"What kinds of AI are there?" It is one of the most common questions in our workshops. And the answer you find everywhere, narrow AI versus general AI versus super AI, is completely useless when you want to decide where to start tomorrow morning.
That framework is about what AI theoretically is. Not what AI practically does. And as a director or manager, you need the second.
Four capabilities
AI has four core capabilities that are relevant to organisations. They sound simple. The combinations are powerful.
Sense. AI can perceive. Read text, listen to audio, recognise images, scan documents. Everything that arrives as unstructured information, emails, customer conversations, invoices, product photos, AI can perceive and convert into structured data.
A logistics company that manually enters hundreds of delivery notes every day can run that stream through an AI layer that reads, recognises, and structures. The work does not disappear. It shifts from entering to checking.
Interpret. AI can interpret. Recognise patterns in data, categorise, assess risk, signal trends. The difference from a spreadsheet: AI recognises patterns in unstructured, messy, incomplete information.
A retailer receiving thousands of customer reviews every month can use AI to identify which themes recur most often, where sentiment is shifting, where a problem is growing before it escalates. No human reads a thousand reviews per month. AI does.
Generate. AI can produce. Write texts, generate reports, draft proposals, create variations. The most familiar application, and the most overrated. Generating without the other three capabilities is a machine producing things without context.
Where it gets interesting: AI that generates based on what it has perceived and interpreted. A quote automatically drafted based on customer history (Sense), market position (Interpret), and internal guidelines (knowledge). That is not a standard template. That is a bespoke first draft in minutes.
Decide. AI can make recommendations. Which complaints require escalation? Which leads are most promising? Which products need reordering? AI does not decide. AI advises. The decision stays with the human. But the advice can be based on more data, processed faster, than any human could ever survey.
This is where a boundary needs to be deliberately designed: advice and authority kept separate. AI may recommend. Humans decide.
The combinations
The real value rarely lies in a single capability. The power is in the chain.
Sense + Interpret: an insurer that automatically reads claims (Sense) and classifies them by urgency and fraud risk (Interpret). What previously took three days takes three minutes. The specialist spends her time on complex cases, not on triage.
Sense + Interpret + Generate: a consultancy that records client conversations (Sense), extracts the key themes (Interpret), and generates a structured meeting summary (Generate). The summary is ready before the consultant is back at the office.
All four: a customer service operation that reads incoming queries (Sense), classifies by type and urgency (Interpret), drafts a response (Generate), and recommends whether it can be sent directly or requires human review (Decide). The employee becomes a reviewer rather than a typist.
What this means for your strategy
Most organisations start with Generate. They have AI write texts, make presentations, draft emails. That produces something, but it is the least strategic application. The real difference lies in Sense and Interpret, the capabilities that let your organisation see things you could not see yesterday.
The question for every AI investment: which capability are we adding? And which combination delivers the most value for this specific process?
Start with the pain. Identify which capability fits. Build the smallest possible proof. Test with real people. And measure not just speed but also quality.
That is an AI strategy that starts with what your organisation can become, not just what it already was.