AI means many things these days. What we call “AI” often refers to a mix of very different tools — with very different purposes.
If you’re building, improving, or rethinking how things work in your organisation, you need to know which tool fits which kind of job — from decision models that have been around for decades (like my favourite Operations Research), to systems that spot patterns and predict outcomes, to new generative tools that create something entirely new.
This is your field guide to the AI toolbox — what’s in it, what it’s good for, and how to pick the right one for the job.
1. Decision-Based AI
Drives productivity through structured process and resource optimisation.
This is the original backbone of applied AI — rooted in Operations Research (OR) and mathematical optimisation, dating back to World War II logistics. By the 1980s and 90s, OR had become standard in complex businesses — especially where decisions had to account for trade-offs, capacity, time, and cost.
And it hasn’t stood still. What used to take hours or days to compute now solves in seconds. Today, decision-oriented AI is embedded in how most large organisations run their operations, pricing strategies, financial plans, and supply chains — even if they don’t call it “AI.”
These tools don’t learn from data. They calculate optimal decisions based on constraints and objectives that you define — whether that’s budget, headcount, time, service levels, or environmental impact. You tell it the rules and trade-offs, and it finds the best path forward.
Real-world examples:
· Retail: Which products should go to which stores to avoid stockouts or waste?
· Banking/Insurance: Where should capital be allocated to maximise returns under regulatory limits?
· Healthcare: How should limited medicine, equipment, or staff be allocated across facilities to meet demand and minimise risk?
· Transport & Logistics: What combination of routes and vehicles delivers on-time service at the lowest cost?
· Airlines & Hospitality: How should seat or room prices adjust based on demand, competition, and seasonality?
· Energy: When and where should energy production be ramped up to meet demand while minimising cost?
· Telecoms: Where should infrastructure be built or upgraded to maximise service without overspending?
· Strategy: Which projects should be prioritised given budget and resource limits?
Who in your team should explore this — and why:
- Operations managers and planners → They make the day-to-day trade-offs this AI is designed to optimise.
- Analysts in supply chain, service delivery, or finance → They’re closest to the constraints, rules, and objectives that shape real-world performance.
- Strategic decision-makers working with complex trade-offs → OR helps them test what’s possible — and what’s optimal — across multiple future scenarios.
2. Prediction-Based AI
Recognises patterns and forecasts what’s likely to happen.
This is the space most people think of when they hear “AI” today — systems that use data to learn behaviours, forecast trends, and make automated recommendations. It’s the heart of modern machine learning (ML).
Most of the use cases — demand forecasting, churn prediction, fraud detection — were being handled long before AI got popular. Econometrics, Actuarial Science, and Operations Research were already modelling these behaviours. The difference is that today’s models are more automated, require more data, and are less transparent.
Prediction-AI is great for high-volume, fast-changing environments where outcomes are uncertain and patterns aren’t obvious.
With Machine Learning focusing purely on learning patterns from data, Operations Research still holds a competitive edge bringing structure to uncertainty — using tools like stochastic models, queuing theory, and probabilistic simulations to guide real-world decisions.
Real-world examples:
- Retail: Recommending products based on customer browsing and purchase history; forecasting demand by store or region.
- Banking/Insurance: Predicting fraud in real time, or anticipating which customers are likely to churn or default.
- Healthcare: Flagging high-risk patients, predicting disease progression, or estimating length of stay.
- Transport & Logistics: What delays are likely, and how will weather or traffic impact routes? Utilities: Predicting equipment failure before it happens using sensor data.
- Utilities: Which assets are most likely to fail next based on sensor data?
- Telecoms: Anticipating network load spikes or subscriber churn based on behaviour patterns.
Who in your team should explore this — and why:
- Data scientists and ML engineers → They’re already building models that can be applied more strategically across the business.
- Marketing and customer experience leads → Prediction AI helps them segment, target, and personalise in ways that manual methods can’t.
- Risk and compliance teams → ML can flag anomalies or early warning signs faster than human review.
3. Generative AI
Excels at creating new content, ideas, and scenarios
The newest wave of AI — and the one currently exploding in visibility. Generative AI doesn’t just analyse — it produces. Text, images, code, synthetic data. It learns patterns and then creates something new based on prompts or goals.
While the foundations were laid over the past decade, the inflection point came in 2022 with tools like ChatGPT. Suddenly, businesses could create at scale — without creative teams or development bottlenecks.
Generative AI isn’t here to make decisions. It’s here to speed up creativity, prototyping, and simulation.
Real-world examples:
- Retail: Writing product copy, social media posts, or chatbot responses; generating design variants for packaging.
- Banking/Insurance: Summarising policy documents, drafting responses to complex customer inquiries, or building simulated data for stress-testing models.
- Healthcare: Generating plain-language patient materials, creating summaries of clinical records, or simulating different treatment paths.
- Transport & Logistics: Drafting training manuals or driver communication; simulating customer support scenarios or route planning under edge cases.
- Internal operations: Creating employee onboarding materials, summarising reports, or generating code for repetitive tasks.
Who in your team should explore this — and why:
- Marketing, content, and communications teams → They can scale content creation and speed up experimentation.
- Customer service and product support leads → It can draft answers, automate chat, and reduce response times.
- Innovation and product teams → They can use it to prototype ideas, simulate scenarios, and accelerate iteration cycles.
- Founders and CEOs → It’s a powerful thinking partner — helping them brainstorm ideas, test narratives, pressure-test decisions, or draft content they don’t have time to write from scratch.
Red Flags: What Not to Do
Even the smartest tools can fall flat when applied carelessly. Here are some common pitfalls to avoid:
Decision-Oriented AI
- Overcomplicating problems that a basic logic model or spreadsheet could solve.
- Trying to model everything before understanding the real trade-offs.
- Building optimisation models in a silo, disconnected from real operations.
- Assuming the model will reveal the objective — when that still needs to be clearly defined by the business.
Prediction-Oriented AI
· Training models on poor-quality or biased data
· Trusting accuracy metrics without understanding real-world consequences
- Confusing high accuracy with actual business value.
- Deploying models before understanding how they’ll be integrated into workflows or decision points.
Generative AI
- Treating outputs as facts without human review.
- Using generic prompts and expecting targeted results.
- Assuming one-size-fits-all — when outputs often need tailoring to brand, audience, or regulation.
- Using it as a decision engine, when it’s best suited as a creative or exploratory partner.
Quick rule of thumb: If you’re unsure how the output was generated — or how it fits into your business rules — don’t automate based on it.
If You Want to Put This Guide Into Action — Ask Your Team:
- What are our most decision-heavy, prediction-heavy, or content-heavy problems?
- Are we solving for efficiency, insight, or scale?
- Are we already using any of these tools — but calling them something else?
- Where are we applying AI for the sake of it, rather than because it solves the real need?
Then what?
Use those answers to map your current efforts to the right type of AI — and spot gaps, overlaps, or missed opportunities. This guide gives you a shared language to realign your teams around the right tools for the job.
If You’re Only Going to Do One Thing After Reading This…
Take your five biggest business problems.
Ask:
Are we trying to decide, predict, or create?
Then work backwards: match the problem to the approach, and explore a small, testable way to apply the right tool. That one exercise will clarify next steps — and prevent months of going in the wrong direction.
Bringing It All Together
AI isn’t a single tool — and it doesn’t solve anything on its own. But if you understand the toolbox, you can move faster and make smarter decisions.
You move from hype to clarity. From buzzwords to action. From guessing to designing.
And you make technology fit your business — not the other way around. That’s what smart business is really about.