BigCat Research
How to write a research brief or RFP and what to ask an agency about AI use, brand tracking and field audits
Short answer: A good brief states the decision the research will inform, the audience, the questions, any method preferences, the budget range, the timeline, the deliverables, who owns the data and how proposals will be scored. Add specific questions on AI use, tracker design and field audits so you can compare agencies on the same terms.
Write the brief around a decision
Start with one sentence: "We need to decide whether to X, and this research should tell us Y." Everything else follows. A brief that opens with a fixed method (for example "we want a survey of 1,000 people") settles the method before the question is clear, so agencies may quote what was asked rather than what is needed.
Include what you already know. Existing data, earlier studies and internal hypotheses save fieldwork cost and let a good agency challenge your assumptions.
RFP template with evaluation scorecard
| Section | Content |
|---|---|
| 1. Background | Company, category, situation, why now |
| 2. Decision and objectives | The decision, 3 to 5 research questions |
| 3. Audience | Target respondents, geography, exclusions |
| 4. Method preferences | Quantitative, qualitative, desk, mixed; open to proposals? |
| 5. Existing material | Prior studies, customer data, brand guidelines |
| 6. Deliverables | Report, executive summary, dataset, dashboard, workshop |
| 7. Timeline | Kick-off, fieldwork window, interim, final, decision date |
| 8. Budget | A range, or a statement that you cannot share one |
| 9. Data and IP | Who owns data and questionnaires, storage, deletion, confidentiality |
| 10. Proposal format | Length, team CVs, references, methods annex |
| 11. Evaluation | Criteria and weights (below) |
Scorecard (adjust to your case):
| Criterion | Weight | Score 1-5 |
|---|---|---|
| Understanding of the decision | 20 | |
| Method fit and rigour | 25 | |
| Team and relevant experience | 15 | |
| Data quality controls | 10 | |
| Use of technology / AI with human oversight | 5 | |
| Deliverables and usability of results | 10 | |
| Timeline realism | 5 | |
| Cost clarity and value | 10 |
Consider sharing the scorecard with bidders: it tells them what you value and gives you a documented basis for your decision.
What to ask an agency about AI
AI can speed up desk research, document scanning, classification of open-ended answers and report drafting. It does not remove the need for judgement. Ask:
| Question | What a good answer includes |
|---|---|
| Where exactly is AI used in my project? | Named stages (for example text classification), not "throughout" |
| Who reviews AI output? | A named human reviewer; sampling of outputs for accuracy |
| How is output validated? | Comparison with manual coding on a sample, checks against sources |
| Where does my data go? | Which tools process it, storage, retention, whether data trains external models |
| Can I opt out? | Yes, with consequences for cost or timeline explained |
| How are sources handled? | Citations retained, reliability assessed by a person |
Limits to understand. AI summaries can sound confident and still be wrong. Classification errors can be systematic. "Synthetic respondents" (models simulating survey answers) are not people: they may help with early hypothesis generation, but should not replace fieldwork for decisions that depend on real opinion or behaviour. An agency that says otherwise should explain its validation evidence.
Choosing a brand health research provider
Brand health tracking repeats the same measures on a defined audience so you can see change. What to check:
| Area | Check | Why it matters |
|---|---|---|
| Questionnaire ownership | You own the questionnaire and can move it to another provider | Avoids lock-in and keeps trends comparable |
| Sample and weighting | Population definition, quotas, weighting variables | Waves are comparable only if the sample is |
| Frequency | Matches decision cycles (continuous, quarterly, annual, pre/post campaign) | More waves cost more and may not add insight if decisions are made less often |
| Dashboards and export | Raw data export, documented variable names | You can analyse beyond the dashboard |
| Metrics | Awareness, consideration, preference, NPS, trust, perceived quality | Read as a chain: where does it break? |
| Competitor set | Which brands are measured and why | A score is meaningless without a reference point |
| Change rules | How questionnaire changes are handled | Wording changes can break the trend |
We do not quote benchmark numbers: they vary by category, market and method, and published norms without context mislead. Ask any provider claiming benchmarks for their source and definition.
Running a dealer service quality or field audit
- Define the standard. Write down what good looks like at each touchpoint (greeting, waiting time, product knowledge, cleanliness, issue handling).
- Build a scorecard. Criteria, scoring scale, weights, what counts as a critical failure.
- Choose methods. Field observation by trained observers, customer and staff surveys, review analysis; combine rather than rely on one.
- Sample locations. Cover regions and dealer sizes; repeat visits at different times.
- Train observers and calibrate. Joint practice visits so two observers score the same visit alike.
- Report by location and standard. Show gaps against the standard, trend between waves, and priority actions for the network.
- Close the loop. Agree owners and re-measure.
Note on terms: some providers use "mystery shopping".
FAQ
1. Should I give the budget in an RFP? A range helps agencies propose the right scale. If you worry about anchoring, ask for options at different budget levels.
2. How many agencies should I invite? There is no fixed number. A short list that lets you compare proposals properly, without asking many agencies to invest in bids they are unlikely to win, is a reasonable aim.
3. How often should brand health be tracked? It depends on how fast your category and campaigns move. Options include continuous, quarterly or annual waves, or pre/post campaign measurement; choose the interval at which you can act on change.
4. Can AI replace a survey? No. It can support parts of the workflow, but opinions and behaviour of real people still need to be measured.
5. Who should own the questionnaire and data? The client, unless agreed otherwise; state it in the brief.
How BigCat approaches this
BigCat Research is an independent research company in Istanbul. Relevant pages: brand health research and dealer service quality.
BigAgent is the working layer BigCat uses to extend research production capacity across desk research, open-source scanning, digital competitor visibility, text classification, review analysis, early finding extraction and report pre-structuring. It is not positioned as a standalone product. Human oversight statement: method choice, source reliability, interpretation and recommendations remain with BigCat researchers; BigAgent does not replace them.
We make no benchmark claims and name no clients on this page.
Author: BigCat Research team. Last reviewed: 3 October 2026.