Blog

Behind the project: Building an AI-powered pricing engine

Kseniia Bacho AI Engineer, Solita

Published 16 Nov 2026

Reading time 5 min

We work with customers across industries to solve real business challenges and to create impact. Every project is different, with its own goals, people, and technology. In this series, we take you behind the scenes to meet the people and discover the choices and everyday work that bring each project to life.

In this story, you learn how an AI-powered pricing engine helps a customer in the information and communication technology industry automate the pricing of used devices, making it easier to handle large volumes of products while improving the speed and quality of pricing.

Automating the pricing of used devices

The project focused on building an automated pricing system for a retailer that buys used devices from consumers and resells them. Previously, pricing was largely done manually. With hundreds of devices, brands, models, and different conditions to consider, the work involved managing thousands of individual prices and was time-consuming.

Our goal was to build a pricing engine that could suggest prices automatically, explain the reasoning behind them, and eventually allow experts to approve prices without having to review every suggestion manually.

We started the project in March 2026, and the system went into production in October. Rather than beginning with a long development phase before showing anything to the customer, we wanted to understand the business and start building something concrete as quickly as possible.

Bringing business experts and engineers together

We used Solita’s CollabAI method, bringing together the people who understand the business and the people who build the technology. 

As I like to describe it, the idea is to put the people who know and the people who build in one room, and let AI do the heavy lifting between their decisions.

We started with two intensive weeks of on-site collaboration. We had full-day workshops every other day, followed by six weeks of a lighter development and refinement period. During the workshops, we learned how the customer’s pricing processes worked, what mattered to the business, and which factors needed to influence the prices.

We also used AI to support the collaboration itself. We transcribed meetings, created summaries, and collected the knowledge from each workshop in Markdown files. This gave us a shared foundation for building the system. We had an initial prototype to show the customer within the first weeks of the project and could start improving it based on their feedback.

The team included an AI Engineer, a Data Engineer, and a facilitator from Solita, working closely with the customer’s Pricing Experts, business representatives, and Full-Stack Developers. The facilitator helped keep discussions focused and made sure we continued making progress during the workshops.

Finding the right approach for the problem

There are many ways to build an AI system, and one of the interesting challenges was figuring out which approach would work best for this particular use case. We experimented with traditional machine learning, an agentic AI approach, and a deterministic algorithm built with AI. After comparing different options, the deterministic algorithm developed with AI has proven to be the most reliable.

A big part of the work involved understanding the historical pricing data and identifying which information and signals should influence the suggested prices. We also needed to make sure the system behaved predictably. For example, we introduced hard limits and conditions to ensure that price suggestions were only made when the relevant signals were stable.

Trust was particularly important because the pricing experts needed to understand and rely on the suggestions. We therefore worked together with the customer to define the rules behind the pricing engine.

We asked what should happen in specific situations, translated those decisions into the engine’s logic, and tested the results against historical data. The customer could then review the suggested prices and the reasoning behind them, and we adjusted the rules together. It was an iterative process of testing, reviewing, and refining the engine.

Making the logic understandable

One important aspect of the collaboration was making the pricing logic accessible to people who weren’t working directly with the code. We documented the rules in plain language, creating a shared artefact that both business experts and engineers could review. The code was then built on top of this description. This meant the customer could easily read the logic, correct our understanding when needed, and help make the rules more precise.

As Engineers, we aren’t necessarily experts in pricing, economics, or the customer’s business. Having the opportunity to work closely with people who understand these areas helped us get beyond the technical requirements and understand how the business actually operates.

Adapting as AI technology evolves

AI technology developed significantly during the six months of the project. We started with one AI model, but new and more powerful models were released along the way. We wanted to make sure we were using the most suitable model for our needs, so we continuously evaluated new options and improved the system.

Each new model requires some research. As an AI Engineer, I need to understand how it differs from previous models, what its capabilities are, and how we can use it effectively. We also need to consider the customer’s requirements around data security and where information is processed. For customers who require data to remain within Europe and comply with GDPR, the choice of models and providers can be limited.

This means that technical decisions aren’t only about finding the most powerful model. We also need to understand the environment in which it can be used and what works for the particular customer.

From manual work to everyday use

The pricing engine is now in production. Pricing experts no longer need to set all the prices manually. Instead, the system generates suggestions with explanations, allowing them to review the results much more efficiently. As trust in the system grows, some suggestions can also be approved automatically.

For me, one of the most rewarding parts of the project is seeing that the system is in use. 

In some projects, you build something and never really know whether people will use it. Here, we can see that it makes a real difference to people’s everyday work.

This project has also given me a lot of experience in working together with people from different industries and areas of expertise. I’ve since been sharing these learnings internally at Solita and running CollabAI workshops to help other Solitans make better use of AI in workshops and team collaboration.

I think there’s a lot of potential in bringing business experts, data, and engineering together from the very beginning. When you capture what people know in a form that both humans and tools can understand, you can do some really powerful things.

AI also creates opportunities for projects that might not have been possible just a few years ago. The pricing experts are still very much part of the process, but they no longer have to spend so much time setting prices manually. They can focus more on reviewing the suggestions and making pricing decisions.

Curious to work on similar projects? Take a look at our open positions and remember to also subscribe to our talent community newsletter.

  1. Culture