Ask a room of business owners what data science could do for them and you'll hear about AI assistants and predictive everything. Ask what would actually move the needle this quarter, and the answer is usually one of two unglamorous classics: knowing what's coming, and knowing your customers.

Forecasting: knowing what's coming

Demand forecasting is the workhorse of applied data science. Take your own sales history, account for seasonality and trends, and produce a forward view you can plan against — stock, rosters, cash flow.

It doesn't need to be perfect to be valuable. A forecast that's 85% accurate beats a gut feel that's 60% accurate, every week, compounding. For a trades business we might forecast job volume by month to plan hiring; for a retailer, units by product line to stop both stockouts and dead stock; for a services firm, pipeline conversion to steady out the feast-and-famine cycle.

The methods are mature — exponential smoothing, seasonal decomposition, gradient-boosted models when there's enough history. The hard part isn't the maths. It's the discipline of keeping the data clean and reviewing the forecast against reality every month so it keeps earning trust.

Segmentation: knowing your customers

The second classic is segmentation — grouping customers by how they actually behave rather than how the org chart imagines them. The simplest credible version is RFM: how recently they bought, how frequently, and how much (monetary).

Run that on a year of invoices and patterns appear that nobody had words for:

• The quiet 15% who generate half the margin and never complain — worth a phone call, not a newsletter.
• The once-big accounts drifting away slowly enough that nobody noticed.
• The small, frequent buyers who'd happily consolidate more spend with you if asked.

Each group justifies a different action. That's the test of useful segmentation: not prettier reporting, but a different decision for a different group.

Why this comes before the shiny stuff

There's a sequencing argument here, not just a value one. Forecasting and segmentation force the groundwork — clean sales data, consistent customer records, a habit of checking predictions against outcomes — that every later AI project depends on. Skip it and you're automating on sand.

The best first data science project is rarely the most exciting one. It's the one that makes next month less of a surprise.

Start with what's coming and who's buying. The fancy stuff stands on those two answers.