CocoVida: From Purchase Intent to Real Trial
CocoVida, a Thai coconut-water brand, commissioned an online panel of 400 Bangkok adults aged 18–45. Sixty-one percent chose the top two boxes on a seven-point purchase-intention scale ('definitely/probably would buy') when shown a concept at THB 35 per 330 ml bottle. The brand team wants to use this to justify a national launch targeting 25% household penetration in year one. After a four-month soft launch in 120 convenience stores, actual trial among the target group was 19%, and repeat purchase within eight weeks was 11%. The panel was 68% female and 74% Bangkok-centric, while category buyers are 45% male and 40% live outside greater Bangkok. The questionnaire showed the product image before price, asked 'Don't you agree that a natural, healthy drink is better than sugary soda?', and measured intent before describing the THB 35 price. Research budget for a second study is THB 900,000.
1. Compute the 95% confidence interval for the 61% intention figure and explain why it does not resolve the 61%-versus-19% gap. Identify the three largest validity threats, in order.
Read the worked answer
Calculation. Standard error = √(0.61 × 0.39 ÷ 400) = 0.0244. The 95% confidence interval is 61% ± 4.8 percentage points, or 56.2%–65.8%. With a design effect of 1.5 from weighting, the interval widens to roughly ±7.4 points, or 53.6%–68.4%. Even the lower bound is far above 25%, so sampling error cannot explain 19% trial. The gap is bias, not noise.
Three largest threats, in order.
- Measurement: intent was asked before price and after a leading question, so it measures agreement with 'natural is better', not purchase at THB 35. Top-two-box intent is hypothetical.
- Coverage and selection: a non-probability online panel that is 68% female and 74% Bangkok-centric cannot represent a category that is 45% male and 40% upcountry. Weighting cannot fix unobserved panel bias.
- Concept-to-market gap: a concept image is not a shelf in a convenience store. Distribution, visibility and habit are missing.
Social desirability and acquiescence are a fourth, smaller factor. The 11% repeat confirms that trial quality was poor, not merely that awareness was low.
2. Redesign the study for THB 900,000. Specify sampling frame, sample size logic, question sequence, and one behavioural validation method. State what you would no longer measure.
Read the worked answer
Budget split. THB 450,000 for a nationally recruited, quota-controlled online sample of about 700 respondents; THB 300,000 for in-store intercepts and a conversion test in 20 stores across Bangkok and one upcountry province; THB 150,000 for analysis, incentives and a small price test.
Sampling frame. Recruit from a large retailer's loyalty database and store intercepts, not the brand's social followers. Set quotas on gender, age, region and category usage.
Sample-size logic. A simple proportion at ±5 points and 95% confidence needs 384 respondents per key cell. With a design effect of 1.5 and 20% attrition, budget about 700 completes for the main sample and rely on intercepts for the behavioural read.
Question sequence. Category behaviour → brand awareness → concept without price → price sensitivity (four-cell test at THB 29, 35, 39 and 45) → purchase intent in a real channel ('would you buy in the next seven days at this store?') → demographics. Replace top-two-box intent with a forced-choice constant-sum or a discrete-choice conjoint on price, flavour and pack size.
Behavioural validation. Run a six-week in-store availability test and compare actual take-up with the model forecast.
Stop measuring long attitude and image batteries; they consume budget without answering the launch question.
3. The CMO asks whether 11% repeat is good. What comparison and what additional metric would you need before answering?
Read the worked answer
11% repeat is neither good nor bad without a benchmark. Compare it with the category's eight-week repeat rate for new beverage SKUs and with the brand's own prior launch. Successful beverage launches often reach 15–25% repeat, with lower figures for trial-led launches, so 11% is a warning but not a verdict.
Repeat alone is insufficient. You need the trial base (19%), repeat rate, and purchase frequency or buying rate. Compute eight-week sales per 100 target households = trial rate × repeat rate × bottles per repeat purchase. Compare that with the launch hurdle. If trial expands, total penetration can still grow even at 11% repeat; the risk is a low-quality trial pool.
Also split repeat by whether the buyer paid full price or bought on promotion. If promotional trial repeats at 4% while full-price trial repeats at 18%, the 11% average hides a weak core. Finally, check availability: a stockout in the top stores can depress measured repeat without reflecting preference.