Mjam Assistant —a customized selection of your favorite restaurants in food delivery app
All existing food delivery platforms offer some kind of restaurant discovery, customized to customer's preferences: it can be about a specific time of the day, for example, lunchtime so users can browse through affordable lunch offers from restaurants nearby. On other occasions, they can choose from a list of their recent orders, a favorite pizza from well know pizzeria.
Rarely apps can use complex algorithms to “remember” user’s preferences and display selections tailored to their taste and previous search.
When thinking of a typical situation when a person or a group choose to order some food online, the usual ordering scenario would probably be somewhat like this:
A: Hey I am hungry.
B: Me too.
A: What do you feel like?
B: Pasta, you?
A: Pasta is ok, maybe pizza too, but please something fast.
B: Ok, but I don’t want anything shitty.
A: Well, how much money do we have?
B: It is ok, I have PayPal, and we can treat ourselves.
A: Ok. I can pick something on my way to work.
B: No problem, I will order …. Yay, our food is at the door in 20!
But … it’s not always that easy as it looks…
The challenge & discovery
In many usability tests, we conducted with users of the Austrian food delivery platform, Mjam (in Vienna, 2017–2019) and from quantitative research, we observed repetitive usability issues and some troubling behaviors, when it came to the restaurant discovery phase of food ordering process, especially with new users.
Let’s have a look at some repetitive insights from the data we gathered:
- Users spend too much time ordering their food. Browsing the restaurant list, going back and forth to the menu page, took them in average 10 min
- They often complained about seeing too many restaurants not relevant to them.
- On average 32% of app users who convert see 9 restaurants before making a decision
- Only 8,22% of app users use filters to narrow down the selection, but from those who use them, 89% select the restaurant from the results.
- Only 12% of app users use sorting, but from those who use it 89% select the restaurant from the results
- 52% trust the restaurants offered when filtered by super cuisine
The conclusion was when users choose to order the food they often find it very hard to find and choose the restaurant they like. They spend a lot of time browsing through the list and checking the menu offers on the menu page, repeating the process multiple times.
But from a few of those who choose to use the functionalities such as filters and sorting, the majority at the end selected the restaurant to order from, from the suggested list.
There were 3 pain points users had within the discovery process we could identify:
- There are too many restaurants on the list, most of them not tailored to customers preferences
- It takes too much time to find the right restaurant on the list
- There is no helpful way to discover restaurants similar to what they like
After the conclusion we believed we had a solution in our UX sleeve we wanted to test:
- If we present filters and available cuisines in a more prominent and more beneficial manner, more of the new users will find food they like easier and faster
- If customers are able to save their preferences so they can easily find restaurants matching their preferences, retention & reorder rate could increase
- If a limited number of restaurants based on their preferences is presented to the users, time to order will decrease and CVR will increase
- The number of restaurants compared before ordering will decrease when presenting a limited number of restaurants based on user’s preferences
- A number of new restaurants will be discovered in a curated list of restaurant, based on given preferences
- USP → we will be able to convert more new users with the new flow because the food discovery is faster and better than with the competition.
Ideation and validation 0.1 — Chatbot
As a result of the first ideation, we adopted a chatbot pattern to create a friendly and interactive solution to get valuable information about user’s preferences when picking the right restaurant. The collected information would be used to generate a curated list of restaurants at the end of the flow.
The solution was tested on a group of Mjam new and recurring users in a usability test in Vienna.
The main goal of the test was to observe reactions to a limited set of results as a result of narrowing the selection by applying the filters through a chat flow.
Although all of the users really liked the idea and friendly chatbot feature, the length of the onboarding process was an issue to all of them.
We immediately responded and presented an iteration of the concept to the second group of users the next day. A friendly but lengthy chatbot concept was replaced with a so-called vending machine concept: all of the preferences were accessible on one screen without any additional steps needed.
Users' reactions to the updated version were positive. They felt the interface was intuitive, especially due to the simple interface and included visual elements.
Although the reaction to the onboarding solution was overall positive, most of the users’ were less enthusiastic towards the preferences selection they had to choose from. Some of them felt there was not enough variety, the others just felt they are not relevant to them.
User research 0.2 — Food preferences
As a next step taken, we needed to define what types of food preferences bring users the most value.
In the current app filters set was limited. They could filter by type of cuisine, delivery cost, minimum order value — not much opportunity for curating the selection of restaurants.
It was important to learn what are the most important choices users think about when they are ordering food.
We understood about the most common, high-level food preferences when choosing the right food were:
Cuisine & Foodtype (& Dish)
Estimated delivery time
Since many users expressed they’d like to see more variety in that list, we added a few more such as food type, menu category, ingredients, taste, and preparation style and conducted a card sorting session.
From the existing data we had, we could easily name the most popular cuisines and top-selling dishes(food type).
The rest preference groups were combined from the available restaurant’s menu content and competitor’s menus analysis.
