SUPPLY CHAIN MANAGEMENT

1. INTRODUCTION

If we have questions we can ask them at the end of the lesson.

The final exam will be closed books and closed notes and is based on a written test (2H) with:

- Numeric exercises
- Open theoretical questions

Damiano Milanato -> dmilanato@[Link]

Fabrizio Dallari -> fdallari@[Link]

2. SALES AND OPERATIONS PLANNING

“Supply Chain Management is the process of planning, implementing, and controlling the efficient,
cost-effective flow and storage of raw materials, work-in-process, finished goods and related
information from point of origin to point of consumption for the purpose of conforming to customer
requirements”

If I buy a good and after I sell it directly, I will be considering such as a perfect seller.
 2.1. The mission of Supply Chain Management is to get the 7 R:
- The RIGHT goods or service (on time in full = the customer service)
- The RIGHT quantity
- The RIGHT quality/condition
- The RIGHT place
- The RIGHT time
- The RIGHT customer
- The RIGHT cost

N.B. On time in full is related to the customer service.

2.2. S&OP Processes

We have to estimate the future demand before doing something or to keep stock. If we are the only
seller, we can wait the demands! Indeed, the customers will not be able to go somewhere else.

Some items will be put in the “make to stop” (e.g. the delivery will occur in 2-3 days)

If I notice that a company’s management is not okay, I can send a mail to explain what goes wrong in
explaining the situation.

 Customer Relationship Management - provides the structure for how relationships with
customers are developed & maintained, including the product and service agreements
between the firm & its customers.
 Customer Service Management - provides the firm’s face to the customer and provides a single
source of customer information (Customer Service).
 Demand Planning – provides the structure for balancing the customers’ requirements with
supply chain capabilities.
 Order Fulfillment & Distribution Planning – includes all activities necessary to define customer
requirements and fill customer orders.
 Manufacturing Flow Management & Inventory Planning - includes all activities necessary to
move products through the plants & to obtain & manage manufacturing flexibility in the supply
chain.
 Supplier Relationship Management - provides the structure for how relationships with
suppliers are developed & maintained.
 Product Development and Commercialization – provides the structure for developing and
bringing to market new products jointly with customers and suppliers.
  Returns Management – includes all activities related to returns, reverse logistics.

N.B. the longest waiting time is 60 days.

2.3. How to obtain a syncronized Supply Chain with S&OP ?

a) Responsive supply chain:

o Supply chain synchronization
o Agile vs. Lean / Responsive vs. Efficient
o SC Risk Management & Resilience
o Off-shoring vs. Near-shoring
b) Accurate forecasts:
o SC Visibility (demand planning vs. sales forecast)
o Matching demand patterns with forecast methods
o New product introduction
o Leveraging Demand Sensing & Big Data = it’s the capacity to catch resents
competitively. It is a forecasting method that leverages new techniques such
as artificial intelligence, machine learning and real-time data capture to create an
accurate forecast of demand based on the current realities of the supply
chain.[1][2] Traditionally, forecasting accuracy was based on time series techniques
which create a forecast based on prior sales history and draws on several years of data
to provide insights into predictable seasonal patterns. However, past sales are
frequently a poor predictor of future sales. Demand sensing is fundamentally different
in that it uses a broader range of demand signals, (including current data from the
supply chain) and different mathematics to create a more accurate forecast that
responds to real-world events such as market shifts, weather changes, natural
disasters and changes in consumer buying behavior.
c) Inventory: proper sizing of the stock.
o Inventory allocation (in the network)
o Optimal Safety Stock (holding cost vs. availability)
o Risk Pooling and Information Sharing (VMI, CPFR, “Beer Game”)
o Distribution Requirement Planning (DRP -> push method)
N.B. during the final test, the teacher will say which methods we will have to use but in the
end, we must be able to justify which one is the best and why!
 2.4. Wortmann classification of production / distribution systems

We have to be careful with the delivery because the clients can cancel the order. Before opening a
second warehouse, we must calculate the demands!!

2.5. The customer Order Decoupling Point:

- In the production-logistic system there is a decoupling point that separates the part of the
system managed on demand from the part that has to be managed on forecast
 2.6. The key components of the SCM / S&OP macro-process

A) Customer service strategy: strategy to serve the market.

