Market analysis tools fail the moment data stops being traceable

Market analysis software changes every year; the failure modes do not. What follows is how to judge anything in the category, and which named sources and products sit inside it.

The category sells two different things under one heading. One is access to data. The other is a way of drawing conclusions from data. They fail differently, and a tool that is excellent at the first can be actively dangerous at the second.

What to take away

What the category contains

Named, the category looks like this. Public data portals are the Census Bureau, the Bureau of Labor Statistics and the Bureau of Economic Analysis. Commercial data aggregators are Data Axle for business lists, Claritas for demographics and segmentation, and Lightcast for labor market data. Mapping and trade area tools are Esri, whose business analyst products layer public counts onto maps, and Placer.ai, which estimates foot traffic from mobile location data. Industry report publishers are IBISWorld and Statista. Most of what the second group sells is modeled down from a national sample, which is the point to carry into any comparison with the first.

The sizing calculators deserve their own warning. A calculator that starts from an industry total and applies a percentage produces the exact figure that the fundamentals mistakes list warns about, faster than you could produce it by hand.

The last row is the one to be most careful with, because the output looks like a measurement of everybody and is a measurement of whoever was in the sample.

Traceability is the whole test

For any figure a tool gives you, you should be able to answer four questions without contacting support.

What is the source dataset, by name, in the form you could cite it in the financial section — County Business Patterns for establishment counts, the Quarterly Census of Employment and Wages for jobs and pay, the American Community Survey for population and income. What period does it cover, and when was it last updated. What geography does it use, in the geography's own terms rather than a marketing label like "your local area". Is the value counted, sampled, or modeled from something else.

If a tool cannot answer all four for a given number, that number cannot go in your plan.

It is not necessarily wrong, but you will be asked where it came from and "the software said so" ends the conversation badly.

Every figure in a plan is a statement you are making. A reviewer testing one figure is really testing whether you know the difference between a source and a screen.

Modeled is not measured

Most detailed local data is produced by taking something measured at a larger scale and distributing it downward using a model. That is legitimate and often the best available, and it means the number carries an uncertainty that the display almost never shows.

Two habits keep you out of trouble. When you quote a number, say whether it is counted, sampled, or estimated. If an estimate has a published margin of error, quote it.

A plan that calls a figure an estimate with a stated range reads as competent. A plan that quotes an estimate to the nearest unit reads as someone who does not know what they hold.

Test it on a place you know

This is the evaluation that works, and it takes an evening.

Choose a street, a town or a trade area you know from experience. Then ask the tool the questions you already know the answers to. How many businesses of a given type are here. What does the population look like. How busy is this location on a Saturday.

Compare with what you know. You are not checking whether the tool is right; you are learning how it is wrong, which is more useful.

Does it miss home-based operators? Does it count a chain's head office as a local business? Does its trade area include the far side of a river with no crossing? Does it place a business at an old address?

Every data product has a characteristic blind spot. Finding yours before you rely on it is the difference between a tool and a trap.

Practical requirements before you pay

Start with what is free

Public statistical agencies publish most of what a small business plan needs, with methodology attached and no license worry. The Bureau of Labor Statistics' employment and wages by occupation series and the Bureau of Economic Analysis' output by industry tables are the two most plans ever need. Add the Census Bureau's County Business Patterns, which counts establishments by NAICS industry code down to the ZIP code, and the American Community Survey for population and income, and most local questions are answered without a subscription.

Read the coverage note before you quote a count. County Business Patterns covers establishments with paid employees, so a market made of owner-operators can look thinner than it is. The Quarterly Census of Employment and Wages covers jobs reported under unemployment insurance, which drops the same one-person operations out again. The American Community Survey is a sample, so every figure it gives arrives with a margin of error the plan should carry.

Sources worth knowing are listed in the market analysis guide. Paid products buy convenience and breadth, not authority.

One input no software sells is a morning spent asking people you intend to sell to whether they would buy. That yields evidence of demand, but a dataset yields evidence of population.

A lender assesses only one of those, and the general questions to ask any planning vendor before subscribing are in the tooling criteria, and they apply here unchanged.

Common questions

Is a purchased industry report worth it?

Sometimes, for structure and vocabulary, and for knowing which questions the industry considers standard. Rarely for a number you can defend at local level, because the local breakdown is usually modeled from a national sample.

Can I cite a tool as the source in my plan?

Cite the underlying dataset, with the period and geography, and name the tool only as how you accessed it. A reviewer wants the source, not your interface.

What about tools that generate the market section as text?

They produce prose with figures in it, and the figures are the problem. Anything auto-written into a market section arrives without the four traceability answers, which is precisely what a reviewer probes first.

How current does my data need to be?

Current enough that the reader believes it describes today, and dated so they can judge for themselves. Always print the vintage next to the figure. An older number labeled honestly is safer than a recent one with no date on it.