Episode 40: Data se campaign decisions lena
Hypothesis, experiment, result aur decision ka loop; sample size ki samajh; aur data ke saath common galtiyan.
Monday ko orders 5, Tuesday ko 1. Neha ghabra gayi: “Ads band karo!”
Ek din ka data mausam hai, chaar hafte ka data climate. Mausam dekh ke ghar nahi badalte.
Decision loop
Example hypothesis: “Agar product page par size chart image ke turant neeche ho, toh conversion 10%+ badhega, kyunki size dar sabse bada blocker hai (Chapter 36 data).”
Achhi hypothesis mein: badlav, expected result, aur reason.
Kitna data kaafi hai?
- Chhote numbers mein random fark bahut hota hai. 3 vs 5 orders = shayad kismat
- Weekday/weekend, salary week, festival: poora cycle dekho (kam se kam 1–2 hafte)
- Fark jitna chhota, utna zyada data chahiye usse saabit karne ke liye
- A/B test calculators (online) se andaza lagao ki kitne visitors/orders chahiye
Data ke common traps
| Trap | Example |
|---|---|
| Correlation ≠ causation | Diwali mein orders badhe aur naya ad bhi chala. Credit kisko? |
| Cherry picking | Sirf best din dikhana |
| Vanity focus | Views double, orders same |
| Survivorship | Sirf jo customers ruke unse poochna, jo gaye unse nahi |
| Attribution bias | Har platform sale ka credit khud leta hai |
Pehle se rules banao
Test se pehle likho ki kya result aane par kya karoge. Baad mein mann nahi badlega.
| Agar | Toh |
|---|---|
| Conversion +10% ya zyada, 3 hafte | Permanent karo |
| ±10% ke beech | Koi fark nahi, simple version rakho |
| −10% ya kam | Wapas purana |
Numbers + log = poori kahani
Numbers batate hain kya hua. Customers batate hain kyun.
- Monthly 5 customer calls (10 min)
- DMs mein common sawaal ki list
- Reviews mein bar-bar aane wale shabd
Jab dono ek hi baat bolein, tab decision pakka.
StyleStreet ki real problem
Reaction mode mein marketing
Neha har din numbers dekh ke kuch badal deti: ek din price, ek din ad, ek din caption style. 3 hafte baad kuch pata nahi ki kya kaam kiya.
Data tha, lekin system nahi tha.
Practical example: StyleStreet experiment log
| # | Hypothesis | Result (3 hafte) | Decision |
|---|---|---|---|
| 1 | Size chart image ke neeche → conversion ↑ | +18% | Permanent ✅ |
| 2 | Free shipping ₹999+ → AOV ↑ | AOV ₹960 → ₹1,080 | Permanent ✅ |
| 3 | Story daily poll → DMs ↑ | +4% (noise range) | Optional, simple rakho |
| 4 | 10% discount pop-up → conversion ↑ | +6%, lekin margin −₹96/order | Band ❌ |
Test 4 ne sales thodi badhayi, lekin profit ghataya. Decision profit se.
Aapka mission: Experiment system banao
Apne ya StyleStreet ke liye:
- 3 hypotheses likho (badlav + expected result + reason)
- Har ek ka main metric aur duration likho
- Test se pehle decision rules likho
- Experiment log sheet ke columns banao
- 3 data traps ke apne examples likho
- 5 customer-call sawaal likho (“kyun” samajhne ke liye)
Quiz
-
Achhi hypothesis mein kya hona chahiye?
- A. Sirf idea
- B. Badlav, expected result aur reason
- C. Sirf number
- D. Kuch nahi
Answer dekho
Sahi jawab: B. Badlav, expected result aur reason
Tabhi test ka matlab samajh aata hai.
-
Ek din orders 1, ek din 5. Kya karein?
- A. Turant ads band
- B. Trend ke liye 1–2 hafte ka data dekho
- C. Price badlo
- D. Sab badlo
Answer dekho
Sahi jawab: B. Trend ke liye 1–2 hafte ka data dekho
Chhote samay ka data bahut random hota hai.
-
Diwali mein orders badhe, naya ad bhi tha. Ad ko credit dena?
- A. Haan, pakka
- B. Sambhal ke: season bhi reason ho sakta hai (correlation ≠ causation)
- C. Nahi, kabhi nahi
- D. Diwali ko
Answer dekho
Sahi jawab: B. Sambhal ke: season bhi reason ho sakta hai (correlation ≠ causation)
Saath hona, karan hona nahi.
-
Discount ne conversion +6% kiya lekin profit ghataya. Decision?
- A. Rakh lo, conversion badha
- B. Band karo ya badlo; profit goal hai
- C. Discount double
- D. Ignore
Answer dekho
Sahi jawab: B. Band karo ya badlo; profit goal hai
Business ka goal profit hai, sirf conversion nahi.
Client project mein output
Project file mein ab Decision System:
- Hypothesis template
- Pre-set decision rules
- Experiment log (4 tests)
- Customer feedback routine
Stage 7 complete! StyleStreet ab andaze se nahi, data se chalti hai.
Golden rule: Mausam dekh ke ghar mat badlo. Hypothesis likho, rules pehle banao, kaafi data lo, aur faisla profit se karo.