What the Road Makes Possible

The last part of Building a Data Team. Stage 5 ended the road. This is what is waiting at the end of it.

The principle. Foundations are not the goal. They are what let one team answer many business questions, instead of starting from nothing every time.

The first five stages can feel like a lot of work with little to show for it. Documents, data controls, rules and meetings do not look exciting to a business.

This part is the answer to the question you will be asked: what do we get for all of this?

Once your data can be trusted and your path is clear, each new idea gets cheaper, because it stands on what already exists. Here is what that looks like for us, in an insurance business. Your list will be different, but the pattern will be the same. Some of these are in daily use, and others are still being built or rolled out.

Know each customer

For years, businesses have sorted customers into a few broad groups. With trusted data, you can go much further.

• Hyper segmentation. Group customers by how they really behave and what they need, not only by age or area. The value: decisions fit the customer in front of you.
• Hyper personalisation. Use those groups to give each person a more relevant message or offer. The value: people engage more, because what they get is useful to them.
• The next best product. Work out which product is the best fit for each customer. The value: fewer unwanted offers, and better conversations.
• A customer's long term value. Estimate what a customer is likely to bring in over time. The value: you can decide where to spend your time and money.

Keep your customers

It costs far less to keep a customer than to win a new one. The hard part is knowing who is about to leave, and why.

• Who may leave. Spot customers who are likely to cancel soon. The value: the retention team calls the right people first.
• Why people really left. Look at what happened in the weeks before someone cancelled, not only at what they said. The value: you fix the cause, not just the symptom.
• What customers are saying. Search thousands of written reviews by asking plain questions. The value: you hear problems sooner, in the customer's own words.

Understand how people drive

Telematics means measuring how a vehicle is driven, using a device or a phone. Once that data is reliable, one set of driving measures can feed many uses.

• Rewards for good driving. Give safer drivers money back. The value: customers have a reason to drive better.
• Fairer prices. Compare what each customer pays with the risk their driving shows. The value: you stop losing money on underpriced risk, and losing good customers who are overcharged.
• Seasonal risk. See how risk changes over the holidays and on wet winter roads. The value: prices and plans can follow the real pattern.
• Early warning. Notice when someone's driving is getting worse compared with their own normal. The value: you can step in before a claim happens.
• Devices that go quiet. Find devices that have stopped recording. The value: no decisions based on data that stopped arriving.

Understand the big risks

Some risks do not depend on one customer at all. They depend on where many customers are.

• Floods and fires. Turn an address into a measure of how exposed it is to natural disasters. The value: you understand your risk before you agree to carry it.
• Too much in one place. See how much of your business sits in one suburb. The value: you know what one bad hailstorm could cost before it happens.

Protect money and work faster

These uses are less exciting, and they often pay back the fastest.

• Automatic underwriting. Underwriting means deciding whether to accept a risk, and at what price. Models can help with the routine cases. The value: faster and more consistent decisions, with people focused on the hard ones.
• Vehicle damage checks. When a car is hit, use photos to assess the damage and estimate the repair cost. The value: claims can move faster, and estimates become more consistent.
• Warning signs of fraud. Compare what a claim says with what the driving data shows. The value: investigators know where to look first, and a person still makes the decision.
• Money owed but not collected. Find the gap between what was charged and what came in. The value: you catch losses that nobody was looking for.

How to measure whether it worked

An accuracy score tells you whether a model is good. It does not tell you whether it was worth building. We judge each use case, including AI, by what it does for the business:

• Money. How much did this save?
• Customers. How many did we keep?
• Claims and risk. Have claims come down? Is our risk lower?
• Speed. How quickly can we respond to a crisis or a request?

AI tools need a few checks before they reach that test. These are my suggestions for anyone starting out:

• Keep a set of test questions with known answers, and run them on every change.
• Check that answers come from your own trusted data, and that the tool says so when it does not know.
• Have a person review anything that affects a customer or money.
• Track what each use case costs to run.

It changes the business itself

The most surprising result was not any single model.

Once people could see what the data showed, the business started asking bigger questions. Inefficiencies we flagged have prompted a redesign of the processes that were creating bad data in the first place, and they have changed how products are designed.

That is when a data team stops being a support function. It starts helping to shape what the business does next.

How to choose your first use cases

You do not need all of these. Starting with too many is a common way to fail. Ask four questions about each idea:

1. What decision will this change? If nobody will act differently, stop here.
2. What is it worth? Put it in money, time or customers, even roughly.
3. Can you trust the data? If not, you have found your next foundation job.
4. Who owns the result? Someone in the business must want this and use it.

Then pick one that is valuable and small enough to finish. A finished small win does more for you than a big plan, because it gives you something to show. It also adds to the foundation, which makes the next one cheaper.

You have arrived when

• A new business question takes weeks to answer, not months.
• Business teams come to you with ideas.
• New use cases reuse data and patterns that already exist.
• You can say what each use case is worth to the business.

Remember where this started: no handover, no map, and everything important inside someone's head. Everything on this list stands on the unglamorous work that came first. Build the road, and you can go to many more places.