Recently, the R&D Team at Everli embraced a super challenging initiative that we call Firebreaks! 🔥🔥🔥🚒
This served as a natural breakpoint as teams wind down their old missions and prepare to start their new ones. It’s an excellent opportunity to pursue other work that’s of interest to them and of value to the organization.
Our colleagues ventured into several workshops and tested themselves in a five days sprint!
After introducing a list of ideas and related projects, they worked on the most voted ones, describing the final output.
Here’s the list of the initiative winners 🏆, followed by details on each of them:
- The Most Ready To Use ⚡ / Quick search on product lists
- The Most Fun 😂 / Alexa skill to fill my cart
- The Most Wow 😲 / Carbon footprint calculator
- The Most Useful/Wanted 🚓 / A/B testing dashboard
The Most Ready To Use ⚡
Quick search on product lists
Why?
It’s all about the context!
When a customer searches a product, he might be confident on the category or selection he is looking into.
Let’s make it easy for the customer to find the desired product and save valuable time ⏰
How?
We introduced a new filter option in our Website sidebar (soon to come on mobile too). This is a keyword filter within the current list of products.
It can be either a category listing products, selection of personal Favorites, Offers or even Cashbacks!
Practical example 🍅
Imagine you’d like to buy some tomatoes, so you would naturally search “pomodoro” across the site…
While the results are relevant to your search at a technical-level, you know that you’re looking for the fresh veggy tomatoes, so you’d go to Fresh Vegetables categories.
There you now have the “Search in this page” field ⚡! Try it…
Even more! Another practical example 🍝
Maybe you’d like to get some pasta …
Good news! You can filter by “gluten free”
Contributors:
@Mihai Tuhari, @Matteo Piran, @Cesare Zotti, @Claudine Rollandin, @Federica Felici
The Most Fun 😂
Alexa skill to fill my cart
Why?
Let’s spread Everli outside the apps and website world!
With the aim to get “peace of mind” to our customer let’s try to add “Everli experience” in other customer life habits!
So how could we help our customers even more? 🤔
Why not add Everli to voice assistants, such as Alexa?!
Brainstorming
So at the very beginning of Firebreaks, it was presented thea idea to “fill your cart cart through Alexa” and the newborn “let’s break Alexa” team started working on it!
But, first fun fact and challenge: we noticed out a clumsy human interaction.
- Human: “Hi Everli! Add tomatoes to my cart”
- Alexa: “Here’s what I found: Cirio tomato sauce, Mutti tomato sauce …” and so on with many other products
- Human: “Add the Mutti tomato sauce” and then
- Alexa: “I don’t understand, please repeat.”
and repeat and repeat and repeat for every product you want to put into your cart!
We noticed that getting the Alexa job done will take longer than the Firebreaks week.
With the aim and spirit of “do something useful”, we changed the direction and did another brainstorming on what we can offer in a timely manner.
After a half day of discussion we put our money on two main features:
- Status of your order
- Offers in store store
Let’s implement it!
First problem: How do we get user data?
We need to link our user account to Alexa systems, so we go through documentation and we discovered how to use our actual login app to add a simple “link account page” to our system and send our token to Alexa system request.
As you can see, the squad didn’t had front-enders 😂
Second problem: How we can design the dialog flow?
Nothing easier for our superheroes Cesare & Riccardo using Flow-Chart!
Showcase:
The funniest part cannot be said in words, let’s showcase:
Where is my grocery? (order paid)
In this particular example we asked Alexa via the command “Ciao Everli” (Hi Everli), where is my order and it answered the following:
- Our order is ready!
- Riccardo (the shopper that has accepted doing the order) is coming soon with your grocery.
- The final amount is 74,60 €.
- If you selected contactless delivery you do not need to open the door before the shopper has finished arranging the bags.
Contributors:
@Adrián Beniel, @Riccardo Peddio, @Mirko Ventura, @Cesare Zotti,
The Most Wow 😲
Carbon footprint calculator
Why?
