Behind the scenes
More AI, more price transparency – dispatches from Hackfest
by Martin Jungfer
Team Skyfall: How Mob Programming fills the warehouse
2.11.2018
Translation: machine translated
Pictures: Christian Walker
Where buyers used to have to replenish stocks manually, a programme called AutoBuy has recently taken over. The engineering team behind the project is Skyfall. They let us take a look at their cards.
"I think it's wrong to put myself alone in the limelight," says Oliver Streuli, Team Leader Skyfalls. Team Skyfall is one of the engineering teams that you as a reader or customer don't have anything to do with directly. That's because the team's job is done when you visit our website. This is what happened with AutoBuy, the team's latest prank.
AutoBuy makes life easier for buyers. In other words, the people who are responsible for ensuring that stock levels of your favourite products are always high have more capacity to deal with special cases and the like.
The thoughts before the first line of code
A lot of work and discussion goes into AutoBuy. Because Skyfall pushes itself. The seven coders are always trying out new things. During the AutoBuy project, they stopped working as individual developers on individual Agile stories, i.e. completing parts of the project individually and then bringing them together.
"The aim was for every member of the team to come into contact with as many stories as possible," says Oliver, "because that way more developers scrutinise the work."
.
This brings together more ideas and more knowledge. In the end, knowledge is distributed more widely across the team. This requires a slightly different organisational approach that cuts each story vertically and then breaks it down into tasks. And then there was mob programming.
The whole team sits around a single screen and only one developer, called the driver, writes code at the keyboard while the other team members discuss. The driver changes every few minutes, which means that every developer involved automatically thinks about a feature.
"We realised that we could learn a lot from each other," says Oliver. Among other things, this applies to the programming style of the individual team members. "If each developer programmes individually, they end up seeing the result of the other developers, but never the thought process that led to it."
What is AutoBuy?
AutoBuy makes the job of buyers easier. A few days ago, the backend component passed a major brand: 75 per cent of all orders are processed via AutoBuy. This gives buyers more time to concentrate on special cases and other tasks. This is because AutoBuy calculates order proposals for the buyers based on a large amount of data from all corners of the overall system. Skyfall's work distinguishes between two types of proposal:
- Normal order suggestions: Essentially an order recommendation
- Suspicious order suggestions: If the price is too high or the suggested order does not match the margin, these suggestions end up on a greylist and the buyer decides manually whether more than just the customer backorders should be purchased.
The order suggestions are calculated for a period of seven days and can be displayed accordingly in the in-house ERP. There, the purchasers, also known as planners or buyers, can manually adjust the order quantities. And they can manually exclude special cases from AutoBuy. But there is one feature that makes you happy as a customer. If you order something that digitec or Galaxus doesn't have in stock, AutoBuy simply orders the product. Previously, a planner had to do this manually if the supplier did not offer direct delivery.
Before AutoBuy, planners had a purchasing assistant as their main tool. This performed a simple linear calculation, calculating regression and outliers based on the sales of the previous eight weeks. AutoBuy is more intelligent. The data scientists at Digitec Galaxus have developed an algorithm that forecasts sales over the coming weeks. Among other things, this takes into account:
- Sales history
- Long-term trends
- Seasonal fluctuations
- Age of the product
- Information on the product category
However, this is only the beginning of the calculation. Other influencing factors then follow, such as:
- Price sensitivity
- User behaviour
- Visits on product pages and other metadata
"Simply taking these factors into account would not have been enough," says Oliver.
The Skyfall team knew that if AutoBuy was to be effective and efficient, the end users of the programme had to be involved right from the start. This is because planners consider a whole range of factors when purchasing, which should be reflected in the configuration.
How these factors interact, how they work and how they should be set, however, is something that only the planners themselves know, having previously done a lot of the work manually and in their heads.
It's human in IT
As soon as the planners sat down at a table with the members of Skyfall, the engineers realised what AutoBuy cannot do and does not see: The human factor.
"The planners have an enormous amount of knowledge about the individual suppliers. They know the people, they know how things work. This is not mapped anywhere in the system and therefore AutoBuy cannot work with it."
Skyfall is still struggling with this after the launch. Some retailers cannot yet be recognised by AutoBuy because their ordering process is more complex or unusual. However, Skyfall is not laying down its arms here: "There is still a lot of potential for the future here," writes Oliver in an internal document.
But that's not all:
- The old assistant has already intercepted and taken into account many special cases. However, this code is not documented anywhere and is not covered by automated tests.
- The data is processed asynchronously in the system. Working with the old purchasing assistant let the employees at Skyfall know that they would quickly run into performance problems with a live calculation or even cause a race condition if the purchaser made manual adjustments at the time of the calculation.
- It was difficult to split AutoBuy into agile stories, because the programme was only actually useful and meaningful once a certain functionality was in place.
- Problems were difficult to foresee. Only after a prototype was working did it become clear which problems could arise where.
Skyfall, however, was convinced by the idea and got to work. But 21,200 lines of code, countless sessions and hours later, AutoBuy was ready to go live.
AutoBuy takes over
AutoBuy made its first official production orders at the beginning of July 2018, and Skyfall and product management were still on alert in case AutoBuy made a mistake or crashed. But planners and engineers can quickly breathe a sigh of relief: AutoBuy works. Initial orders are processed smoothly, suggestions come in reliably and are checked correctly and products that have been ordered but are not in stock are ordered automatically.
Skyfall is proud.
"AutoBuy saves me about 45 minutes of work a day," says planner Chantal Stössel, responsible for Fun Sport at Galaxus.
Skyfall is proud.
Until now, she had to clear delivery backlogs twice a day. This is now done automatically. AutoBuy also prevents her from buying stock corpses, as the programme also makes recommendations for non-purchases. But she is particularly enthusiastic about the small details: "AutoBuy simply remembers more than I can remember." AutoBuy also remembers the little things, niche products and accessories that she previously only remembered afterwards, and more by chance than systematically.
In the end, you win, as a customer and as a reader. Buyers have more time to take care of other things - almost four hours more per week, after all - and the warehouse is filled more strategically, intelligently and completely.
Today, a few months after the first tests, Skyfall can proudly announce that AutoBuy is making more than two thirds of all purchases. This doesn't mean that the programme buys everything itself, but AutoBuy monitors three quarters of all items - an item can be "10 iPhones", for example - and suggests purchases to the planners.
Apropos: Team Skyfall wants to let you know that our engineering team is looking for reinforcements. If you want to get involved, please take a look at the following open positions:
- Software Developer Zurich
- Software architect with hands-on mentality Zurich
- System Engineer / Site Reliability Engineer
In addition, there are other positions in Product Development.
If you want to learn more about how we develop, then take a look here.
Journalist. Author. Hacker. A storyteller searching for boundaries, secrets and taboos – putting the world to paper. Not because I can but because I can’t not.
3 comments
later