E.g. my items are perfect. I’m excellent in customize but the client has to wait 2 months (it’s
the way you are serving your clients)
B) System Configuration: set out everything in order to guarantee the production (e.g open/close
some stores/warehouses, …)
C) Supply Chain Management: every time we have to check the operating machines
D) Logistics & Manufacturing execution: the stock management through customers’ services (est-
ce que vous avez bien reçu votre stock? Avez-vous assez?)
E) System management & control: starts of the KPs

The strategy is reviewed every 5 years.

2.7. Hierarchical levels of S&OP process

 2.8. S&OP / Supply chain planning

The allocation problem is linked to the stock

 Unreliable items from suppliers

 Underestimation of the demand

2.9. Supply chain planning workflow & 5 typologies of PLANS

Why inventory planning comes before distribution planning?

 I have to plan the demand before choosing where to distribute

At the end of the fiscal year, it’s better to have a stock as close as 0. We prefer to have less stock than
to have happy customers.

2.10. Output of the planning process

 2.11. Gaining competitive advantage through the supply chain

The most importance in the supply chain is performance.

(le point rouge 85%)

The the same cost I increase the customer service levering (CSL est l’axe des abscisses)

La grosse ligne noire est la COST TO SALE SERVICE

2.12. S&OP as a balancing act

Demand > Supply Demand < Supply

Customer Service suffers (out of stock) Inventory increases
- It is a short term planning.
- Medium – long term service
Costs rise Production rates are cut
Quality often gets lost in the drain due to short- Profit margins get squeezed by overheads
term pressure

I have to analyse if I have enough stuff to take care the next months.

CONTINUER

2.13. Sales & Operation Planning: the junction box

I have to accurate the rolling of the demand payment by category, by region, by countries and I have
to estimate the future demand of sales.

The supply has to have infinity proposition.

If I will prepare an overestimating I will forecast to must and I will have an overstock.

Demand forecast: the supply chain/production/…. They will provide some information (maximium
transportation capacity, …)

La bulle du milieu: after the meeting we obtain a physical capacity, in the medium long term we will
not have enough capacity to provide the others.

Operation plan: what I have to produce by category, by months

Demand plan: visible operation plan

The financial plan is the hole process, from the fabrication until the payment.

2.14. Sales & Operation Planning: overview

In the first part of the month, we need to plan the maximal capacity (how many mozzarellas can I make
with Xkg of cheese) - > rouge

If we order an item on the internet and that they do not have a stock, they can take it from a physical
store. That’s why we have to follow this daily. -> bleu

The financial marketing target will tell us what’s the maximal quantity that we can have (in term of
money). -> vert

We have to detail a medium – long term term.

The final executive is the process that the boss will decide if sth is wrong.

2.15. Sales & Operation Planning: the process

IMAGE

I generate some outputs to be present at the end of the meeting (alternative demand senarios,
financial information, …) We have to win the statement but we never know. If I don’t won the
statement, there will be no items.

The demand planning is what a machine will understand or do.

The first meeting without executive, we will discuss the reconciliation of plans.

Image

Scp are the constrained of the capacity

Every month, we will have to organize a meeting to check and control everything.

2.16. Traditional vs. S&OP-driven planning

There is no charing information between sales planning

!! les 3 premieres c’est du Middle – long term

Les 3 dernières c’est du Small – middle term

All the companies are structured such as production, …

2.17. Volume vs. Mix

In the last case on the right, on parle de stock to store. The customer orders are B2B.

The forecasting plan is variant. All the decision about S&OP have always to be transform in production

2.18. How to model capacity constraints?

Image

2.19. The S&OP planning horizon and granularity

Image

This is an example.
Mth + 1 and mth +2 are frozen moment. The increasing capacity can go on before knowing the plans
and the delay of the decisions.

2.20. THE S&OP CYCLE

Image

If the demand increase, we have to check if we have enough capacity to ensure this augmentation.

If I change sth to the sales plans, we will have changes with the capacity.

2.21. THE S&OP DISTINCTIVE FEATURES

Involves general management, marketing, sales, operations, finance, promotions and product
development. The supply chain is a function related between marketing, logistics, etiology, …

Performed at aggregate level (Product Group/family)

Includes Marketing Plan, Sales Plan, Promotion Plan, Production and Purchase Plan, Inventory Plan,
Order Backlog. DEMANDER DES EXPLICATIONS
The Planning Horizon should be sufficient to plan Labor, Equipment, Facilities, Material, and
Financials required to accomplish the Operations Plans.

