When we help companies define their data strategy, one of the top priorities is to increase revenue with minimal waste. Depending on the company’s business model, what they do, what they sell, and volume of transactions, propensity models can be the highest impact initiative to implement.
What is a propensity model, from a business perspective?
A propensity model tells you who is most likely to take a specific action - buy, churn, upgrade, or click.
It lets you focus resources on people most likely to say “yes” and stop burning time and money on longshots. You stop wasting time, money and opporunity showing the wrong offer to the wrong people.
That translates to higher ROI, lower customer acquisition and fewer wasted cycles.
You stop treating all leads the same. You start betting where the odds are in your favour.
What problem do they solve? Why do you need it?
You’re wasting money.
Wihout a model, you spend the same on everyone - hot leads, cold leads, tire kickers. A propensity model tells you where to lean in and where to walk away.
It answers the question: “Who should I spend time and money on, and who should I ignore?”
What’s required for a successful implementation?
Propensity models are often the fastest way to increase revenue using data, which is why they’re among the first AI applications high-performing companies implement.
To succeed, start with a clear strategy:
- What problem are we solving? Is it really a problem for the business?
- What actions are we predicting?
- What will we do with the results?
- Who needs to act on this? Or will it be incorporated into scalable/automatable processes?
At a minimum, you need:
- Data that describes your customers (behaviour, demographics, etc.)
- Behaviour;
- Demographics;
- Etc.
- A history of the action you’re trying to predict:
- Part purchases;
- Clicks;
- Etc.
If you have those, you’re already in the game.
Why not just use ChatGPT or some other LLM?
LLMs are generalists. Propensity models are snipers.
LLMs generate words. Propensity models generate probabilities at scale.
This makes them incomparably more powerful an accurate than LLMs when solving this particular issue. Using an LLMs instead of a dedicated propensity models can be costing your company millions.
Using an LLM instead of the proper model is like using a blender to cut wood. It’s the wrong tool and it’s costing your money.
Will my company benefit from a propensity model?
Ask yourself:
Strategically:
- Is this your top priority for revenue growth?
- Is the outcome clear and measurable?
- Does everyone involved know how to act on the result?
Business Model:
- Is it hard to predict what a customer will do today?
- Do you have enough volume (hundreds, thousands, or millions) to justify building this?
Technical:
- Do you have the right data?
- Can you deploy the model into your current systems?
We’ve implemented high-performing propensity models across multiple industries:
- Banking: Predicted which clients were most likely to buy insurance, investment products, or credit lines, improving conversion and reducing outbound effort. We used demographics and past customer behaviour.
- Real Estate: For a UK-based platform, we matched client profiles and inquiries to thousands of venues, providing venue experts with a prioritized list of properties most likely to be chosen by each inquirer—even when limited client data was available.
- Retail: For a client in Denmark, we predicted which restaurants were most likely to purchase specific products, enabling their field sales team to focus visits on high-probability buyers and close deals faster.
All three resulted in measurable sales growth and cost reduction—less noise, more signal.
Final thoughts
If you have enough volume propensity models can be the most effective AI models to apply to your business.
They tell you exactly where you place your bets, so you stop guessing. They allow you to scale faster.
If you’re interested in implementing these and selling in a smarter way, we should talk.