Borås Energi och Miljö: From Pilot to Full-Scale AI Rollout – with Local On-Prem Deployment

Introduction

When Borås Energi och Miljö (BEM) began exploring generative AI, it was not about finding a single "magic" solution – but about making everyday work easier across many different roles: technicians, customer service, HR, finance, and engineers. During 2025, BEM, together with Borås City, ran a pilot with Intric in which a large number of AI assistants and processes were developed and validated. Following a successful evaluation, they are now moving forward by deploying Intric with local on-prem hosting and planning a rollout across the entire organisation. The result is a portfolio of applications that both relieves pressure on experts, improves quality, and makes knowledge more accessible.

"As a company of societal importance, we must use new technology in a responsible way. We see AI as a tool for creating real value – for our employees, our operations, and our customers," says Magnus Kårestedt, CEO of Borås Energi och Miljö. "To succeed with that, we need to encourage curiosity, courage, and collaboration around ideas."

— Magnus Kårestedt, CEO

The Organisation at a Glance

• Organisation: Borås Energi och Miljö AB (BEM), municipally owned energy and environment company

• Mission: Energy, waste, water, and wastewater infrastructure in Borås, with a goal of becoming a sustainable circular city

• Scale: Approx. 380 employees, annual revenue approx. SEK 1.3 billion

• Operations: District heating, water supply, and waste services, with a focus on local resource efficiency and climate neutrality

• AI journey 2025: Pilot with Intric together with Borås City, with a dedicated pilot group, resulting in a decision to roll out Intric broadly

The Challenge

BEM identified several recurring obstacles that were hindering efficiency and quality:

• Knowledge existed – but was hard to access: Manuals, procedures, regulations, and system documentation were scattered and time-consuming to search through.

• Complex specialist environments: Everything from plant information, drawings, and control systems to ERP and HR/payroll rule sets.

• Onboarding and dependency on key individuals: New employees often needed colleagues' help to navigate instructions and ways of working.

• Quality risks in manual steps: For example, when reviewing supplier documentation or naming/applying metadata to technical information.

• Scaling required the right foundations: Several parts of the organisation highlighted the need for governance, data quality, and secure operations before moving from "proof of concept" to lasting value.

Why Intric

In the cloud-based pilot, Intric was used to quickly build and test AI applications close to the needs of the business, with assistants responding based on the organisation's own documents and able to be validated on an ongoing basis.

A key insight from the pilot: BEM clearly saw that the greatest benefit would come from using data with a higher classification level – information from operational systems, sensitive documents, and processes that could not be handled in the cloud. This motivated the decision to scale up with Intric in a local on-prem deployment following a successful evaluation.

Intric's granular access control and deployment flexibility were decisive factors. The ability to run on-prem in their own environment, combined with clear control over access to assistants and data, created the conditions for secure integration with core operational systems and handling of sensitive information – without compromising on GDPR and other regulatory and security compliance requirements.

How the Journey Began

The journey gained momentum through a working method in which ideas and needs came from multiple functions – not just IT. A group of AI ambassadors (representatives from different parts of the organisation) was appointed and met regularly to share experiences and work in a structured way. Colleagues took initiative and developed support for everything from "chat with manuals" in control system environments to assistants for customer service, HR, finance, and the web. The focus was on quickly turning needs into working prototypes, evaluating the value, and prioritising the next steps.

How It Was Carried Out

1. Gather use cases from across the organisation – both small friction points and more strategic needs.

2. Train employees in the pilot on AI and Intric to become so-called AI ambassadors.

3. AI ambassadors build assistants themselves using existing knowledge (PDFs, procedures, manuals, regulations, tariffs, contracts, industry standards, and legislation).

4. Iterate with feedback: adjust prompts, supplement source material, and manage limitations.

5. Evaluate using a clear model: bring the pilots together, make the value visible, and present findings to the full management team.

6. Make a decision on next steps: following the pilot and management-anchored evaluation – deploy Intric with local on-prem hosting and a plan for rollout.

Results

• Over 25 AI applications/assistants were developed and evaluated across technology, customer service, HR, finance, and web.

• Faster responses in complex documentation – increasing actual use of manuals and instructions.

• Higher quality through standardisation (e.g. naming conventions, metadata, consistent report texts).

• Public chatbot on the website received 336 incoming questions during November (launched 20 October).

• A well-founded evaluation model was established, with findings presented to the full management team ahead of the decision.

What stands out most is the breadth: instead of chasing a single "big AI project," BEM built a portfolio of assistants tied to concrete work tasks. Technical teams use AI for support with extensive manuals, reverse engineering, and creating simpler tools. Customer-facing and administrative functions receive help with everything from waste tariffs and HR queries to accounting suggestions and monthly narrative texts.

Examples of AI Applications

• Technology / Automation: "Chat with manuals" for ABB 800xA; support for reverse engineering and conversion of legacy code

• Technical information management: Assistant that interprets title blocks/name fields in scanned drawings, eliminating the need to manually open each file

• Web & Customer: Chatbot on borasem.se (launched 20 October; 336 questions in November alone), drawing on website content and relevant documents

• Customer service: Assistant providing quick, clear answers from instructions/manuals – particularly valuable during onboarding

• HR: HR assistant for general questions about HR, payroll, Flex HRM, legislation, and collective agreements

• Finance: Agresso assistant, accounting/coding assistant, financial reporting assistant (K3 support), bank reconciliation assistant, and assistant for drafting/summarising monthly narrative texts

"Our broad initiative with AI ambassadors across the entire organisation has been crucial for spreading interest and rapidly building knowledge. The AI labs have also become important forums where we share insights and solve challenges together."

— Fredrik Andersson, Enterprise Architect at Borås Energi och Miljö

Lessons for Other Organisations

1. Start broad – but validate quickly. Many small and medium-sized use cases can create clear momentum, but require structured prioritisation.

2. The knowledge base is a product. Documents must be maintained, supplemented, and version-controlled – especially when rules and ways of working change.

3. Scaling requires governance. On-prem deployment, data quality, and clear frameworks (e.g. GDPR and sensitive information) need to be part of the plan, not a "later problem."

4. Make it easy to get started. Simple entry points and clear evidence of value lower the threshold for colleagues who are not yet convinced.

5. Access control is key. The ability to securely manage different types of information and control access at the user level creates the conditions for broad adoption without security compromises.

Next Steps

Following the pilot and evaluation, BEM is taking the next step with a deployment of Intric in a local on-prem environment and is planning a rollout to the entire organisation. The ongoing focus for a successful implementation will be: secure operation and data management, connections to key systems and data sources, improved data quality/structure, and a continued clear and actionable AI policy and strategy.