MVP development specifics for an AI product

In this post, we'll discuss building MVPs for consumer products backed by Artificial Intelligence. The development process of AI-driven applications can include either using off-the-shelf AI resources or creating custom models, or fine-tuning them with more data.

AI-driven products have different product development cycles when compared to “traditional” tech products. Yet, they are still user-facing solutions, and the general iterative approach to product development benefits AI-driven startups as well.

Data-driven vs data products

When creating solutions backed by AI, it is important to understand the difference between data-driven and data products, as AI MVP development is more related to the latter. Filip Romeling provides some good examples in his guide to developing your AI-MVP:

• Gmail is a product that uses data but isn’t a data product itself. But a spam filter integrated into it is a data product that identifies and classifies spam and malicious mail.
• Netflix, a streaming platform, is a data-driven product. The algorithm that suggests what to watch based on multiple parameters is a data product within it.

Consumer products based on AI are products with the primary goal to use data to facilitate an end goal and provide value to and solve problems for its customers.

AI-driven product development lifecycle

Let’s now look closer at the AI product development process which can be defined as an iterative six-step cycle. We’ll focus on the first three steps that include this specific work with AI implementation.

• Design sprint or scoping is a phase when the product team looks at the problem from a customer perspective, turns ideas into testable hypotheses, and creates an MVP to validate them.
• Data analysis/processing. It is an optional step and is relevant when you use specific data of your own or if you need to train the custom AI model. It requires the enrollment of data science engineers who will work with the information your AI product uses, processes, and provides.
• AI solution development and experiments. This step might include the training part and multiple tests before the solution becomes production-ready. If you use off-the-shelf models and instruments, your team will have time to focus on the engineering part.

Even though the development process is somewhat different with AI-driven products than with other tech products, overall, it shares the core principles with the traditional lean startup approach: launching product versions quickly, gathering and analyzing the feedback, and proceeding with the necessary pivots and adjustments.

MVP step in AI-driven product development

The MVP development step allows you to identify the weaknesses of an idea early on and avoid investing in a product with multiple potential shortcomings. In complex AI product development, where there is a lot of unpredictability and uncertainty, this phase lowers lots of risks.

Overall, from our team’s experience, the MVP step is absolutely vital, no matter what product you build.

One of our recent insights from exploring the AI-MVP specifics is that at this step off-the-shelf ML models and solutions are more effective to use. Training your own model takes time and resources. A ready-to-use one will help launch an MVP in weeks, quickly test your hypothesis, and start with the next phase where you have a validated idea, some market traction, and feedback to work on.

With this data and on the condition that your product needs it, you can invest in training a Machine Learning model. But not sooner.

Launching an AI product MVP in 7 weeks – Case Study

Recently, our team had a chance to work with the seasoned ML expert Mercan Topkara and helped her build an MVP for her new product – Hashtag In Real Life (#inreallife). We won’t disclose all the inner cuisine of the project as it is still under development, but we’d gladly share some high-level insights we received during this collaboration.

Product background

Our team collaborated with the product founder Mercan Topkara previously on several projects. Together, we worked on scaling a recommendations tool for an EdTech marketplace to millions of users and separately on building a podcast recommendation platform.

Mercan has 16 years of experience in building large-scale natural language processing solutions for content platforms and marketplaces. In the middle of summer 2023, she reached out to our team with this great new product idea – create a platform that would offer people ideas with meaningful real-time experiences that would match their interests, location, and abilities.

The product's mission to encourage real-world exploration among both adults and children resonated with our worldview a lot, and we were more than happy to onboard.

Team and process

For this project, we built a team of two engineers and a DevOps engineer. Our team collaborated with Mercan during the whole MVP development step which covered scoping, design, implementation, and launch.

One of the smartest decisions we’ve made together with Mercan was to use off-the-shelf AI solutions, namely, Open AI and LangChain, to build the Hashtag In Real Life MVP. This greatly sped up the whole process and allowed us to release the MVP in 7 weeks.

The MVP for the Hashtag In Real Life allowed users to:

• Specify what kind of experience they are looking for stated as their aspirations in the text. We extract location, group size, age range of the group, budget, and location from this text, and make it easier for the user to update them if needed. We also ask them if they have any restrictions we should take into account (like allergies) in text format.
• Generate a list of activities based on the provided parameters.
• Save personalized suggestions to revisit after, as well as share them with other people.

The scheme below illustrates how a client request was processed on the back end during the MVP step.

Next steps

We successfully launched product MVP in the defined terms and are already planning the second phase. The first user feedback was received and processed. Building an MVP on a ready model allowed us to finish the development in minimal terms.

Our team is already looking forward to making the platform more robust and fast. We plan to work on the result's relevance in the next iteration as well. So, stay tuned for the #inreallife product updates.

Insights for AI startups

• Start product development with the MVP step and pay thorough attention to MVP scoping. Include only the most important functionality that leads a customer from state A to state B.
• Use off-the-shelf models or AI technology process solutions (OpenAI, PaLM API, Vertex, and others) to build an MVP and invest in training models after the idea is validated.
• Like no other solutions, AI-driven products require constant care and monitoring. Once the product goes live, together with your engineering and data science teams keep fine-tuning it and validating the results it provides.

Have an idea for an AI startup in mind?

At Wise Engineering, we build dedicated engineering teams to build consumer products backed by AI for different segments and industries.

Artificial Intelligence is nothing but an experiment. Complex and exciting at the same time. We’d gladly go on this journey with you and your team (or become your team if you don’t have any).

Contact us to discuss what we can achieve together.