How does predict delivery times and why can you trust it?

According to the report published by in December 2021, delivery time is the most crucial factor influencing customer satisfaction when ordering food with home delivery, highlighted by as many as 40% of those surveyed – more than the quality of the meal itself! The key factor, however, is not the delivery time itself but the accuracy of its estimation. Customers need to know exactly when the driver will deliver their meals. If it’s too early, they may not get their timing right. However, if the dish arrives too late, the customer’s degree of annoyance may be so high that even an excellent dish might not save the reputation of the restaurant, which might lose a loyal customer as a result.


To address this issue, has developed and introduced a system to automatically predict the delivery time of an order to the customer. Was this necessary? Let us take a look at how this whole process works in detail.

What is the method we use today to determine the arrival of the order?

John is a student who last month began working as a waiter in a small pizzeria. The restaurant is located in a quiet district of Kraków. John has his shift every Friday afternoon, during which he takes orders. He answers the phone, writes down customer preferences and all the necessary details while suggesting pizzas of the day. John is asked about the estimated delivery time very often. He then glances at the slips of paper with the orders remaining to be prepared and has to remember where the drivers have gone and calculate how long it will take them to get back to the restaurant before they can set off with the next order. The waiter gives the time he has calculated in his head and wishes the customer a good day. If John knows how the kitchen works, knows the topography of the city and has a system at his disposal which tells him the location of the drivers and the orders they still have to deliver, then he is able to quote delivery times to customers. Of course, they won’t be exact, calculated to the minute. However, this sounds like a long and complicated analysis. The waiter’s job is to sell, answer questions and take orders. It should not be the waiter’s responsibility to calculate delivery times – especially when you already know just how important punctuality is to customers.

How does Papu’s artificial intelligence determine the estimated delivery time of an order?

Let the computer do this work for you. So how do you go about it? Using your senses. John uses his hearing to know what is going on in the kitchen. He also uses his sight to spot the location of the drivers delivering orders. Based on the data collected, he begins to make conclusions to predict the delivery time. As time goes by, he gets better and better at estimating it.

Artificial intelligence (AI) uses an identical principle. The first stage is data collection. Where does the system source this data? The Papu software you see on the computer screen located in the bar and on the phone used by the driver is only a very small part of it. Papu’s databases accurately record all the necessary data, including the time, day of the week, location and how the customer placed the order. We store the location of drivers and the actual delivery times, which we then use in the process of map integration. As a result, we are able to assign more than 20 unique characteristics, including: the workload of the kitchen staff and of the drivers, the time of day and week, the journey time, driver vehicle type, town and even a specific restaurant for each order. All this data is then used to teach our AI system. 

But does artificial intelligence have any drawbacks?

Yes. There are characteristics that we do not take into account. A new driver named Martin had never delivered food; moreover, he had just moved into the city. It is clear that he will be slower in his work. However, it can also go the other way, as a new car works much better than an old one. It is important for you to know that a lot depends on your employees. If your driver logged into the Papu driver app and failed to enter the correct details or has done so inaccurately, then he proceeded to fool the AI Papu system. Incorrect information is translated into conclusions drawn by artificial intelligence. 

How to control the work of artificial intelligence

After a diverse mixture of attitudes ranging from “Wow, how cool is this AI” to “maybe it’s no use after all”, there comes a moment for reflection. Is it possible to control the work of AI? Yes, of course. Papu’s artificial intelligence is not left to fend for itself. The Papu team monitors its accuracy on an ongoing basis. This allows further automated algorithms to continually improve its performance. We call this problem “data drift” – the world is changing, and Papu is changing with it. Some elements are developing more slowly, for example, facial recognition will still work in a thousand years. Others develop significantly faster. We have restaurants undergoing a revolution within a space of one month! By automating the process of monitoring and keeping up with changes, we are able to maintain a high degree of accuracy. 

You should know this – overview 

  • Precision in estimating delivery times is a factor more important to customer satisfaction than the food itself.
  • Providing the correct delivery time is a difficult task that requires multi-faceted analyses.
  • The employee who is responsible for calculating delivery times relies primarily on intuition.
  •’s artificial intelligence, which uses more than 20 characteristics to describe an order, performs this task more accurately and faster than a human.
  • However, it is not immune to deliberately introduced falsifications.
  • has also taken care to monitor and educate the AI so that it keeps up with changes that restaurants undergo on a regular basis.

You may also be interested in



Papukurier sp. z o.o.
ul. Sowińskiego 18A, 60-283 Poznań
tel. +48 506 199 047

NIP: 7792433088
REGON: 36202821600000
KRS: 0000567424
Kapitał zakładowy: 34 850,00 zł
Sąd Rejonowy Poznań – Nowe Miasto i Wilda
VIII Wydział Gospodarczy KRS