How many responses do we actually need?

For a large population, roughly 385 completed responses gives you 95 percent confidence with a 5 percent margin of error. For a small finite list — say around 800 customers — the requirement drops to about 260. Those are the arithmetic floors. The frame matters more than the count, because 400 responses from the wrong people is worse than 150 from the right ones.

What if we do not reach that number?

Then the report says so and adjusts what it claims. Small samples are still genuinely useful for finding themes, generating hypotheses and ruling things in for further work. What they cannot do is support a percentage stated about your whole customer base, and we will not write one.

Can you analyse a survey we have already run?

Yes, and that is the cheaper end of this work. The limit is honest — wording problems, a biased frame and missing answer options cannot be repaired after fielding. We will clean it, analyse it properly, and tell you plainly which of your questions can carry a conclusion and which cannot.

Should we offer an incentive?

It reliably lifts response rate and it changes who responds, and both effects need accounting for. An incentive pulls in people motivated by the reward rather than the topic, which shifts your mix. Whether that trade is worth making depends on how badly your natural response rate skews, and we work that out before fielding rather than after.

We need to hear from people who are not our customers. Is that possible?

Yes, through a research panel such as Prolific or Respondi. Panel responses are bought per completed interview and that cost sits outside this fee as a separate line, because it scales with sample size and screening strictness rather than with our work.

What if the findings contradict what we expect?

They get reported as found. Agreeing in advance what the answer will be is the one thing that makes a survey worthless, so the finding stands and the write-up explains the evidence behind it. Where the data is genuinely ambiguous, the report says that too rather than picking the more comfortable reading.

Data & Analytics

Survey and market research analysis

Asking the right people the right questions, then reporting what the answers can and cannot support — including who never replied.

From: $500–$6,000
Typical time: 2–6 weeks
Usually bought by: Product, marketing

What this actually solves

Your analytics can tell you what people did. It cannot tell you why they did it. It will show you precisely how many people abandoned the checkout, on which step, on which device, and it will never once explain whether they left because of the shipping cost, because the card form rejected them twice, or because they were only ever browsing. Behavioural data records outcomes. Reasons have to be asked for, and asking is a different discipline with different failure modes.

The other case is starker — you are about to do something you have no history for. A new product, a new market, a price change, a service you have never offered. There is no existing data to interrogate because the thing has not happened yet, and the only route to evidence is to go and ask people. That is where a badly built survey does real damage, because it produces numbers, and numbers get quoted in a board pack long after the caveats have been forgotten.

Which leads to the single most important thing about this work, and the thing that gets skipped most often. Who you ask determines the answer more than what you ask. A company surveys its active customer list, finds that 91 percent are satisfied, and concludes retention is healthy. The people who left in frustration were never on that list. The survey asked the people who stayed whether they had stayed. Every project here begins by naming who is structurally absent from the sample and what their absence would do to the conclusion.

What we do, and in what order

Fix the research question and the decision behind it. Not the topic — the decision. What will you do differently depending on the answer? If nothing changes either way, the survey should not be run, and that is a cheaper conversation to have now than after fielding. Two to three days.
Define the frame and calculate the sample. Who gets asked, drawn from what list, and who is missing from that list. Then the arithmetic — around 385 completes for 95 percent confidence and a 5 percent margin of error on a large population, around 260 where the population is a finite list of roughly 800. One to two days.
Write the questionnaire and stress-test every item. Each question is checked for leading wording, for two questions hiding inside one, for answer options that overlap or fail to cover the field, and for scales that quietly reverse direction halfway through. Order effects are considered, because an early question primes the ones after it. Length is capped deliberately, since response rate falls as the form grows. Three to five days.
Field it. Built in Google Forms, Typeform, SurveyMonkey or Qualtrics depending on how it needs to look and how the data has to come out. Response rate is the binding constraint on email fielding, so send timing and a planned reminder sequence do more for the sample than any other lever. One to three weeks, depending on reminders.
Clean, then analyse. Partial completions, straight-liners who select the same point down every scale, and speeders who finished faster than the questions can be read all come out before anything is counted, and the number removed is reported. Then frequencies, cross-tabs by the segments that matter, significance testing only where a comparison genuinely warrants it, and open-ended responses coded into themes. Three to five days.
Write the report. Findings, the evidence under each one, the recommended action, and a limits section that states plainly who is not represented and what therefore cannot be concluded. Two to four days.