S&OP’s output drives Departmental Plans

S&OP bridges business strategy with business planning

2.22. THE S&OP CRITICAL SUCCESS FACTORS

a) Ongoing, Routine S&OP Meetings
Routine meetings on periodic basis (from quarterly to monthly). At least 3 meetings:
1. Establish unconstrained demand plan & forecast
2. Rough-cut supply plan & constrained demand plan
3. Fine tuning and alignment demand & supply plans
b) Structured Meeting Agenda
Fixed Agenda with a pre-specified amount of time (e.g. 2 – 4 hours meeting), in order to:
1. Review of how well previous plans were met and perform a root cause analysis of plans
variance
2. Discussion to align of demand side plans (Mktg, Sales) with supply side
3. Closure of the meetings: single number planning environment
c) Pre-work to support Meeting inputs
 Baseline forecast, budget, orders, plant capacities, on-hand inventories.
 Data should be aggregated and translated for management
 Demand plan should include known impacts on future demand
 Supply plan should include known capacity problems, future shut-down, etc.
 Also external information should be brought into the process
d) Cross-functional participation
It is required an active participation during the meetings
 Demand side managers: Sales, Customer Service, Marketing
 Supply side managers: Manufacturing, Logistics, Procurement, Supply Chain
  Finance is involved to marry the operations plans with financial objectives
Participants must be empowered by the executives to make decisions
e) S&OP Management
The S&OP needs to be organized and run by a responsible organization, in order to:
 Schedule meetings, Setting the agenda, Ensure pre and post meetings work
 S&OP manager must not dominate, rather drive to consensus
f) Collaborative process, leading to consensus
As the output of the S&OP is a consensus-based plan, a collaborative process is needed
 Every stakeholder must be able to quickly create, review and revise plans
g) Joint Demand & Supply Planning to ensure balance
 A common mistake: to presume that sales & marketing plans are given and inflexible
Problem 1: Sales & Marketing has no role to play at the meetings
Problem 2: Missing the opportunity to exploit excess in supply capacity
 Therefore, also Marketing and Sales plans are rough-cut plans to be revised during
S&OP
h) Process measurement
Like any other process S&OP should be measured so it can be improved over time
 Operational KPIs: performance measures: demand forecast accuracy, variance to
baseline and budget, customer order backlogs, plant utilization, etc.
 Financial KPIs: SC costs and financial index (gross margin, cash-to-cash cycle, …)
i) Support of Integrated Demand vs. Supply Tool
 A common problem: S&OP is supported by a myriad of asynchronous spreadsheets
Step 1: implement a demand forecasting tool, since the main input of the process is
the baseline forecast (that should be generated by statistical tools, to keep it unbiased
and unconstrained)
Step 2: integrate supply-side tools with demand side ones

Séance 3.

There are 3 aggregation dimensions!

Mid-term model is also for budgeting.

Why I to forecast and then I have to decide for the rest (production, …)
It presumes that demand for a product follows a S-shaped curve growing slowy in the early stages,
achieving rapid and sustained growth in the middle stages, and slowing again in the mature stage
If an item has a pattern, we want that the product life cycle is matching. If the life of a product is easy
to forecast, this is better because we know until where we are going.

We need to have an historical number of sale in order to be able to forecast.

The ARIMA model will not be explained.

Sales and Operations Planning includes Supply Chain processes related to the planning of Sales,
Production and Distribution activities across Supply Chain Network multi-level

The Sales and Operations Planning processes:

- The Medium term process

The Demand planning, Material and Capacity Requirments Planning (MRP/CRP), Distribution
Requirements Planning (DRP)
- Short term process
Operations Scheduling (job loading, allocation and sequencing), Transportation Scheduling
(vehicle loading and routing)

The objective of the macro-process called S&OP (Sales & Operations Planning) is the generation of
feasible operation plans, using in an optimal way the available business resources (production,
distribution and inventory capacities) to satisfy the market demand for the company’s finished
goods.

Analyzing :

a) Availability of resources along time periods of the planning horizon

b) Sales demand by product, country / customer, period
The S&OP process aims to determine:

- Demand plans: trade promotions, sales plan, …

- Supply plans : production plan, purchasing plan, logistics transportation plan

The sales pan is also called MFR.

Lorsqu’il y a un problème de prévisions, cela peut

entrainer une perte de clients. Maintenant, si le reste
de la prévision était correcte, le problème devrait être
vite résolu.

