A bad sample returns clean numbers.
That is what makes it dangerous.
Ask the wrong people the wrong way and you still get a formatted deck with confident percentages in it. Nothing about the output tells you it is wrong — which is why market research is the discipline most capable of producing an expensive decision built on nothing.
The sample decides the answer before the question is written.
This is the failure that produces the most confident wrong conclusions, and it is almost always accidental — the easiest people to survey are your existing customers, who are by definition the ones who already said yes.
And demand research now has a second surface. Homeowners increasingly ask an AI assistant who to call rather than typing a search, which means the question is no longer only what people search for — it is what ChatGPT, Google AI Overviews, Perplexity and Gemini say when they are asked about your category in your market. We research both, because a market where AI platforms already recommend three competitors is a different opportunity from one where nobody has been named yet. That is what running an AI-first agency means in practice: the research changes, not just the reporting.
Take a real example. A home services business wants to know whether customers would pay a premium for a same-week appointment, so it surveys its customer list. The answer comes back strongly positive. Every person asked had already chosen this company, so the sample structurally cannot contain the people who called, heard the lead time, and hung up — who are precisely the people the question was about.
The result is not a small error. It is an answer to a different question: how do our existing customers feel about a service they already buy. That is worth knowing and it is not what the deck says on the cover, and nothing in the formatting distinguishes the two.
This is not an obscure concern. Federal statistical agencies publish standards for exactly these mechanics — the U.S. Office of Management and Budget's statistical policy directive on survey standards covers coverage, instrument design and nonresponse as first-class methodological requirements. Almost no commercial market research cites anything like it.
The third failure is the quietest. Nonresponse is treated as noise, when the people who did not answer are usually different from the people who did in ways that matter — busier, less satisfied, or not customers at all. A study with a low response rate is not a smaller study; it is a study of whoever felt like answering. That is why we start from evidence about people who already acted, covered in conversion rate optimization.
A leading question still returns clean data.
Four wording faults produce most of the damage, and every one of them is invisible in the output. The chart looks the same either way, which is exactly the problem.
Four faults we find constantly
- The presumed answer. "How much do you value our fast response?" assumes it is valued and asks only about degree.
- Two questions in one. "Was our team professional and on time?" cannot be answered by someone whose technician was one but not the other.
- Unbalanced options. Four positive choices and one negative one produce a positive average by construction.
- Your vocabulary, not theirs. Asking about a term customers never use returns confusion recorded as opinion.
- Order effects. A question that follows a positive one inherits some of its sentiment.
- No option to disengage. Forcing an answer from someone with no view manufactures one.
What we change
- Ask whether before how much. Establish that a thing matters before measuring the degree to which it does.
- One idea per question. Split anything containing "and" into the two questions it actually is.
- Symmetric options. Equal weight either side of a neutral midpoint, with an honest way to opt out.
- Their words, taken from your own records. Language pulled from calls and reviews, not from your service menu.
- Order tested. Sensitive or evaluative questions placed where they cannot be primed.
- Piloted before it runs. Read aloud to a handful of real customers first, which catches most of the above.
Each one needs a different method. Most projects use one for all three.
Market research fails as often from method mismatch as from bad sampling. These are the three questions businesses actually need answered, and they are not answered the same way.
How many are there?
Household counts, income distribution, housing age, business establishment counts and industry employment are all published. The American Community Survey and Bureau of Labor Statistics data answer most sizing questions outright.
- How many households in this service area could plausibly buy
- How the area differs from the one you already serve
- How many competing establishments operate there, per the County Business Patterns program
- Whether new construction is rising, via building permits data
What do they want?
A survey asks people to predict their own behavior. Your call recordings, lost-quote reasons and site search queries record what they actually did and said when money was involved — which is a different and better class of evidence.
- The exact words customers use, which belong in your copy — see content writing
- The objection that appears before every purchase
- The service people search your site for and you do not offer
- The question asked so often it should be a page, per content strategy
Would they actually buy?
People are poor predictors of their own future purchases, and they are polite. "Would you be interested in this?" reliably overstates demand, because agreeing costs nothing. Stated intent is a hypothesis, not a finding.
- Put a real offer in front of real traffic and count who acts
- Measure a booking or a deposit, not a raised hand
- Conversions defined per Google's documentation
- Test design and readiness sit under CRO
Six public sources that answer questions people pay to have estimated.
All of it is collected by federal statistical agencies under published methodology, all of it is free, and most of it is more rigorous than anything a marketing budget would buy.
Where to look first
The order matters. Start here, exhaust it, and only then commission primary research for the questions that genuinely remain — which is usually fewer than expected.
- Households and population by area. The American Community Survey publishes detailed demographic and housing characteristics down to small geographies, updated annually.
- Income and housing characteristics. Same source — including housing age and ownership, which for most home services businesses is a better demand proxy than income alone.