The goal was to learn what comes first to user’s minds when they start the discovery process when ordering food.
The results would help us build a more accurate step by step curating process and better selection results.
The results showed most of the users would take this path below to narrow down their restaurant selection:
1 — Cuisine
2 — Food Type
3— Menu category
4 — Time / Budget
5 — Menu category
6 — Ingredients
7 — Taste
8 — Preparation style
We had enough to proceed to the next step — a decision tree.
Ideation — Bubble tree
I believe you all might remember Apple’s Music onboarding. Do you remember that red and black bubble interface which allowed you to enter your music preferences and save them?
As you were selecting your favorite genre and producers, you’d see more and more detailed suggestions. The suggestions got more and more specific towards the end so that the final results were a very precise list of suggested artists and albums, tailored to your taste.
Yes, you’re right … That was another existing experience we wanted to mimic and test with our users.
As a first step, we created a decision tree, based on the results from the card sorting session, as a foundation of the user flow.
The user would need to choose the type of cuisine first, which will trigger and display matching suggestions for the dietary choice as a next step, which would then reveal matching occasions and taste suggestions, followed and finished by the meal types…. We named the new feature Mjam Assistant. ❤
I know it sounds a bit complicated. We were worried too. :)
Besides the complexity, to develop a fully working feature, another essential component was needed — a proper dish tagging (eg. the dish on the particular menu is spicy and vegan), which, unfortunately, didn’t exist at the time.
Nevertheless, unfavorable constraints didn’t prevent us from proceeding with the idea and planning MVP for the Assistant.
As we wanted to test the feature as soon as possible with less possible effort, we decided we will build Assitant with a simplified version of the preference list.
We would use only existing filters and measure how many users will save them and return them to the saved list next time they open the app.
User flow and UI design
At first Assistant MVP was created only for new users, introduced as an onboarding feature.
When users first open the app and enter their delivery address, they would trigger the Assistant. Within the next 3 simple steps, they would be asked to select their preferred cuisines and narrow down the selection by applying few more filters in the next step.
All selected preferences are then saved in a new section of the app called Meins (Mine). As a result of saving their preferences, they can now see the top 10 matching restaurants on the list. They can always refine their saved preferences by easy access to the Assistant feature from that page.
Super excited by the execution of the experiment, we could start to collect the eagerly awaited data.
Unfortunately, the first GA numbers were not as encouraging as anticipated.
Although the experiment showed positive results across the whole funnel (users visiting a menu page improved by 2.83%, Users adding something to their cart improved by 3.5%, users placing an order improved by 3.6%), ~70% of customers didn’t complete the assistant first step.
There was a bit more encouraging data from analyzing those users who completed the Assistant flow. They seem to be more committed to placing an order and:
- 93% visit a menu (vs 70% on average)
- 86% add the item to the cart (vs 60% on average)
- 83% go to checkout (vs 75% on average)
There seemed to be an improvement in engagement as expected:
- 95% of users who used the assistant come back to it at least once again.
- 60% of those customers place an order (vs 35% on normal listing page)
Since the results were more positive for the users who actually used the feature we decided to take few more steps towards improving the Assistant experience:
- Run user interviews to get more qualitative input
- Promote Assistant to increase the discovery
- Once the food labeling project is established; update the assistant to include extra information: allergies, spicy, vegetarian, vegan…etc.
Final step — Usability test on the live feature
As one of the most concerning insights from the qualitative data was ~70% of customers didn’t complete the assistant first step, we wanted to understand why?
In the following remote usability test with Mjam users we were focused on few main topics below:
- Why most of the users skip the Assistant onboarding?
- How they perceive saved filters? Do they notice this is related to the choice they made in Assistant?
- Are the results of onboarding expected and relevant?
- Meins: why more users don’t interact with the feature in Meins?
- How can we increase engagement?
- Where would they look for Assistant in the app?
- Why more users don’t save their preferences?
The top insight we gathered from the interviews:
- All users are aware of Assistant but no one is using it actively
- They are not using Assistant because they don’t see a real benefit from it since it’s using the same filters as the ones available on the Restaurant list
- They would consider using assistant if filter selection would be based on more detailed/refined preferences such as dish level selection, dietary preferences, recommendations/popularity, new/upcoming, etc.
- No bigger issues with UI design
Mjam Assistant as an MVP gave us a lot of insight and direction for the future development of the feature. It was clear that the feature would have a great value to all types of users if we provide a more refined set of preferences they could select from. The feature would be beneficial for the new users to help them discover restaurants they like faster, as well as for returning users to always return to their saved list of preferred types of restaurants.
There are definitely new next steps in the plan, such as exploring alternative solutions for filter selection in order to show more relevant results and more.
For that, as always … thank you for reading so far, and stay tuned. :)
I couldn’t achieve anything of the above without my dearest and the best Mjam Product team members, so thank you: Hesham, Antonia, Ola, Christian, and of course the amazing Mjam Tech team.