L’équation est souvent utilisée dans la vie de tous les jours.

The importance of demand planning process:

The critical factors related to products

- Number of SKUs
- Product mix (assortment planning)
- Product life cycle
- Marketing actions (promotion planning)
- Product shelf life (ex: seasonal collection)
- Critical components purchasing lead time

The critical factors related to markets

- Number of B2C / B2B customers

- Customer service level
- Competitors’ commercial actions
- Delivery lead times
- Sales channels behavior (retail, wholesale)
The sell-out:

The final consumer demand (client), estimated by POS manager and the final B2B customer demand
(manufacturing company), which purchases components / WIP from a manufacturer (supplier), for
assembling then components into more complex finished goods.

The sell-in:

- The B2C -> POS demand, estimated and filtered and modified by distributor (wholesaler
modifies the retail Sell-Out demand, adding safety stock and following suppliers’ price
discounts)
- The B2B -> finished products demand, estimated by manufacturer (supplier)

The differences between Sell-In and Sell-Out:

- For the manufacturer -> the distributor: additional stock (pricing), safety stock
- For the distributor -> Point of Sales: local promotions (ex: discounted price)
- Sell-Out: forecast made by distributors, store managers, …
- Sell-In: demand plan generated by distributor (client) to the manufacturer
- The manufacturer estimates Sell-In demand (Sell-In forecast)
- Safety Stock: 10% of Sell-Out demand

The demand forecasting is super important

Under-forecasting

- Customer service level reduction

Potential periods measuring stockouts
- Safety stock level increased
- Continuous review of demand & supply plans
- Reduced sales and market quota

Over-forecasting

- Inventory holding costs increased

- End of life / shelf life (physical or technology obsolescence)
- Capacity allocation

The underforecasting:

The actual sales have been underestimated by 20%, in all the past months

The actual service level: 80% (240 pieces lost)

The overforecasting:

The actual customer demand (actual sales) has been overestimated by 50%, in all the historical
months

The actual service level: 100% (0 lost sales units)

The overstock final level: 600 pieces; 50 pieces on averages

THE DEMAND PLANS

The sales forecast:

- The quantitative forecast of the future demand

Typically prepared using quantitative Time Series methods

The sales plan:

- It defines the commercial plan for the sales; it involves different sub-plans:
Statistical forecasting (Sales Forecast)
Marketing actions (Trade Promotion Management)
Sales Plan prepared by sales rep or channel managers

The demand plan:

- It is the Sales Plan «feasible», that is: constrained to the distribution and production
constraints (Supply Planning)
Production, distribution, storage, supply constraints

THE SALES FORECAST VS. SALES PLAN

The sales forecast:

- Quantitative forecast of the future demand

Typically prepared using quantitative Time Series methods

The sales plan:

- It is the integration (generated during commercial meetings) of:

Statistical forecasting (Sales Forecast)
Marketing actions (Trade Promotion Management)
Commercial orders current portfolio
Financial budget constraints

The typical differences:

Slow movers / obsolete vs. best seller / class A products

Old vs. new innovative items
Forecast vs. Order Netting, salesforce constraints
Promotion actions (increase of the forecast baseline)
 Top Management & Finance inputs (revenue target, profitability)

Example 1

 The sales plan SP is:

Example 2

Comments:

Maximum value, by period, between orders OA and forecast SF. Delta Promo expressed into sales
quantities (not percentages). The user (sales manager) can correct the forecast plan (mix of statistical
forecast, orders, promotions). SKU1 / Jan05 → estimated effect of cannibalization on SKU1 of promo
action on SKU2 / Jan05. Management intervention on SKU2 / Mar05: sales reduction due to
obsolescence (end of lifecycle). Management intervention on SKU1 / Mar05: sales increase due to
half-year budget constraint (new promo action to be defined for reaching the expected budget). KPI:
punctual difference between forecast SF / Sales Plan SP

The lesson learned:

The calculation of the Sales Forecast has to be done before, from a Demand Planning workflow point
of view, the Sales Plan definition. Sales Forecast is the first input to the Sales Plan, it defines its
volume size and its time profile. Adding to the Sales Forecast the delta promotional and netting the
customer orders currently in the Order Portfolio, we obtain the plan to submit to the Management
approval, delivering as output the Sales Plan. Often and incorrectly, companies define Sales Plan in
an independent way (and previously) from the Sales Forecast, not leveraging the powerful
information value added inside the statistical forecasting plan