- Industry employment and wages. Bureau of Labor Statistics data covers employment, wages and occupational detail by area and industry.
- Business counts by area. County Business Patterns reports establishment counts and employment by industry and geography — a supply-side check on a market you are considering.
- Residential construction activity. Building permits data is a leading indicator for any trade whose demand follows new housing.
- Consumer spending categories. The Consumer Expenditure Surveys report what households actually spend across categories.
- The methodology behind all of it. Federal surveys operate under published statistical standards — you can read exactly how the numbers were produced, which is rarely true of a commercial panel.
Six rules every research project runs to — and why each one exists.
Three protect the method, two protect the conclusion, and one protects you from paying for something you already have.
Public data first
The sample is stated, always
Who was asked, how they were reached, how many did not respond — reported alongside every finding, not in an appendix.
Why: coverage and nonresponse are first-class concerns in federal survey standards. A finding without its sample is a number without a meaning.
Questions piloted before they run
Read aloud to a handful of real customers first, and rewritten wherever someone hesitates or asks what it means.
Why: it catches leading wording, double-barrelled questions and internal jargon in an afternoon, before the instrument reaches hundreds of people.
Stated intent is labeled as intent
"Would you buy this" produces a hypothesis. It is reported as one, never as demand.
Why: agreeing to a survey question costs nothing and people are polite. The only real test is a live offer measured in conversions.
Your own records read first
Calls, lost quotes, site searches, form drop-off and cancellation reasons, before any new data is collected.
Why: it is behavior rather than opinion, it is free, and it comes from people who already tried to buy. Storage and retrieval sit under CRM.
A finding that stops the project counts
Research that can only confirm the plan is not research, and we will say so before taking the work.
Why: the FTC's advertising guidance requires objective claims to be substantiated — including claims you make on the strength of research you paid for.
We will tell you the study is not worth commissioning.
Most market research questions we are asked have already been answered — by a federal statistical agency, at no cost, with published methodology, or by records the business already owns and has never read. Saying so ends a paid engagement before it starts.
We say it anyway, because the alternative is selling a study whose findings we already know. A survey commissioned to confirm something is not research, and a research bill for a question the Census answers for free is a fee with a deck attached.
When primary research genuinely is warranted, every finding arrives with its sample stated in the same breath — who was asked, how they were reached, and how many did not respond. A percentage without that context is a number without a meaning, and we will not put one in front of you.
What separates this from a standard research engagement.
Not a competitor teardown — a description of where the practices differ, so the choice can be made on substance.
| Practice | Standard research engagement | AllegiantOMNIVIZ™ |
|---|---|---|
| First step | Design a survey | Check whether public data already answers it |
| Your own records | Not examined | Calls, lost quotes and site searches read first |
| Sample disclosure | In an appendix, if at all | Stated beside every finding |
| Who gets surveyed | The customer list | Including people who chose otherwise or bought nothing |
| Question design | Written and sent | Piloted aloud with real customers, then rewritten |
| Stated intent | Reported as demand | Reported as a hypothesis, tested with a real offer |
| Unnecessary studies | Sold | Declined, with the free source named |
Four research line items you can stop paying for.
Two are already available free. One measures the wrong thing. The fourth is a deck priced as a decision.
Market sizing that repackages public data. Household counts, income distribution, housing characteristics, industry employment and business counts by area are published by the Census Bureau and the Bureau of Labor Statistics at no cost. Reformatting it into slides is a service. Charging for it as research is not.
Customer surveys sent only to customers. Useful for satisfaction, structurally incapable of answering a demand question — the people who did not buy are the ones the question is about, and they are the ones the sample excludes by construction.
Purchase intent reported as demand. Stated intent is a hypothesis. It costs a respondent nothing to agree, and it reliably overstates what people will actually do when money is involved. The honest version puts a real offer in front of real traffic and counts conversions.
Persona documents with no research behind them. A named fictional buyer with a stock photograph, invented demographics and a list of pain points somebody imagined in a workshop. It looks like research, it is used like research, and nothing in it came from a customer. If personas are built at all, every attribute traces to a record or a source.
The pattern beneath all four: they produce an artifact that resembles evidence. The deck is the deliverable, and the deck looks identical whether or not anybody learned anything.
- U.S. Census Bureau — American Community Survey
- U.S. Census Bureau — County Business Patterns
- U.S. Census Bureau — Building Permits Survey
- U.S. Census Bureau — Data
- Bureau of Labor Statistics — Data Tools
- Bureau of Labor Statistics — Consumer Expenditure Surveys
- OMB — Standards and Guidelines for Statistical Surveys
- FTC — Advertising and Marketing Business Guidance
- Google Analytics Help — About conversions
- Google Search Central — Creating helpful, people-first content
- Google Search Central — Evaluating third-party SEO advice
- Google Search Central — Spam policies for Google web search
The questions operators actually ask.
