Market Research With No Budget: Start From the Decision

Primary versus secondary. Qualitative versus quantitative. The taxonomy will not help you decide anything on Monday.

The skill is knowing which question blocks a decision and buying the answer at the lowest honest price. Where a sub-discipline takes over we hand off: competitor analysis is a method of its own, and the cheapest to run first.

Start with the decision, not the method

Write the decision as a sentence, then what you would do under each possible outcome. If every outcome leads to the same action, the research is a hobby. We call it the "what would change my mind" test.

1. Write the decision. Not "understand the market" — "do we launch the ₹2,999 plan or the ₹6,999 plan first?"
2. Write the plausible outcomes — the two or three answers the research could return.
3. Write the action under each, one line, decided in advance.
4. Check for a tie. Identical actions across outcomes means stop and go build.
5. Set the threshold. "If fewer than four of twelve name price, we stop treating price as the problem."

A threshold set in advance cannot be moved later to suit the answer you wanted. It also sizes the effort: a reversible decision needs one conversation, not a study.

The four questions a small business actually needs answered

Nearly every brief collapses into four questions, each with a cheapest credible method: the least effort that still gives an answer you would bet on.

Question two is where customer segmentation begins, and STP marketing carries the targeting logic from there. Question three is a value proposition question in a research costume.

Where to find real data without paying for it

Five free sources cover most of what a small business needs.

• Search demand data. Google Trends reports relative interest, not volume: each point is divided by total searches for that geography and time range, then scaled 0–100. It answers "rising or falling here", never "how many". Search Console's Performance report shows which queries bring traffic to your site.
• Marketplace and app-store reviews. Unpaid depth interviews that already happened; read the three-star ones first.
• Google Business Profile reviews and Q&A. Questions on a local profile are objections your site failed to answer. You cannot buy the sample: offering free or discounted goods for reviews is "considered fake engagement and is strictly prohibited".
• Government and public statistics. Better in India than most markets; the tables are below.
• Your own sales calls and support tickets. What people say in public is a performance; what they type into a ticket at 11pm is not. The public layer is social listening.

Platform sources, none showing a last-updated date, read 1 October 2026: Google Trends Help, "FAQ about Google Trends data"; Search Console Help, "Performance report (Search results)"; Google Business Profile Help, "Tips to get more reviews".

Talking to customers properly

The highest-value method available with no budget, and the most skipped. Twenty minutes with a recent buyer beats a month of desk research.

How many conversations is enough

Guest, Bunce and Johnson analysed 60 in-depth interviews and found saturation — the point where new interviews stop producing new codes — within the first twelve, with 92% of all codes in one country's dataset present by then (Field Methods, Vol. 18, No. 1, 2006, pp. 59–82).

The caveat is theirs: that was a narrowly defined, homogeneous group. So twelve is twelve per segment, not in total. Three is an anecdote with a spreadsheet around it.

How to ask without leading

• Ask about the last time, not general preference. "Walk me through how you bought the last one" beats "what do you look for?"
• Never name your solution first, or you have told them the answer you want.
• Ask about money already spent — what they bought, what it cost, what else they considered.
• Follow a complaint with "what did you do about it?" A problem nobody acted on is not worth building for.
• Avoid anything containing "would". Then stay silent: the second thing they say is usually truer.

You are listening for problems, not validating an idea. That output becomes a buyer persona.

A worked research plan with a ₹0 cash budget

Three weeks of part-time attention, no research spend. Every row ends in a decision; one that does not is a row to delete.

Reading that last row is a measurement job: our marketing analytics guide covers the tracking to fix before the test.

How research goes wrong

Five failures account for most bad research, and four feel like good research while you do them.

1. Leading questions

"How important is fast delivery to you?" scores high with anyone, because the question has already said delivery matters. The unled version is "what nearly stopped you buying?" The tell: your findings agree with the brief.

2. The say-do gap

People are not lying; they do not know what they will do. Sheeran and Webb conclude that "intentions get translated into action approximately one-half of the time", and that experimentally raising intention by a medium-to-large amount produced only a small-to-medium change in behaviour (d+ = .36) (Social and Personality Psychology Compass, Vol. 10, No. 9, 2016, pp. 503–518). The tell: evidence made of future-tense sentences.