Analysis runs in Google Sheets for smaller sets and in Python with pandas, or R, once the cross-tabs and testing get heavier.

What moves the price

The range is $500–$6,000 over 2–6 weeks, and it is set by four things more than anything else.

Design and field, or analysis only. Taking an existing dataset and reporting it properly is the light end. Designing the instrument, running the fieldwork and chasing the response rate is a different scope.

Where the respondents come from. Your own customer list costs nothing to reach. A research panel is bought per completed interview and gets more expensive the tighter the screening criteria, and that cost is quoted separately from the work.

How many segments have to be compared. Every additional cut — by region, tenure, age band, product line — multiplies both the sample you need and the analysis, because a segment with thirty responses in it cannot carry a claim on its own.

What usually goes wrong

A large claim built on a small base. Forty-two people responded, twenty-seven of them mentioned price, and the report says 64 percent of customers think the pricing is too high. That sentence is false. The true sentence is longer and less quotable — twenty-seven of the forty-two who responded raised price, which indicates it is worth investigating, but a sample that size cannot be projected onto the whole customer base. The longer sentence is what your credibility is made of, and it is the one that goes in the report.

The reachable list is not the population in question. Asking current customers why people churn is the classic version, and it is undefeatable once fielded — the churned are not on the list, and the angriest never reply to anything. Refunds tell you about people who bothered to ask for one. Top customers tell you what kept the people who stayed. Every report here carries a standing section answering one question: who is missing from this data, and would including them change the answer?

Questions that supply their own answer. Leading phrasing, double-barrelled items that ask about two things and accept one response, option lists that overlap or leave gaps, and a five-point scale that runs positive-to-negative on page one and negative-to-positive on page three. Unlike a tracking error, none of this is fixable afterwards. The flaw is baked into every response you collected, and the only remedy is to field it again — which is why the questionnaire gets picked apart line by line before a single person sees it.

See where the site stands before you brief anyone.

If the site is already live, start from what it scores rather than from a guess. The audit reports what it checked, what it found, and — the part most tools leave out — what it could not check, and why.

See the website audit Talk about this work

Questions people ask before they commit.

How many responses do we actually need?: For a large population, roughly 385 completed responses gives you 95 percent confidence with a 5 percent margin of error. For a small finite list — say around 800 customers — the requirement drops to about 260. Those are the arithmetic floors. The frame matters more than the count, because 400 responses from the wrong people is worse than 150 from the right ones.
What if we do not reach that number?: Then the report says so and adjusts what it claims. Small samples are still genuinely useful for finding themes, generating hypotheses and ruling things in for further work. What they cannot do is support a percentage stated about your whole customer base, and we will not write one.
Can you analyse a survey we have already run?: Yes, and that is the cheaper end of this work. The limit is honest — wording problems, a biased frame and missing answer options cannot be repaired after fielding. We will clean it, analyse it properly, and tell you plainly which of your questions can carry a conclusion and which cannot.
Should we offer an incentive?: It reliably lifts response rate and it changes who responds, and both effects need accounting for. An incentive pulls in people motivated by the reward rather than the topic, which shifts your mix. Whether that trade is worth making depends on how badly your natural response rate skews, and we work that out before fielding rather than after.
We need to hear from people who are not our customers. Is that possible?: Yes, through a research panel such as Prolific or Respondi. Panel responses are bought per completed interview and that cost sits outside this fee as a separate line, because it scales with sample size and screening strictness rather than with our work.
What if the findings contradict what we expect?: They get reported as found. Agreeing in advance what the answer will be is the one thing that makes a survey worthless, so the finding stands and the write-up explains the evidence behind it. Where the data is genuinely ambiguous, the report says that too rather than picking the more comfortable reading.

Often taken together.

Start with the brief, not the invoice.

Send what you are trying to achieve. You get a written scope, a price and a date — or an honest “this is not the right service for that”.

Talk about Survey & market research analysis All Data & Analytics services