AI implementation at a roofing company
In brief
A company working in flat and pitched roofing asked me to build AI support for three recurring steps in its estimating and sales process. The solution helps the team assess incoming requests, check quotes before they are sent and prepare detailed pricing requests for suppliers. It is built around the company's own rules, documents and terminology.
The challenge
The estimating team constantly receives documents to assess, including specifications, bills of quantities, quotes and supplier emails. Three moments in that process were particularly time-consuming and left room for error: deciding whether an incoming request was worth pursuing, checking that a quote was complete before it went out, and drafting pricing requests to suppliers line by line.
The approach
The project started with a day on site, not with a project plan written in advance. Together with the estimating team, we reviewed dozens of possible use cases and divided them into three groups: what could be built straight away, what required a connection to other systems, and what fell outside the scope of a short project. We then selected three concrete cases from the first group.
The solution was built using the company's own documents, including files, historical emails, quotes and reference lists. We did not rely on generic examples. This material helped shape and refine the solution throughout the project, including support for a less common type of file that only became apparent later.
Testing was carried out on a broad set of company-specific files with known outcomes. This made it possible to compare each result with a real human decision and test whether the defined rules were being applied correctly.
The result
A working solution was delivered in just three days of work.
The three selected use cases were turned into practical tools that the team can use without having to learn a specific command or formulate a technical prompt. Every output can be traced back to its source, with clear documentation of what has been demonstrated and what has not yet been proven.
The rules and knowledge used by the three tools are defined once and shared across them. This avoids repeating the same business logic in different places and having it gradually drift apart.
This first project also established the groundwork for the next phase. The same approach, business rules and technical foundations can now be reused as I develop additional AI use cases for the company.
What this shows
This project shows a particular way of working. It starts from the real work, not from the technology. It builds on the client's own documents and processes rather than generic examples. It tests the solution against decisions that are already known and makes clear what those tests actually prove.
The scope also remained deliberately narrow. This made it possible to deliver something useful within the project while keeping a clear path towards a follow-up phase, rather than trying to build a complete solution all at once.
AI becomes useful when it is built around your real work. If you have a process where AI could save time, reduce repetitive work or support better decisions, let's talk.