What AI does, explained without jargon
A short guide to AI, machine learning and digitalisation, written for business people. It helps you know what you're buying and what questions to ask.
Why AI, ML and digitalisation
Most companies already have the data they need, but struggle to get to it. Digitalisation brings it together in one place, and a layer of AI lets you ask about it in everyday language.
What AI actually is
A program that is very good at recognising patterns in large amounts of data. It isn't conscious and doesn't know things the way a person does. You use it as an assistant, and you check what it gives you.
What machine learning is
It's like a new employee who learns how things work by watching what happened month after month. The difference is that it looks at all the data at once and forgets nothing.
Automation and AI
Simple automation moves data from one place to another by a fixed rule. AI reads the content and can make a decision based on it. Both are useful, for different things.
Why now
Companies that have their data in one place can decide based on numbers. The later they start, the more history they have to put in order.
Digitalisation takes over the repetitive work, and people are left with more time for decisions.
Predictions and machine learning
Machine learning turns an experienced person's intuition into numbers and applies it the same way every day. It answers the question every company asks: what comes next?
What it can predict
Demand in the coming days, the money coming in and going out, the risk of an invoice being paid late, and anomalies, meaning the places where you lose money without noticing.
Most likely, not certain
A prediction is the best estimate based on patterns that repeat, with a margin of uncertainty shown. For the next few days it is precise; for months ahead it is only a guide.
The same result from the same data
A prediction model gives the same result every time it receives the same data, so you can check where each number comes from.
When it works and when it doesn't
If the input data is wrong, the prediction is wrong too. Models work well on patterns that repeat and cannot foresee a sudden crisis or a very large new customer.
The model tells you what is most likely to happen next. What you do about it is your decision.
How a language model works
You don't need to be technical to understand the basic idea. People who understand it use AI better and know when not to rely on it.
It predicts the next word
A language model, like the one behind a chat, doesn't know things the way a person does. It picks the most likely next word, based on a huge number of examples. That's why, when it has no clear data, it can make up answers that sound right.
Grounding
To keep it from making things up, we tie it to your approved data and ask it to answer only from there. When it can't find the answer there, it has to say so.
Temperature
A setting that controls how varied the model's answers are. For business numbers we keep it low, so answers are precise and consistent.
Language models and machine learning
A language model is good at text and explanations; machine learning is good at predictions from numbers. We use each one where it does well.
An AI is as good as its data and as the guidance it gets.
Terms worth knowing
In brief, without jargon.
- Prompt
- The instruction you give the AI. The clearer it is and the more context it has, the better the answer.
- Hallucination
- An answer the AI makes up, which sounds convincing but is wrong. You prevent it through grounding.
- Grounding
- Tying the AI to verified sources, such as your company's data, so it answers from them.
- Token
- The chunk of text a model works with, about three quarters of a word. Cost is measured in tokens.
- Read-only
- Access for reading only: the AI can read the data, but cannot delete or change it.
- AI agent
- An AI that carries a task through to the end: it reads, decides and acts.
- LLM
- Large Language Model, the model that understands and writes text, for example Claude.
- Model (ML)
- A program that has learned patterns from historical data in order to make predictions.
Not sure where to start?
Answer six questions about your company. Within one or two working days we send you your company's AI plan, written by a person on the Askera team.
Want to see what AI can do in your company?
We'll do a free assessment and tell you where it's worth starting.