03-01 Качественные методы | Demand Planning Course
03-01 Качественные методы
·
·
Когда нет данных — как прогнозировать?
Иногда у нас нет истории продаж: новый продукт, стартап, резкий слом рынка. В такие моменты мы не «считаем по формулам», а собираем умы: спрашиваем экспертов, ищем похожие запуски и готовим несколько сценариев. Это и есть качественные методы (Qualitative Methods) — разумная структура вокруг «умных оценок».
Метод Дельфи (Delphi Method)
Оценка PERT (PERT Estimate)
Метод аналогии (Analogy Method)
Сценарное планирование (Scenario Planning)
Когда использовать?
Новый SKU или рынок без истории (New Product Introduction).
Сильная инновация или смена архитектуры продукта (Disruptive Innovation).
Ранний период после запуска (Early Sales Window).
Цель — получить реалистичный диапазон (range) и договорённый базовый сценарий (Base Case) до тех пор, пока не появятся данные для количественных методов (Time Series / Causal).
Метод Дельфи (Delphi Method)
Способ быстро собрать мнения нескольких экспертов и превратить их в один понятный прогноз-коридор: база (Base) и границы p20–p80. Без споров и доминирования.
Когда применять
Новый продукт, данных мало.
Рынок нестабилен.
Нужно решение быстро.
Пошаговый алгоритм (5 шагов)
Сформулировать вопрос прогноза (Forecast Question): что, на какой период и в каких единицах прогнозируем.
Медиана (Median) — центральное значение, устойчивое к выбросам.
Интерквартильный размах (Interquartile Range, IQR) — ширина центральных 50 % значений.
FVA (Forecast Value Added) — вклад шага процесса в улучшение прогноза.
Миф (Myth)
«Мнение отдела продаж всегда точное». На практике оценки нужно усреднять и проверять через Дельфи и аналогии.
Ловушка (Trap)
Использовать одно число без условий нельзя. Всегда фиксируйте Базу + p20–p80 + Условия.
Мини-тест (Mini-Quiz)
Что показывает показатель IQR?
When there is no data — how to forecast?
Sometimes we have no sales history: a new product, a startup, or a market disruption. In such moments we cannot rely on formulas — instead, we gather minds, ask experts, look for similar launches, and prepare several scenarios. This is what Qualitative Methods mean — a rational structure built around smart judgments.
Delphi Method
PERT Estimate
Analogy Method
Scenario Planning
When to use:
New SKU or market with no historical data (New Product Introduction).
Periods of turbulence — crises, sanctions, pandemics (High Uncertainty).
Strong innovation or a disruptive change in product architecture (Disruptive Innovation).
Early post-launch phase while data are still accumulating (Early Sales Window).
The goal is to obtain a realistic range and a mutually agreed Base Case until enough data appear for quantitative models (Time Series or Causal).
Delphi Method
The Delphi Method helps to collect expert opinions quickly and transform them into a single forecast corridor: a Base and the limits p20–p80 — without discussions or domination by louder voices.
When to apply
New product with limited data.
Market volatility or uncertainty.
A decision is needed within days, not “when models are ready.”
Five-step procedure
Formulate the forecast question. Define what is forecasted, for what period, and in which units. Example: “How many units of SKU X will we sell in September (retail, pcs)?”
Select 5–8 experts from Sales, Marketing, Supply, Finance, and eCommerce. All answers must remain anonymous.
Round 1 — individual estimates: each expert provides Min, Most Likely, Max and key assumptions.
Aggregate and share anonymized results: the tool calculates the median (Base) and corridor p20–p80.
Round 2 — review and convergence: experts revise estimates; if the spread is moderate, stop; otherwise one more round.
Output
Base — the number used for planning.
p20–p80 — realistic range around the Base.
Assumptions — conditions under which the forecast is valid.
Triggers — signals for shifting toward Best or Worst.
Parameter
Value
Base
10 200 pcs
p20–p80
9 300 – 10 800 pcs
Assumptions
Distribution 60 %, price = competitor, 2 promo weeks
Triggers: by week 2, distribution < 25 % → stay near p20; > 45 % and sales per store > 1.1 → move toward p80.
Practical use: plan production and procurement on the Base; keep inventory and capacity within the p20–p80 range; check triggers weekly.
PERT (Three-point estimate)
The PERT Method (Program Evaluation and Review Technique) converts three expert estimates — Min, Most Likely, Max — into a single number using the formula:
PERT = (Min + 4 × Most Likely + Max) / 6
Why multiply Most Likely by 4?
This weighting (λ = 4) was adopted in the U.S. Navy project-management standard: it emphasizes the most probable value while still accounting for extremes.
Min
Most Likely
Max
λ
PERT
7 000
9 000
12 000
3
≈ 9 200
7 000
9 000
12 000
4
≈ 9 167
7 000
9 000
12 000
8
≈ 9 100
IQR — Interquartile Range
IQR = p75 − p25 shows how tightly expert opinions are clustered. A smaller IQR means better consensus; a large IQR indicates divergence and the need for another Delphi round.
Group
p25
p75
IQR
Comment
1
9 400
10 000
600
Narrow – good agreement
2
8 200
10 800
2 600
Wide – disagreement
Analogy Method
The Analogy Method uses a similar past launch and carefully adjusts its results to current conditions.
When to use
New SKU but with comparable launches in the same segment.
Comparable price, channel, and marketing support.
Five Steps
Select an analog: check similarity in segment, price tier, channel, season, media, and region.
Normalize sales curve: bring analog’s post-launch sales (12 weeks or months) to a common timeline.
Set four multipliers: Distribution Ratio, Price Multiplier, Media Multiplier, Cannibalization Multiplier.
Compute combined factor: Combined = Distribution × Price × Media × Cannibalization.
Multiply the analog curve by Combined to get Base / Best / Worst.
Scenario
Distribution
Price
Media
Cannibalization
Combined
Base
0.83
0.95
1.10
0.90
0.78
Best
1.00
0.975
1.15
0.95
1.07
Worst
0.67
0.925
1.00
0.85
0.52
Scenario Planning
Scenario Planning prepares several plausible futures — Base / Best / Worst.
When needed
High uncertainty (listing, weather, promo, competition).
Key drivers: Distribution, Price, Media.
Five Steps
Identify Drivers and set levels for Base/Best/Worst.
