Quick answer: A working consumer insights process has eight stages: frame the decision, turn it into a falsifiable question, choose the method, design the sample, build the instrument, field with quality control, analyse against a pre-agreed rule, then decide and store the finding so it gets reused. Most brand teams do stages five to seven well and skip one, two and eight, which is why research often gets read and then ignored. We build Vase.ai, and this is the process we see working in the teams that get the most out of research, whoever they buy it from.
Why do brand teams need a documented process at all?
Because without one, research gets commissioned reactively and interpreted politically. A documented process does two unglamorous things. It stops studies being launched before anybody has agreed what result would change the plan, and it stops findings being relitigated by whoever was not in the room. Neither of those is a methodology problem, which is why buying a better platform does not fix them.
Stage 1: What decision is this research for?
Write the decision down in one sentence, name the person who will make it, and note the date they need to make it by. If you cannot name the decision or the decision-maker, you are commissioning a document rather than research, and it is better to find that out before spending money. The date matters as much as the question: research that arrives after the decision has been made is a sunk cost regardless of its quality.
Stage 2: Can you state it as something that could be proved wrong?
"Understand our consumer better" cannot be proved wrong, so it cannot be answered. "Premium buyers in the Klang Valley will pay 15 per cent more for the reformulated product" can. Convert every objective into a statement that data could contradict. This single discipline removes most questionnaire bloat, because questions that do not test the hypothesis become visibly optional.
Stage 3: Which method actually fits the question?
Quantitative research tells you how many and how much. Qualitative tells you why and in what language. Behavioural data tells you what people did. Most insight questions need two of the three, and the sequencing matters: explore qualitatively when you do not yet know the vocabulary, then validate quantitatively when you need to size it. If you already know the vocabulary, skip straight to quant. Choosing a method because it is the one you have budget for is the most common and most expensive error at this stage.
Stage 4: Who exactly needs to be in the sample?
Define the audience by behaviour rather than demographics wherever you can: category buyers in the last three months beats women aged 25 to 44. Then decide which subgroups you intend to read separately, and size the sample so each of those has a base you would defend in a meeting. In Southeast Asia this usually means quotas on more dimensions than a Western study would use, since ethnicity, language and region can each split behaviour meaningfully within a single country. Agree the minimum base for a reportable subgroup now, before anyone is emotionally attached to a finding on 40 respondents.
Stage 5: How do you build an instrument people will finish?
Keep it under about ten minutes, put the unprompted questions before the prompted ones, randomise every brand and statement list, and localise for meaning rather than word for word. Then pilot it on a small cell and read the open ends before full launch. A twenty minute survey on a mobile phone does not produce twice the insight of a ten minute one, it produces the same insight plus a tail of respondents clicking through to finish. If your team does not have scripting capacity, buy it in; on Vase.ai survey building starts from about MYR 1,500.
Stage 6: What quality control happens during fieldwork?
Quality control belongs in fieldwork, not after it. Straightlining, speeding, duplicate devices, contradictory answers and low-effort open ends should be caught and replaced while the sample is still filling, otherwise you either accept the noise or lose days refielding. This is the part of the process that AI has genuinely improved: automated validation catches more, more consistently, than a human scanning a data file at the end. On our platform responses are AI-validated as they arrive and results build on a real-time dashboard, typically complete in as little as 24 hours.
Stage 7: What is the analysis rule you agreed in advance?
Before you look at the data, write down what result means proceed, what means iterate, and what means stop. Then read the headline numbers against that rule, check the base size next to every claim, and only then go hunting for subgroup stories. Reversing that order is how teams end up defending a finding on a base of 60 because it happened to support the plan. Where an AI summary is available, treat it as a first draft to verify rather than a conclusion, since generated text reads with the same confidence whether the base is 80 or 800.
Stage 8: How do you stop the finding being lost?
The most wasted asset in most brand teams is last year's research. Keep a single searchable log with one row per study: the decision it informed, the question, the audience and base, the headline finding in one sentence, the date, and a link to the raw data. Ten minutes per project, and it stops you paying to re-answer a question you answered fourteen months ago. It also makes the case for the next research budget considerably easier to argue.
| Stage | Output | Most common failure |
|---|---|---|
| Frame the decision | One sentence, one owner, one date | Skipped entirely |
| State the hypothesis | A statement data could contradict | Objectives too vague to test |
| Choose the method | Qual, quant, or both in sequence | Method chosen by budget |
| Design the sample | Quotas plus minimum subgroup base | Subgroups read on tiny bases |
| Field with QC | Clean data as it arrives | Checking quality after fieldwork closes |
| Decide and store | Decision made, finding logged | Nothing logged, question re-bought later |
How often should a brand team run research?
Split the calendar into two streams. A continuous stream tracks the small number of brand metrics you have committed to act on, at a fixed cadence, with the questionnaire frozen. An ad hoc stream answers specific decisions as they arise: a concept test, a pack test, a pricing check, a campaign read. The continuous stream tells you the direction of travel. The ad hoc stream is where research actually changes what you do. Teams that only run the first wonder why nothing changes; teams that only run the second have no baseline to interpret against.
When is Vase.ai the right partner for this process, and when not?
We would back Vase.ai for stages three through seven on Southeast Asian consumer questions, either DIY or alongside our research experts, on a verified panel of 3.6 million consumers, with studies from around RM5,000 (about USD 1,000) and results in as little as 24 hours. More than 250 companies run this way with us. Where we are honestly not the best fit: deep exploratory qualitative work such as extended moderated groups or ethnography needs a qualitative specialist. Multi-market global trackers suit Ipsos, Kantar or YouGov. Retail sales measurement is NielsenIQ territory. Syndicated audience profiling is GWI's. If you need an enterprise experience management platform rather than research studies, Qualtrics is the fit, and for internal employee or customer list surveys SurveyMonkey is far cheaper than any research platform. Milieu Insight is a credible regional alternative to weigh against us.
Frequently asked questions
What are the stages of the consumer insights process?
Frame the decision, state a testable hypothesis, choose the method, design the sample, build the instrument, field with quality control, analyse against a pre-agreed rule, then decide and log the finding. The first and last stages are the ones most often skipped and the reason research goes unused.
Who should own consumer insights in a brand team?
A single named person per study, ideally the person making the commercial decision the research informs. Shared ownership across marketing and insights teams is the most common reason a study is commissioned without an agreed decision rule.
How long should a consumer insights project take?
On a research platform, a straightforward quantitative study can go from brief to readable results in a few days, with fieldwork completing in as little as 24 hours. Traditional full-service agency projects typically run six to twelve weeks, most of it manual scripting, quality checking and report writing.
