Artificial Intelligence & Construction
How Frennly built an intelligent technical assistant for the metal construction industry with Swiftask
Yvan Legeay, managing partner at Frennly, explains how his agency used Swiftask to build an AI chatbot able to answer complex technical questions from construction professionals.
THE CONTEXT
A digital agency that explores artificial intelligence
Frennly is a digital agency specializing in digital transformation and the development of custom applications.
The Frennly agency
Frennly is a digital agency specializing in digital transformation and the development of custom applications. The agency works directly with its clients on their digital projects and also collaborates with several communication agencies to support them with their technical needs.
Yvan Legeay, Associate Director of Frennly, quickly became interested in the possibilities offered by artificial intelligence.
“As an agency specializing in digital transformation and the delivery of applications for our clients, we quickly became interested in how AI could be integrated into projects we might be asked to carry out.”
THE CHALLENGE
Making complex technical documentation accessible
The opportunity arose during the creation of a new website for CTICM.
The client: CTICM
The Industrial Technical Center for Metal Construction (CTICM) has specialized expertise and produces a vast amount of documentation to support metal construction professionals and project developers.
CTICM had already established a technical support service. Before the AI solution, professionals had to consult a directory of experts on various thematic topics and call them by phone. An engineer would then respond to users’ questions based on their documentation sources.
Limited availability
Experts were not always available to answer calls, leading to delays and user frustration.
Difficult access to information
Despite the wealth of available documentation, users struggled to quickly find answers to their technical questions.
Suboptimal use of resources
Highly skilled engineers spent time answering questions that were sometimes quite basic, instead of focusing on higher-value tasks.
“This idea was born almost two years ago, at the dawn of large language models (LLMs). It was almost a crazy thought to imagine having an intelligent chat that could answer highly technical questions.”
Identified challenges
The challenge was significant: to create "the ChatGPT for metal construction," capable of answering complex technical questions by leveraging CTICM's documentation database.
THE DECISION PROCESS
Why Swiftask?
Facing this challenge, Frennly explored several options
Alternatives considered
The team had already had the opportunity to meet Swiftask and discuss AI-related topics. This pre-existing relationship made it easier to submit the project to the Swiftask team.
Building from scratch
The agency had approached AI developers for a custom solution. This option involved developing a full chat software, building a RAG (Retrieval Augmented Generation) system, and using a large language model API.
Issue identified
"A complex project to manage and also very expensive in terms of financial resources to develop with no guarantee of results. Moreover, it was impossible to know the monthly usage costs of this tool in advance."
Off-the-shelf chatbot solutions
Other existing solutions on the market could have been integrated, but they did not meet the specific needs of the project.
Swiftask
Swiftask emerged as the ideal solution for several reasons:
An existing software foundation
Swiftask already offered a platform to build on for creating an agent and connecting the document base, without having to develop everything from scratch.
Domain expertise in prompting
The Swiftask team brought specialized skills to develop an agent capable of effectively answering technical questions.
Technical flexibility
Frennly could retain its domain knowledge as a web developer. The agency created a visual interface connected to Swiftask's APIs to query the agent created on the platform.
Integrated user experience
The chat is displayed directly on the client's website, offering a seamless and integrated user experience.
THE SOLUTION
A custom AI agent integrated into the website
The implementation took place in several stages
Swiftask support
The Swiftask team supported Frennly in integrating an agent capable of storing a large volume of technical data and responding to user inquiries.
Agent training
The core of the solution relies on training the agent with the technical data produced by the engineers at the Industrial Technical Center for Metal Construction (CTICM).
Web integration by Frennly
The agency created a visual interface connected to Swiftask's APIs. The chat appears directly on the client’s website, providing a seamless and integrated user experience.
Operational conversational agent
The agent can now answer both simple and complex questions posed by users directly on the website.
“The particularity is that we didn’t necessarily use many of the platform’s features, but rather focused on a very specific use case. This hybrid approach allows us to combine the best of both worlds: the power and intelligence of Swiftask for the AI component, and Frennly’s expertise for web integration and user experience.”
ADOPTION
Managing change
At Frennly: a smooth onboarding
On Frennly's side, getting up to speed with the tool was fairly straightforward. The technical teams quickly understood how to integrate the solution and use it effectively.
At the client: cultural change required
On the other hand, supporting the end client required more effort.
“Where it required cultural change and support, it was on the client's side”
Yvan Legeay
Managing Director, Frennly
The CTICM team had to be introduced to artificial intelligence, taught how to train the agent, and how to use it to respond to client requests while providing the best possible answer.
Two profiles within the client
Within the client's working group, the situation was mixed:
The believers
“Some were 100% convinced that AI could answer precise, technical questions”
The skeptics
“And then there were those who needed to be convinced that the tool would provide relevant answers.”
Strategic preparation phase
Before launching the agent, the team set up a contact form. This form served a dual purpose:
Allow users to ask their questions
Collect these questions to train the agent
This preparation phase proved crucial for what followed.
Timeline
Start in September • Testing and POC • Production launch in December 2025
THE RESULTS
A measurable reduction in inquiries
Although the timeframe is still short, the initial results are encouraging.
30% reduction in inquiries
In November, before full deployment, CTICM received 45 inquiries via the form. By January, this number had dropped to 30. Users now get their answers directly by interacting with the agent.
30 conversations handled
The agent has handled around 30 conversations since going live. Without the Swiftask solution, these inquiries would have had to be managed manually by engineers, either by phone or through the contact form.
Freeing up engineers’ time
The main objective has been achieved. CTICM engineers can now focus on higher-value tasks instead of answering basic questions.
24/7 technical assistant
The agent handles first-level support, available 24/7, providing immediate responses to construction professionals.
“The benefit isn’t so much in enterprise users having the tool in their hands, but in using it as a conversational agent directly integrated on a website to address a specific challenge: providing a true online technical assistant.”
MEASURED IMPACT
Key figures
THE TESTIMONIAL
Swiftask as an all in one solution
“For me, it’s an all-in-one solution and a truly simple answer amid the jungle of AI tools available to us."
"We are often faced with the problem of a client asking, ‘Which is the best AI tool, which one should I use?’ Swiftask addresses this concern: you don’t need to create accounts on every platform. With Swiftask, we can answer the question, ‘Which tool do I integrate?’ We are connected to many things."
"It’s a real asset in building agents, given the range of skills and configuration options available. It’s not a tool 100% specialized in one domain—it’s highly multidisciplinary, which is a major advantage.”
CONCLUSION
A win win partnership
The project carried out by Frennly for CTICM perfectly illustrates how Swiftask can be integrated into complex digital projects. By combining the digital agency’s industry expertise with the power of the Swiftask platform, the result is a tailor-made solution that precisely meets the end client’s needs.
This collaboration also demonstrates Swiftask’s flexibility: the platform does not impose itself as a rigid solution but adapts to the needs and technical skills of the partners using it. Frennly was able to maintain its role as a web developer while leveraging Swiftask’s artificial intelligence for the conversational agent component.