SALES PLAN VS. DEMAND PLAN

Demand Plan :

- It is the output of the collaborative process of negotiation and balancing of the Sales Plan
between “Demand” and “Supply” funtions.
 - The Sales Plan is submitted to the feasibility analysis with the “Supply” functions (production,
distribution), comparing:
Demand: sales plan (company profitability)
Supply: production capacity constraints, inventory capacity, distribution / transportation
capacity, purchasing budget constraints
- The Demand Plan (constraint plan) is production and logistics feasible, satisfying all the
“supply” constraints; it represents the official final plan for internal diffusion to the company
functions and to the external Supply Chain partners

Example:

Demand planning process

a) Demand analytics

The hierarchical analysis OLAP-based (on line analytical processing) / historical sales analysis

The sales analysis for performance evaluation

The forecast accuracy calculation

b) Sales forecasting
 The sales cleaning: outliers’ removal from historical demand patterns

The best fit on parameters (holt-winters)

The sales forecast generation

c) Marketing intelligence: products

Actions on products:

It defines the marketing and development plan on products, defines promotions on slow-movers
low-sales products, discounting sales prices, defines conjoint offering of couples of groups of articles,
with the same packaging, sold during special events or marketing events, update the price lists for
finished goods, defines marketing campaigns on specific product groups or lines, using proper
multimedia advertising channels, prepares commercial catalogs to be distributed on POS, accurately
projecting the layout of catalog or Web pages and defines location of products on POS shelf (store
layout planning)

d) Marketing intelligence: customers

Actions on market:

It evaluates new opportunities to open new stores / POS at new geographic areas, evaluates sales
perspectives on potential prospects, defines new promo actions (contents, duration, presentation &
communication, economic parameters of the offer) on item/customer clusters, introduction of new
item lines on specific target markets, project of commercial relationship of collaboration with single
customers (wholesalers B2B), integrating logistic plan (Vendor Managed Inventory, CPFR) and
definition of the contact way with consumers in B2C (web sites, mailing list, multimedia advertising,
call centers for customer care activities)

e) Demand intelligence

The data mining:

- Clustering (on groups of items or customers / stores)

- Sequence clustering
- Classification (ABC, promo, …)
- Association rules (market basket analysis)
The role of demand planner into the organization

The typical tasks:

- Forecasting ownership (forecast champion)

- Input data (quantitative, qualitative) research
- Demand analytic (KPI monitoring)
- Preparation and control of Demand Plans
- Human interface to company’s key users and external business experts
- Organize periodic S&OP / DP meetings
- Collaborative inter-functional negotiation
- Distribution of demand plans to the functions

The key competencies

- Math – statistics: knowledge of Sales Forecasting / Data Mining algorithms and big data
analysis methods
- Relational – communication
- Organization: managers during the meetings, identification of best functional resources
(forecast group definition)
- Business knowledge: expertise on markets where company operates

Times series

A time series is a sequence of measurable data (orders, euro, kg, litre) points (D1, D2, D3, … , Dt, …),
obtained typically at successive times, spaced at (often uniform) time intervals (days, weeks, months,
quarters, years, …)

The time-series forecasting techniques

How? Multiple methodologies’ usage

Time series analysis

The objectives of the Time Series analysis:

- Outliers detection
- Historical series component analysis
 Basic typologies
o Continuous series
o Sporadic series
 Components
o Seasonality (default length: 1 year, 1 sales cycle)
o Trend (linear, non linear)
o Cyclicity (long term trend)
o White noise (erratic unforecastable component)
The demand components

Df = f (Tt, St, Ct) + εt

Systematic Random
Part Part

Before start forecasting for the future it is necessary to analyse historical data to identify

Trend and seasonality

a) Without seasonality

b) With seasonality

L: seasonal time-lag
Demand components: seasonality

FORECASTABILITY: DECISION TREE ABOUT PATTERNS

How easy is to forecast a historical series?

Which methods have to be used?

