The mid-week banana curse is real. And it will get you at some point, if it hasn’t already.

You try to be good. To be healthy. To spice up the morning breakfast routine or midday snack with that delightful yellow staple. But then you realise you’ve only gone and run out of the precious fruit. Or worse – the last of the bunch has gone so black that not even your trusty NutriBullet can make it palatable.

You could run down to your local corner shop or nearest supermarket, but it’s more likely than not that the stock they have will be green.

This is because retailers, by necessity, must err on the side of caution when it comes to fresh produce.

Bananas belong to a list of fruits that are rarely ready for consumption when you need them. Mangoes? Avocados? Forget it.

Getting these fruits to your kitchen table at peak ripeness has long been – and remains to this day – a real challenge for food retailers worldwide. A recent survey has shown that food waste amounts to more than 1.5% of revenue for an average grocery retailer. Shocking, right?

Sliced Avocado

Can AI help?

It’s important to understand that despite a number of grand claims, AI is not magic. There is no HAL or Skynet, and no self-learning solution can let you simply plug in your data to solve your problems. You might see such lofty claims from some in the analytics industry but trust me, they aren’t true.The reality, however, is that carefully collected and managed data at various levels of the retail operations can be fed into machine learning (ML) algorithms to build solutions that will reduce the arbitrariness in a rule or human-based decision process.

Let’s talk about some of the aspects of retail where more recent developments in machine and deep learning can truly make a difference.

bskt is a newcomer in the retail industry, but one built on a technological foundation and with a data-driven perspective in mind. Every operation, transaction and customer interaction is recorded and analysed, giving a lot of flexibility for valuable data science work.

For instance, we are creating new types of demand forecasting models by training complex machine learning algorithms on features that were not accessible before. Internal data such as food provenance, food content, whether a food is organically made or not can now be engineered as features and can potentially have a large impact on the consumer demand forecast.

Black and Gray Laptop Computer

External variables

External data such as weather forecast, local events, or even social media for a specific food item can also be incorporated, providing better accuracy while reducing over or under stock. All of which is crucial for perishable fruits such as bananas.

Retail is by nature dynamic, and having a flexible and quickly evolving data foundation allows us to test and incorporate new factors rapidly into the forecasting models.

Another aspect of stock management is price elasticity, or how strongly a price change will affect a product’s demand. Machine learning makes it possible to incorporate price changes into the model via markdown features, for example, allowing for a better understanding of the demand under different circumstances and more precise stock management.

Panoramic Photography of Green Field

bskt operational efficiency

At the operational level also, the ability to fulfil the bskt promise to the customer – turning shopping minutes in to a few taps – means optimising every step of the process, beginning with the app purchase button.

This means using machine learning to predict the staff count needed to prepare and deliver the order depending on the time of the day and other changing parameters such as the occurrence of a live sporting event etc.

This also means optimising the route to customers’ homes based on the multiple drop off points between the fulfilment centre and the final customer to be served. It means deciding on the optimum number of orders

The final piece of the puzzle

Yellow Banana Fruit on Yellow Surface


Finally, in order to know exactly when a fruit will be ready for consumption, or the time between now and its eventual delivery, we are experimenting with the latest techniques in computer vision and training a convolutional neural network to estimate fruit and vegetables ripeness.

This kind of approach sounded like science fiction not so long ago, but it is now possible to run deep learning models on existing in-store or warehouse camera systems using cloud-based GPU instances.

I will dive deeper into the method and its implementation in a future article.

This is a non-exhaustive summary of the AI-based applications we are working on, mainly to illustrate how we are planning to say goodbye to the green and black bananas that plague our kitchens.

Data and ML are combined and broadly used in many areas across the bskt operation in order to provide customers with an exceptional experience.

Follow us for more insights into the technology that powers bskt and see how we’re changing retail for good.