To do our part to help saving our doomed planet. 🌍
The carbon footprint calculator aims to:
- Encourage the customers to buy less plastic: each shopping will get them an amount of points which can eventually be converted to discount codes and more.
- Give customers an easy way to contribute to the cause: at the checkout, they can decide to donate (“round up”, 2%, 3%).
The collected money will then be invested to organizations in the environmental fields: local to the cities we serve or to Treedom, to grow a forest.
Similar to our initiative for Earth day.
What?
It is not a small project: to have a final product in 5 days, we had to reduce the scope a bit.
The whole project is split in 3 macro-areas:
- Data mining and machine learning models: we built these algorithms to build up our product-footprint data
- APIs: clearly, the data we stored needs to be served. Also, we want to store the accrued points per each order and user
- Frontend/UI: the final piece, to connect the APIs to our website and apps
So, we focused more on the first 2 steps above, to have solid APIs and reliable models and attach them to our service at a later stage.
How?
Data mining
We figured that we already have much of the data we needed in our DBs, at least to get started with.
We used our products’ details information to evaluate a cart:
- Non-packaged items, family size packs, bio items, KM0 products, sustainable production processes will increase the points of the cart
- Little plastic bottles, mono-dose products, etc. will decrease the total score of the cart
For example, to calculate the amount of plastic, we look at the gross weight and diff it with the net weight of packaged products.
Through some heuristics and quick analysis on the average existing orders’ footprint, we picked sensible values for range for the points, eg. sustainable products will not get you +100 points, but +5.
All of the computation was done through Jupyter notebooks. The result was then stored on AWS DynamoDB.
The plan for the future is to move this mining/computation part on AWS Sagemaker.
APIs
We decided to go with a cloud solution for this project:
- AWS API Gateway as the entry point of our interface
- AWS Lambdas (python) to read and write data to DynamoDB
I know, it looks super easy – and it was. We profited of AWS to quickly draft APIs and have them deployed in a matter of minutes. (…then we spent few hours making them actually work…)
Putting it together
Here is a chart summarizing the flow:
Contributors:
@Giulia Tumminelli, @Jessica Lois, @Mattia Uttini, @Marco Pernigotti
The Most Useful/Wanted 🚓
A/B testing dashboard
Aim of the project
This project aimed at providing specific answers to frequently asked questions, like:
- How much time should my AB test run to achieve significance?
- What is the optimal split traffic we should adopt to detect any tangible effect?
- How is my currently running test doing?
In the past these answers were provided exclusively by Data Scientists and Product Analysts to the product management.
They performed such estimations manually. As the volume of planned and running tests is constantly
increasing, a more scalable solution was definitely required.
Enter the Awesome AB testing dashboard!
This fast, fancy and user-friendly solution will allow PMs to easily run scenarios for future AB tests as well as check the status of those that are running with zero effort.
The awesome AB testing dashboard is a simple web app, comprising two different sections.
AB test scenario planning
The first one allows users to forecast testing scenarios, by selecting
- The type of metric to be probed
- How much uplift we would expect
- What are the probabilities of doing a 1st and 2nd type error
Scenarios are produced in real time by querying our main DB and surfaced via a nice plot.
Test visualization and assessment
The second one can be used to monitor currently active tests, as well as inspect those that are terminated.
Users have to simply select the name of the test, the metric they want to inspect et voilà.
Underlying techonology
The technology powering our solution is streamlit (an easy-to-use python library to build and share data web apps); the computational backbone is a collection of python modules that performs standard significance tests.
Interactive visualization of business metrics has been enabled via plotly.
Currently, the app is running on local devices, however plans have already been made to deploy it on our Everli servers 🚀
Contributors:
@AlessiaD, @andrea-pio-d-antonio, @AndreaPaglia, @Fracappo87, @giuseppellrusso, @mihaela-izabela-floreapeddio, @rpeddio
Congrats to everyone! 🎩