Answered against primary sources where they exist, and answered honestly where they do not.
Do we need to commission a study to size a new market?
Usually not. Household counts, income distribution, housing age and ownership are published in the American Community Survey down to small geographies, and competing establishment counts are in County Business Patterns. Between them they answer most sizing questions at no cost, with methodology you can read. Commission primary research only for what genuinely remains after that — which is usually about how buyers decide, not how many there are. Related: competitor analysis.
Why can't we just survey our customer list?
You can, for satisfaction. For a demand question it is structurally incapable of answering, because everyone on that list already chose you — the people the question is about are precisely the ones the sample excludes. Federal survey standards treat coverage as a first-class methodological concern for exactly this reason. The fix is including people who requested a quote and chose a competitor, and people who enquired and bought nothing. Their records are in your CRM.
How many responses do we need for the results to be valid?
Fewer than most people assume, and the number matters less than who they are. A large sample drawn entirely from one group is worse than a small sample that includes the people you are trying to understand. What matters as much is nonresponse: the people who did not answer are rarely a random subset, so a low response rate does not produce a smaller study, it produces a study of whoever felt like responding. Federal statistical standards address both. See CRO on sample size for tests.
Does asking whether people would buy something tell us if they will?
No, and treating it as though it does is how products launch into demand that was never there. Agreeing to a survey question costs a respondent nothing and people are inclined to be encouraging, so stated intent consistently overstates behavior. It is a hypothesis worth having and it should be labeled as one. The real test is putting an actual offer in front of actual traffic and counting who acts, measured as conversions rather than as interest. Related: landing pages.
What research can we do that costs nothing?
The most valuable kind. Read your recorded inbound calls for the question customers ask every week. Read your lost-quote reasons — those are buyers who evaluated you and chose otherwise, which is the population a survey struggles to reach. Read your site's internal search queries for services people expected you to offer. Read where people abandon your forms. It is behavior rather than opinion, it comes from people who already tried to buy, and the language belongs in your copy per content writing. Google's people-first guidance is the same instinct.
Are buyer personas worth building?
Only if every attribute traces to a record or a published source. The common version is a named fictional person with a stock photograph, invented demographics and pain points somebody imagined in a workshop — it looks like research, it gets used like research, and nothing in it came from a customer. A defensible version cites where each attribute came from: a demographic from the ACS, an objection from your call recordings, a decision criterion from lost-quote notes. Same sourcing discipline as content strategy.
How is this different from competitor analysis?
Market research studies demand, competitor analysis studies supply. This page is about the buyers — who wants the service, how many of them exist in an area, what they call it and how they decide. Competitor analysis is about who else is already answering that demand and how well they do it. They inform each other and neither substitutes for the other: a market can be large and badly served, or well served and too small. Google's guidance on evaluating claims applies to research vendors as much as to SEO ones.
Can AI do our market research?
It will summarize published material quickly, and it is genuinely useful for finding which federal dataset covers a question. What it cannot do is collect primary evidence about your specific buyers — it was not on your calls, it has not read your lost-quote notes, and it cannot survey anyone. It will also produce confident-sounding figures with no traceable source if asked to, which is the failure mode this page exists to warn about. Generating volumes of derived content from it carries the risks Google names in its spam policies. See AI SEO.
How do we know whether a new service area is worth entering?
Three public checks before anything else. Household counts and housing characteristics from the ACS tell you the size and type of the demand base. County Business Patterns tells you how many firms already serve it. Building permits data tells you whether the area is growing, which matters for any trade whose demand follows new housing. Then the operational question nobody researches: whether you can actually serve it at your response time. Related: local SEO.
What makes a research finding actually usable?
That it could have come out the other way. A study designed so that every possible result supports the plan is a document, not research, and it is worth stopping before it runs. Beyond that: the sample is stated beside the finding rather than buried, the question wording is available for you to read, nonresponse is reported, and the conclusion distinguishes what was measured from what was inferred. The FTC's advertising guidance matters here too, since claims you make on the strength of research need to be substantiated by it. Related: reputation management.
Find out what your own records already say about demand.
A demand review: your inbound calls, lost-quote reasons, site search queries and form drop-off read for what buyers actually asked for, plus the public data that sizes your market at no cost. If a study is not warranted, we will tell you that instead. Findings are yours whether or not we work together.
- Call recordings read for the recurring question and the recurring objection
- Lost-quote reasons analyzed — buyers who evaluated you and chose otherwise
- Site search queries, including services people expected and you do not offer
- Form drop-off points identified by step
- Market sizing from public federal data, with the sources named
- A plain statement of which questions genuinely need primary research
Explore the wider program: all services, competitor analysis, content marketing, conversion rate optimization, local SEO and the A.R.C. Report.
We will read what your own records already say about what buyers want, and tell you honestly whether a study is worth commissioning.
No cost, no commitment. We will follow up by email or phone to walk you through the findings.
