Project background
01
How could in-app recommendations feel more relevant to everyday life?
Bolt Food is a habit-driven app: people use it to order lunch, re-order favorites, restock groceries, or get dinner after work. These actions begin from internal triggers (hunger) or external triggers (notifications, links, app icon, social posts). Business goals are repeat usage and improved experience, but food ordering is a convenience, not a necessity. That is why the app must become more relevant, trusted, and emotionally engaging over time.
Our brief was to shift Bolt Food from a passive list of options into a context-aware companion that adapts to users and knows when to simplify or when to suggest something new.
To understand how habits form, I audited the app using Nir Eyal’s Hooked Model (trigger → action → variable reward → investment) and mapped the experience across awareness, choosing, waiting, and receiving. Research methods included a service safari, interviews, journey mapping, prototype testing, and a co-creation workshop with 10 participants (3 groups). The goal was to see how people pick food, compare choices, respond to context, and how Bolt can learn from repeated use without forcing lengthy onboarding.
Our brief was to shift Bolt Food from a passive list of options into a context-aware companion that adapts to users and knows when to simplify or when to suggest something new.
To understand how habits form, I audited the app using Nir Eyal’s Hooked Model (trigger → action → variable reward → investment) and mapped the experience across awareness, choosing, waiting, and receiving. Research methods included a service safari, interviews, journey mapping, prototype testing, and a co-creation workshop with 10 participants (3 groups). The goal was to see how people pick food, compare choices, respond to context, and how Bolt can learn from repeated use without forcing lengthy onboarding.
personas
02
Personas given by the client
research background
03
Background to user clustering concept
Hooked Model Analysis
In the current solution, triggers exist, but often miss the moment. We noticed a travel notification at 9 A.M. while the user is at home heading to work. This is a reminder without much relevance.
Actions in Bolt Food are mainly browsing restaurants, exploring food, adding to cart, placing orders, and reordering. But food ordering is not always simple. It involves mood, hunger, time, price, dietary rules, language, trust, and sometimes social pressure.
Given that Bolt is mainly providing mobility services, we noticed an opportunity for inter-service actions. Bolt has food, groceries, rides, scooters, and more, but these habits are not yet connected into one ecosystem relationship.
Current rewards are useful, but predictable: discounts on restaurants or delivery, subscription offers, cashback, and fixed percentage deals. These reduce pain, but they do not create much surprise or emotion. We wanted rewards to feel more like winning, not just saving: real, rare, contextual, and earned.
Investment already exists through order data, ratings, feedback, history, and shared photos. The next step is making that data visibly improve the next ordering experience.
Actions in Bolt Food are mainly browsing restaurants, exploring food, adding to cart, placing orders, and reordering. But food ordering is not always simple. It involves mood, hunger, time, price, dietary rules, language, trust, and sometimes social pressure.
Given that Bolt is mainly providing mobility services, we noticed an opportunity for inter-service actions. Bolt has food, groceries, rides, scooters, and more, but these habits are not yet connected into one ecosystem relationship.
Current rewards are useful, but predictable: discounts on restaurants or delivery, subscription offers, cashback, and fixed percentage deals. These reduce pain, but they do not create much surprise or emotion. We wanted rewards to feel more like winning, not just saving: real, rare, contextual, and earned.
Investment already exists through order data, ratings, feedback, history, and shared photos. The next step is making that data visibly improve the next ordering experience.
What We Learned About Data
We grouped users through four main parameters: location, time, taste profile, and price.
Time and location are the most variable. A person behaves differently at home, at work, while travelling, during lunch, late at night, on a Friday, or on a Sunday morning. These changes affect what they want, how much time they have, and how open they are to exploration.
Taste profile behaves differently. Allergens, dietary restrictions, and religious limitations are not playful variables. If someone is vegan, lactose intolerant, gluten-free, or eats halal, that does not change because it is Friday or because they are at work. Taste is not always a compass for exploration. Often, it is a hard filter.
Price also shapes behavior. We can estimate price tolerance from grocery orders, average meal prices, order size, and whether someone usually orders solo, for family, or for a group.
Groceries became especially interesting because they provide ingredient-level habitual data. For example, lactose-free milk in a Sunday grocery order may later help Bolt understand that lactose warnings matter in food ordering too.
Time and location are the most variable. A person behaves differently at home, at work, while travelling, during lunch, late at night, on a Friday, or on a Sunday morning. These changes affect what they want, how much time they have, and how open they are to exploration.
Taste profile behaves differently. Allergens, dietary restrictions, and religious limitations are not playful variables. If someone is vegan, lactose intolerant, gluten-free, or eats halal, that does not change because it is Friday or because they are at work. Taste is not always a compass for exploration. Often, it is a hard filter.
Price also shapes behavior. We can estimate price tolerance from grocery orders, average meal prices, order size, and whether someone usually orders solo, for family, or for a group.
Groceries became especially interesting because they provide ingredient-level habitual data. For example, lactose-free milk in a Sunday grocery order may later help Bolt understand that lactose warnings matter in food ordering too.
Daily behavior changes between breakfast, lunch, dinner, and midnight snacks. Weather can also shift decisions: rain may increase delivery, while sun may make pickup or scooter use more attractive.
Weekly patterns can form around weekdays, like Taco Tuesday, sushi nights, Friday treats, or Sunday fridge restocking.
Monthly patterns include salary days, celebrations, and tighter budget periods. Yearly patterns include holidays, seasons, and returning from a trip to an empty fridge.
User maturity:
We also found that maturity changes the product goal. New users are more explorative because everything is new, so the goal is to reduce friction, confirm value, and prompt a second order. Occasional users need light encouragement to repeat and experiment. Frequent users need personalization and targeted discovery. Mature users are already in routines, so the goal is retention, high-value offers, and occasional novelty to prevent churn.
We also found that maturity changes the product goal. New users are more explorative because everything is new, so the goal is to reduce friction, confirm value, and prompt a second order. Occasional users need light encouragement to repeat and experiment. Frequent users need personalization and targeted discovery. Mature users are already in routines, so the goal is retention, high-value offers, and occasional novelty to prevent churn.
