Testing against real material
Before the solution goes live, we test it against examples from your business and measure how often it gets things right. Evaluation continues once it is running.
- Home
- Our expertise
- AI solutions
AI that works with your information.
We build AI solutions for businesses: assistants that answer from your documents, agents that carry out steps in your systems, and AI that interprets information in automated workflows. Applied AI is one of our core strengths, and we use it where a task needs information to be interpreted, summarised or assessed.
What AI can do
From assistants to AI in the background.
- 01
Internal assistants
Employees ask questions and get answers based on your documents, guidelines and systems, with a reference to the source.
- 02
Document understanding
Contracts, applications, invoices and reports are read and interpreted, and the details are extracted and passed on.
- 03
Agents
AI that carries out several steps in a row: retrieving information, filling in systems, preparing drafts and handing over to a person where needed.
- 04
AI in automated workflows
AI as part of a workflow that runs without anyone using it directly, for example to classify requests, summarise calls or enrich records.
- 05
Analysis and research
Compiling material for decisions, meetings and reports, from your own sources and from public ones.
How an assistant can work
Answers you can check.
- A question is asked
- Retrieves from your sources
- Answer with source
- Action in a system
- Review where needed
An employee asks what applies under a customer contract. The assistant searches the documents and systems that person has access to, answers with a reference to where the information is, and can create a case or update a record if needed. Steps that affect customers or money go through a person.
AI or conventional automation
AI where it is needed, conventional software where that is enough.
AI fits when
- The material is unstructured: text, documents, email or calls.
- The task needs interpretation, summarising or judgement.
- The questions vary and cannot be anticipated with rules.
Conventional automation is enough when
- The rules are clear and the result should be the same every time.
- The information is already structured in a system.
- A mistake can be prevented with a rule or a check.
Often the solution is a combination: AI that interprets and suggests, within a system of rules, checks and integrations.
Quality and security
Control over what the AI does and what it can reach.
Source references
Answers based on your documents show where the information comes from, so it can be checked.
Permissions
We limit access to the information the solution needs and design it so that an assistant follows the same permissions as the person asking.
Human review
Steps where a mistake is costly, such as decisions about customers or money, go through a person before they are carried out.
Data, models and cost
Assessed and documented for each engagement.
Data processing
Where data is processed, how long it is kept and which terms apply with each provider are assessed for every engagement and documented.
Model choice
We choose model and provider based on the task, the requirements for quality, security and cost, and your own guidelines.
Running costs
AI is paid for by usage. Before we build, we estimate what the solution will cost to run at the usage you expect.
- How do you build an internal AI assistant with company documents?
- Documents and systems are connected to the assistant so it can find the relevant parts and answer from them, with a reference to the source. Most of the work lies in choosing the right material, keeping it up to date, managing permissions and testing the answers against real questions from the business.
- How can AI access internal systems without exposing sensitive information?
- By limiting access to the information the solution needs, letting the assistant follow the user’s permissions, logging what is retrieved, and choosing providers and terms that meet your requirements. Which requirements apply and how they are met is assessed and documented for each engagement.
- What does it cost to build and run an AI agent?
- The development cost depends on how many steps and systems the agent handles and how much needs testing. The running cost depends mainly on how much it is used and which model it uses. We estimate both before we build.
- When should a business choose conventional automation instead of AI?
- When the rules are clear and the information is already structured. Conventional software is then cheaper to run, gives the same result every time and is easier to troubleshoot. AI is needed when the material has to be interpreted.
- What happens when the AI gets something wrong?
- It will happen from time to time. That is why we test against real material, show sources, check results against rules and data, and have a person review the steps where a mistake is costly.
- Which AI models do you use?
- It depends on the task and your requirements. We choose model and provider for each engagement and design the solution so the model can be replaced.
Let’s start with your everyday work
Where would AI help most in your business?
We start with a conversation to understand your needs and explore the possibilities together. You do not need to know what could be improved – we help you explore that.