Most advice about small business data analytics assumes you have a dashboard, an analyst, and a quiet afternoon. You have none of those. You have a delivery run that left forty minutes late, a wholesale account that stopped reordering in March, and a fuel card statement you’d rather not open.
The good news is that a business running its own local delivery already generates more useful data than a retail shop does. Every order carries a timestamp, an address, a basket size, and a driver’s route. That is a complete record of what your customers want, where they want it, and what it costs you to get it there. You don’t need to collect anything new. You need to look at five things a day and change one thing a week.
This guide covers the whole system: which numbers to capture, where they already live, and how to turn them into a decision instead of a spreadsheet nobody opens.
The Bottom Line
- Last-mile delivery now accounts for roughly 53% of total shipping cost, up from 41% in 2018, so delivery data is where the margin question actually gets answered.
- Five daily numbers are enough to start: orders, on-time rate, cost per drop, average order value, and reorder gap.
- The weekly review matters more than the tooling. One decision per week beats a dashboard you check once a quarter.
- Your point-of-sale system, your route history, and your bank feed already hold everything you need for the first three months.
Lower your delivery costs by 23%
How we reduce costs:
- No delivery vehicle expenses
- Optimized local routes
- Pay-per-delivery model
- Average 23% delivery cost reduction
What small business data analytics means when you deliver your own orders
Small business data analytics is the practice of using the records your business already produces (sales, orders, routes, costs) to answer a specific operating question and then change something because of the answer. That last clause is the part people skip.
For a bakery or a florist running local delivery, the questions are narrower and more useful than the generic version. Not “how are sales?” but “which delivery window makes us late?” Not “are customers happy?” but “how many weeks after a first order do people typically stop?” Those are answerable from your own records this week.
The cost side is what makes this urgent. Last-mile delivery has become the dominant line in shipping spend, and U.S. delivery costs rose roughly 12% between 2024 and 2025, with labour alone making up about half of last-mile expense and fuel another 10% to 25% (Net Zero Insights; SmartRoutes). When over half your shipping cost sits in the final leg, guessing about that leg is expensive. A related pattern shows up on the spend side too: the business spending habits that hold a company back are usually the ones nobody measures.
There’s a second reason to bother. Businesses that run on data rather than instinct report measurably better productivity and profitability than comparable peers, and the gap has widened as analytics tooling got cheap (Hydrogen BI). The advantage isn’t the software. It’s that a decision made against a number gets reviewed, and a decision made on a hunch doesn’t.
The five numbers to capture every day
Start here. These five take about six minutes to record and cover demand, service, cost, and value.
- Orders placed. Count only. Split by channel if you sell through more than one: phone, web, standing wholesale order.
- On-time delivery rate. Of yesterday’s drops, what share landed inside the promised window? If you promise a day rather than a window, count anything that arrived on the right day.
- Cost per drop. Total delivery cost for the day divided by the number of stops. Include driver time, fuel, and any courier fees. This is the number that moves most and gets watched least.
- Average order value. Total revenue divided by orders. Two weeks of this tells you whether a promotion actually made you money.
- Reorder gap. For every order placed yesterday, how long since that customer’s previous order? You’re building a distribution, not a single figure.
Six minutes, one row in a spreadsheet, every trading day. After three weeks you have something no benchmark report can give you: your own baseline. After three months you can see seasonality.
The reason to log the reorder gap daily rather than pull it quarterly is that it’s the early warning. Revenue looks fine right up until the month the repeat buyers don’t come back, and by then you’re looking at a hole you can’t fill with new customers. Cohort analysis shows exactly when customers stop reordering, and that post covers how to build the table and read it, which is the natural next step once you have two or three months of gaps recorded.
Where your delivery data already lives
You almost certainly don’t need a new system. Four sources cover the first year.
Your point-of-sale or order system. Every order record has a timestamp, a customer identifier, a basket, and usually a delivery address. Export it as CSV. This single file answers order volume, average order value, product mix and, with a pivot table, reorder gaps.
Your route history. Whether you use routing software, a driver’s phone, or a printed manifest, you have a record of what stops happened in what order and roughly when. This is where on-time rate and cost per drop come from. Route optimisation platforms that handle multi-stop planning keep this history automatically, which turns a manual count into an export.
Your bank feed and card statements. Fuel, vehicle maintenance, courier invoices, and subscriptions. Categorised properly, this is your real cost base, and it’s the only source that catches the spending you forgot you signed up for.
Your website or ordering page. Free tools such as Microsoft Clarity record session replays and heatmaps at no cost, which tells you where people abandon an order. Behavioral analytics for local delivery goes deep on which of these signals are worth watching and what each one tells you to change.
A useful example of how little you need: one local retailer combined nothing more exotic than point-of-sale records and local weather data, found the pattern connecting the two, and cut costs 18% during slow weeks by adjusting staffing and stock against it (Coupler.io). Two ordinary datasets, one question, one change.
