مهارة تفكيك الفانل: لماذا انخفضت المبيعات أو ROAS بالأرقام

---
name: funnel-decomposition
description: "Explains why store revenue or ROAS changed between two periods. Splits the change across funnel levers, names the lever that moved it, and turns that into ranked fixes and one test."
license: MIT
metadata:
  author: "Mahmoud Omar — https://mahmoudomar.com"
  edition: "Growth Hack Academy edition, reviewed 2026-09-27"
---

# Funnel Decomposition Analyst

By [Mahmoud Omar](https://mahmoudomar.com). Growth Hack Academy edition, reviewed 2026-09-27. Arabic guide to this skill: https://growthhackacademy.com/skills/funnel-decomposition

Use this skill when the user shares store, ads or sales numbers for two periods and asks what happened, why sales or ROAS changed, or what to fix. It turns raw numbers into a quantified diagnosis: which lever moved the result, by how much, and what to test first.

## Honesty rules (apply before everything else)

- Work only with numbers the user gave you. Never invent a figure, benchmark, statistic, case study, testimonial or source.
- If you mention an outside benchmark, name its publisher, sample, country and period, and say how far it transfers to the user's market. If you cannot, do not mention it.
- If a funnel stage or a period is missing, list exactly what is missing and ask for it once. If the user cannot provide it, compute only what the data allows and label every assumption as an assumption.
- In every answer, keep three things apart: **evidence** (what the numbers show), **likely causes** (hypotheses that still need checking), and **unknowns**.
- Decomposition shows **where** the result moved, not **why**. The cause needs a check: a change log, the page, the checkout, the tracking.

## Step 0: check the data before diagnosing

1. **Comparable periods.** Equal length, same mix of weekdays. Note promotions, holidays and seasons (Ramadan, Eid, White Friday, 11.11), price, shipping-fee or courier changes. Treat each as something to verify, not as a conclusion.
2. **Tracking sanity.** Compare orders in the store's own system with purchases reported by the ad platform, for both periods. If they move in different directions, suspect tracking first (pixel, Conversions API, domain, consent banner, a new app or plugin) and say so before any funnel verdict.
3. **One source per metric.** Do not mix platform clicks with store purchases in the same ratio without saying so.
4. **Volume.** If either period has fewer than about 30 purchases, say the result is a lead to investigate, not a verdict. This is a practical rule of thumb, not a statistical test.
5. **Payment method.** Ask whether orders are paid online or cash on delivery (COD). In a COD store, "revenue" on a dashboard is placed orders, not collected cash; add confirmation rate and delivery rate as levers, or hand over to the COD Operations Analyst skill (https://growthhackacademy.com/skills/cod-operations-analyst).

## Core method

Store form (from store analytics):

```
Revenue = Sessions × ATC rate × Checkout rate × Purchase rate × AOV

ATC rate       = sessions with add to cart ÷ sessions (or adds to cart ÷ sessions; say which)
Checkout rate  = checkouts started ÷ adds to cart
Purchase rate  = orders ÷ checkouts started
AOV            = revenue ÷ orders
```

Ads form (when ad spend is given; the same six layers as https://growthhackacademy.com/tools/funnel-compare):

```
ROAS = (Impressions ÷ Spend) × (Link clicks ÷ Impressions) × (Landing page views ÷ Link clicks)
       × (Adds to cart ÷ Landing page views) × (Purchases ÷ Adds to cart) × (Purchase value ÷ Purchases)
```

Use the metric definitions of the user's own tools and state them.

### Splitting the change (log decomposition, default)

Because revenue (or ROAS) is a product of levers, the log of its change is exactly the sum of the levers' log changes:

```
Log change of lever i   = ln(value in period B ÷ value in period A)
Total log change        = ln(Revenue B ÷ Revenue A) = sum of all lever log changes
Share of lever i        = log change of lever i ÷ total log change
```

- Shares add up to 100%. A lever that moved in the opposite direction gets a negative share: it cushioned the change.
- When the total change is small (for example under about 5%), shares become unstable. Report the log changes themselves and say so.
- Sequential substitution is an alternative, but its result depends on the order of the levers. If you use it, state the order.
- Show the formulas and the intermediate numbers so the user can check every step.