3. Surveying your happiest customers

The people who answer your survey are the people who like you enough to answer it. The expensive truth sits with enquirers who did not buy, one-time buyers, and people who chose a competitor. The tell: every respondent came from your mailing list.

4. Sample sizes too small to mean anything

For qualitative work, use the saturation basis above. For a percentage, the arithmetic is not optional: the normal-approximation formula n = z² × p(1−p) ÷ e², at 95% confidence (z = 1.96), p = 0.5 and a ±5-point margin, gives n = 385 for a large population.

Pew Research Center puts the average margin of error at about ±1.3 percentage points for an unweighted sample of 2,000, and is explicit that this sits "in addition to any systematic biases caused by noncoverage, nonresponse or self-selection" ("Variability of survey estimates", 26 January 2018). A clean margin of error on a crooked sample is a precise wrong answer. The tell: a percentage from fewer than 100 responses.

5. Confirmation bias in the write-up

The data was fine; the summary quietly promoted the four quotes that agreed and demoted the three that did not. Count every mention rather than remembering the vivid ones, and hand the notes to someone else. The tell: no section titled "what argued against this".

Judging a secondary figure before you quote it

Secondary data is free and mostly fine. Market-size projections are the exception, and the figures people quote most. Six questions for any published number:

• Who collected it, and do they sell something the number flatters?
• Actual or projected? A 2030 figure is an opinion about 2030.
• What was the sample, and how was it drawn? If unstated, it is not a measurement.
• Is the methodology public, not just the press release?
• Does the primary source say what the secondary source claims?
• Corroborated independently, or is everyone citing one release?

Bluntly: most "India's X market will reach $Y billion by 2030" lines in circulation are unsourced vendor projections with no published sample, no stated method and no base-year citation. We will not reproduce one here even as an example, because repeating it is how it survives.

Compare a quotable figure. MoSPI's Household Consumption Expenditure Survey 2023-24 press note (27 December 2024) gives its field period, August 2023 to July 2024; its sample, 2,61,953 households; and monthly per capita consumption expenditure of ₹4,122 rural against ₹6,996 urban. Anything short of that goes in your notes, not your deck.

Doing this in India

The free data here is better than you think, and your home city is not the country.

• MoSPI — the Household Consumption Expenditure Survey for spending patterns; the Annual Survey of Unincorporated Sector Enterprises if you sell to small businesses. ASUSE 2023-24 (29 January 2025) counted 7.34 crore unincorporated non-agricultural establishments, up 12.84%.
• data.gov.in — the Open Government Data platform; its metrics page reported over five lakh resources from central ministries and state departments, stamped "23 Dec, 2025".
• Census of India for population, household and language structure; TRAI's monthly releases for anything depending on connectivity.

The limits of English-only online research

If your survey, review mining and keyword research are all in English, you are sampling one slice and calling it the market.

Table C-16 of the 2011 Census (Office of the Registrar General, Paper 1 of 2018) records 19,569 raw mother-tongue returns, rationalised into 1,369 mother tongues and grouped into 121 languages. Hindi is the mother tongue of 43.63% of the population, Bengali 8.03%, Marathi 6.86%, Telugu 6.70%, Tamil 5.70%.

The fix is cheap: hold conversations in the language the customer buys in, read reviews in the script they were written in, and check a non-English variant of every keyword you target.

Why a Mumbai sample does not describe a tier-2 market

Spending, connectivity and the competitive set all differ, and only the first is obvious.

HCES puts urban MPCE at ₹6,996 against ₹4,122 rural. TRAI's July 2026 data (press release 116/2026, 28 August 2026) puts urban wireless tele-density at 145.84% against 60.48% rural. And ASUSE 2023-24 reported internet use for business purposes at 26.70% of unincorporated establishments: most did not report using it for business at all, so an online-only channel is not sampling them.

Treat any finding from your home city as a hypothesis about everywhere else.

Where to go from here

Pick the decision that is stuck, write what you would do under each outcome, and choose the cheapest method that could move it. Desk research and twelve conversations fit in three weeks; the search and content work built on them typically takes three to six months.