The weekly routine that turns numbers into a decision
Daily capture is worthless without a weekly read. Block thirty minutes, same slot each week.
- Chart the week. Five numbers, seven days, one line each. Do this in a spreadsheet; charts are for spotting shape, not for presenting.
- Find the one anomaly. Not all of them. The single biggest deviation from your own baseline.
- Ask why once, then check. Cost per drop jumped Thursday. Was it a longer route, a smaller order count over the same mileage, or an expedited courier you paid for? The route history answers this in two minutes.
- Change one thing. Move a cut-off time, re-sequence a route, drop a delivery window, call three lapsed accounts. One change, written down, with the date.
- Check it next week. Did the number move? If not, revert and pick a different lever.
Thirty minutes a week is realistic precisely because it attaches to a routine you already run. If you don’t yet have a fixed operational rhythm to hang it on, the structure in this guide to daily restaurant operations management transfers directly to any business with a production and delivery cycle.
This is the whole method. It works because a single change against a recorded baseline gives you a clean read, whereas five simultaneous changes give you noise. Businesses with a strong data habit make decisions faster than their peers not because they process more information, but because they’ve already agreed what they’re looking at.
What your customers’ buying patterns change about the plan
Analytics tells you what happened. Knowing who your buyers are tells you which changes are worth testing. Two generational shifts matter most to a local business that delivers, and both come down to expectations about speed and price.
Millennial households now shop 683 times a year at an average of $33 a trip, and the great majority buy both online and in store rather than choosing one (Numerator). That omnichannel default is why a clumsy handoff between your ordering page and your delivery promise costs real orders. The detail is in millennial spending habits local businesses should plan for.
Gen Z shops differently again. Discovery happens on social platforms, price comparison is reflexive, and roughly 80% of them shop online partly because comparing prices is so easy (GWI). If your analytics show a high browse-to-order drop-off among newer customers, that’s usually a price-visibility or delivery-fee problem rather than a product problem. Gen Z shopping habits worth watching closely covers what to do about it.
The practical instruction: segment your reorder gap by when the customer first ordered. If newer cohorts lapse faster than older ones, something in the current experience has changed, and it’s usually delivery cost or delivery speed.
Analytics tools small businesses actually use
You can run the whole system above in a spreadsheet, and for the first six months you probably should. The discipline is the product; the software is a convenience.
When you outgrow it, the practical options for a small operation are the free or low-cost tier of general BI tools such as Zoho Analytics, Google Looker Studio or Tableau’s entry offerings, connected to the CSV exports you’re already producing. Microsoft Clarity covers on-site behaviour for free. Your routing platform likely already reports on-time rate and stops per hour without extra configuration.
The AI layer is the newer question, and it’s worth understanding before you pay for it. Most small operations get more from a clean weekly review than from automated insight generation, but the direction of the category is real. This overview of AI for business intelligence is a reasonable primer on what these tools now do and where they still need a human to ask the question.
Two warnings from the pattern of businesses that stall here. First, a tool that requires a weekly import you won’t do is worse than the spreadsheet you will do. Second, buying an analytics subscription before you’ve defined the five numbers just adds a line to the very cost base you were trying to examine.
What to do in your first month
Week one: start the daily log. Five numbers, nothing else, no analysis.
Week two: keep logging, run the first thirty-minute review, resist the urge to change anything yet. You’re still learning your own baseline.
Week three: make your first single change and record it.
Week four: check whether it moved. Export three months of order history and build your first cohort table.
By the end of month one you’ll have a habit, a baseline, and one verified improvement. That’s a better position than most businesses reach with a dashboard, because the dashboard tells you what happened and the habit tells you what to do next. Do this for a year and the improvements compound, not because any single change is large, but because you stop repeating the expensive mistakes you previously couldn’t see.
Frequently asked questions
What data should a small business track first?
Start with orders placed, on-time delivery rate, cost per drop, average order value, and the gap since each customer’s previous order. These five cover demand, service quality, cost, and retention. Everything else is refinement, and none of it matters until you have a baseline for these.
Do I need analytics software to do this?
No. A spreadsheet with five columns handles the first several months comfortably. Software becomes worthwhile when the manual export takes longer than the analysis, which for most local operations is somewhere past the first year.
How much data do I need before the numbers mean anything?
Three weeks gives you a rough baseline, three months gives you a reliable one, and a full year gives you seasonality. You can start acting on obvious anomalies from week two, but hold off on structural changes until you can distinguish a bad Tuesday from a trend.
What’s the most common mistake?
Measuring everything and changing nothing. The second most common is changing several things at once, which makes the result unreadable. One change per week, checked against a recorded baseline, beats a full dashboard nobody acts on.