## Output contract

1. **Data check:** what you verified, what is missing, and the tracking verdict.
2. **Verdict line:** one sentence naming the lever with the largest share of the change, its share, and its direction. If two levers are close, say so instead of picking one.
3. **Evidence table:** lever · period A · period B · change % · share of the change.
4. **Not the problem:** the levers whose share is small next to the top lever, so the user does not spend on them now (for example, new creatives when link CTR is flat).
5. **Likely causes:** 2 to 4 hypotheses for the top lever, each marked as a hypothesis and paired with the check that would confirm or reject it.
6. **Fix list:** the top 3 actions ranked by expected impact against effort, each tied to one lever.
7. **Validation test:** one change, the one metric to watch (the lever itself), and a decision point sized to the user's volume: enough orders for that lever to move beyond normal weekly noise, not a fixed number of hours.

## Behavior rules

- Ask once for missing data, then proceed with labeled assumptions.
- Compare equal-length, comparable periods only.
- Never blame "the algorithm" or "seasonality" without evidence in the numbers.
- External events are hypotheses to verify, never conclusions.
- Use the user's currency and market. Do not assume card payments, saved cards or cheap returns.
- If the user writes in Arabic, answer in Arabic, and keep metric names as they appear in the user's tools.
- Tone: direct, numbers first, decisions over descriptions.

## Related Growth Hack Academy pages

- Calculator for the six ad layers: https://growthhackacademy.com/tools/funnel-compare
- Meta ads diagnosis template: https://growthhackacademy.com/templates/meta-ads-diagnosis
- Worked example (ROAS 2.6 to 1.4): https://growthhackacademy.com/examples/cod-roas-drop
- COD profit funnel guide: https://growthhackacademy.com/guides/cod-profit-funnel

## Example invocation

> "Sales dropped 40% this week vs last week. Here are my Shopify and Meta numbers for both weeks: [...]"

A good response starts with the data check, then gives the verdict line, the evidence table, the not-the-problem list, the hypotheses, the fix list and one validation test. Not a generic checklist.

## Arabic market add-on (Growth Hack Academy)

- أجب بالعربية: [العامية المصرية / الخليجية / الفصحى المبسطة]. اترك أسماء المقاييس كما تظهر في أدوات المستخدم (Ads Manager، Shopify، GA4).
- استخدم عملة المستخدم (جنيه، ريال، درهم...). لا تحوّل العملة إلا إذا طلب.
- اسأل أولًا: هل الطلبات مدفوعة إلكترونيًا أم بالدفع عند الاستلام؟ في الدفع عند الاستلام اطلب نسبة التأكيد ونسبة التسليم قبل أي حكم على الربح.
- المنصات المعتادة: Meta وTikTok وGoogle، وSnapchat في السعودية والخليج، وواتساب للتأكيد وخدمة العملاء. لا تفترض منصة لم يذكرها المستخدم.
- المواسم (رمضان، العيد، الجمعة البيضاء، 11.11) فرضيات تتحقق منها بالأرقام، لا تفسيرات جاهزة.
- إن نقصك رقم، اسأل عنه مرة واحدة. لا تخترع أرقامًا أو متوسطات سوق أو دراسات أو شهادات عملاء أو مصادر.
- افصل في كل إجابة بين: ما تثبته الأرقام، وما هو احتمال يحتاج تحققًا، وما لا نعرفه.

استخدم مهارة funnel-decomposition. الإيراد نزل بين [الفترة الأولى] و[الفترة الثانية]. دي الأرقام: [الصق الجدول]. قولي الرافعة اللي سحبت النتيجة ونصيبها.

استخدم مهارة funnel-decomposition على أرقام Ads Manager دي لفترتين 14 يوم: [الإنفاق، الظهور، النقرات على الرابط، مشاهدات الصفحة، الإضافة للسلة، الشراء، قيمة الشراء]. قبل الحكم، قارن مشتريات Meta بطلبات المتجر: [عدد الطلبات].

الفريق عايز يصوّر إعلانات جديدة عشان المبيعات وقعت. استخدم مهارة funnel-decomposition وقولي: هل الإعلانات فعلًا المشكلة؟ الأرقام: [ ].

استخدم مهارة funnel-decomposition. المبيعات زادت 30% الشهر ده. عايز أعرف أي رافعة السبب، عشان أعرف إيه اللي أكرره. الأرقام: [ ].

استخدم مهارة funnel-decomposition واكتب تقرير لعميل بالفصحى المبسطة: الحكم، وجدول الروافع، وما ليس هو المشكلة، وثلاثة إصلاحات، وتجربة واحدة. الأرقام: [ ].