44% in doors.
The great thing about this model is that we’re no longer left guessing which brands will grow and which will shrink. Not only are we able to identify that the top sports drinks will grow by 44%, we’re able to identify exactly what those top sports drinks are.
Why the model matters
This model helps cut through the fog that has previously surrounded product and brand growth. Like all of the models in Helio, we don’t think any one model is the magic bullet. This model specifically predicts how future distribution will vary based on existing distribution. Other factors — such as product, brand, team, etc. — all matter as well, and it is this mosaic of data constructed from all our models that gives an information advantage. We believe this can help a variety of different stakeholders.
Our mission is to help entrepreneurs thrive by giving them the capital and resources they need. This model gives us powerful new data that we will be able to share with the entrepreneurs to help them grow. Not only are we now able to tell these entrepreneurs which of their products are predicted to grow the most, but we can also tell them which retailers will contribute most to their growth. This will allow them to target certain retailers over others and use our predictions to take their growth into their own hands.
We’ve touched a bit on the benefits here, but with the new model we can look at the predicted door growth of companies and the predicted door growth of every product that company owns. Investors can use this data to make targeted investments in the brands with growth patterns that meet their hypotheses. This is often, but not always, the highest growth companies. Some investors actually like investing into companies where they aren’t expected to expand distribution – with the hypothesis they can add value by fueling that distribution themselves. This model allows investors to pick companies with a predicted growth that works for them.
Buyers at retailers have hundreds, and sometimes thousands, of brands vying for their attention and trying to capture shelf space in their stores. Often times, it can be hard to tell the difference between these products and difficult to determine why one might grow more than another. They typically have very little data, and none of it is predictive. The new model would allow retailers to immediately determine which brands and products have the highest growth potential and select those products for their shelves.
As often happens with Helio, we also expect to see these insights applied in ways that we might not anticipate yet.
A view of the future
In December of last year, our CEO Ryan wrote a post about where Helio is today and discussed some of our platform’s limitations. He made mention of the Gartner framework for analysis, included below.
At that time of the post, Ryan noted that Helio was very good at descriptive and diagnostic analytics, but was not as strong at predictive and prescriptive analytics. While we still have room to grow, we think the new Future Distribution Model is a big step towards being able to describe what will happen and effect how it will happen at scale.
We look forward to sharing more outputs of this model with you in the near future and continuing to build Helio to predict and prescribe other factors that are important to entrepreneurs, investors, and retailers.