First check: new items vs. permanent items

Second check: continuous series vs. sporadic series

Third check: regular patterns on continuous series

o Zero density δZ = nz
nz + nv

If a historical series contains at least 50% zeroes, it can be considered having a sporadic
behavior;
It has to be performed, in addition, a control about the shape of the series. Example: items
sold or promoted into specific periods of the year (sport events, racing competition, etc.) are
characterized by high sales «in season» and zero sales «out of season» → they must be
processed using Time Series forecasting methods (removing «out of season» periods with no
sales)

o Variation factor CVD = σD = ( T * Σ (Dt – 1/T Σ Dt)2 )1/2

μD Σ Dt
historical sales typologies

We distinguish at least 2 different typologies of demand patterns:

The sales forecasting: forecast accuracy

Some historical series intervals

- Training set
- Test set (forecast calculated over the past) Et = Dt – Ft
- Forecast horizon (in the future)

Measuring forecast accuracy

Forecast error in period t is the difference between actual demand and forecast for that period

Et = Dt – Ft

The forecast accuracy: KPIs

The time extension:

- Punctual measures (calculated only on one time bucket)

- Global measures (calculated over a time interval)
The units of measures:

- Absolute measures (in quantity or values)

- Relative measures (percentages)

The usage:

- Parametric Best Fit (choice of best parameters in Holt-Winters)

- Model Best Fit (choice of best algorithm)

Measuring forecast accuracy

ME: Mean Error

ME = Σ Et
n

Indicate BIAS:

- ME < 0 : Average Demand < Average Forecast

- ME > 0 : Average Demand > Average Forecast

It allows positive and negative errors to partially offset each other

MAD: Mean Absolute Deviation

It measures absolute (without regard to sign) error consistency

Opposite errors do not compensate each other

It doesn’t show error’s correlation

MAPE: Mean Absolute Percentage Error

It allows to compare different series on a % scale

It penalizes more errors in periods with a little demand

It doesn’t work with values of demand

The forecast accuracy: mape calculation

SDE: standard deviation of Errors

By squaring the individual error SDE penalizes more high absolute errors than small errors

It refers to n values (n-1 represents the number of data independent from each other)

Often used for estimating safety stock

Forecast accuracy: which error metric to use?

For measuring forecast error it is correct to use:

- Mape: mean absolute percentage error

For comparing performances among SKUs, countries, etc.
To immediately understand how far is your forecast from actual
- MPE: mean percentage error
For highlighting the percentage magnitude of BIAS error (underestimation or overestimation
of demand)
This KPI has to be analyzed secondly, in case of high value of MAPE

MAPE and MPE are not the proper metrics when you want to measure forecast error in presence of
sporadic series (i.e.: series having majority of zero values)

Setup of the threshold values for MAPE (alerting)

- Based on vertical industry benchmarks

Example: in the Food @ Beverage market, a sales forecast over a couple SKUL / all (country) /
month having a MAPE not greater than 10% can be considered (almost) excellent (best in
class)
- Based on continuous improvement
To reduce progressively the threshold value of acceptance for the forecast (25%, 20%, 15%,
…)
- Based on ABC classification (pareto on revenue)
MAPE < 15% -> class 1 items
 MAPE between 15% and 25% -> class B items
MAPE between 25% and 35% -> class C items

Sales cleaning for continuous items

Sales cleaning: general workflow

Sales cleaning: steps

a) Periods: identify which periods have to be analyzed and, eventually cleaned

b) Algorithms: define the parameters for setting-up the chosen cleaning method
Moving averages -> length of average
 Single exponential smoothing -> alpha parameter for smoothing outliers
Confidence interval -> number of standard deviation
c) Pre-cleaning: remove seasonal & trend components to the historical series (still to be
cleaned)
d) Cleaning
Certify whether a value has to be cleaned or not
Clean the certified outlier
e) Post-cleaning: rebuild seasonal & trend components to the historical baseline (already
cleaned)

Sales cleaning: confidence interval

Two-step method:

- Cleaning of non promo periods

 The historical series S1 to be cleaned in phase 1 is composed by only the non promo
historical periods (the promo periods are temporary removed from historical series)
 Phase 1 Sales Cleaning is run for removing from historical series S1 outliers related
to pas stock out, outliers characterized by abnormal sales (big customer orders),
statistical “white noise” (all the outliers that are not linked to past promo events)
- Cleaning of promo periods
  In pahse 2, the historical cleaned S1 series is relinked to the series of non promo
periods (still not cleaned), rebuilding an historical series having only non promo
periods already processed / cleaned
 Phase 2 Sales Cleaning applies only over promo past periods, to remove the only
effect of trade promotions (which can be now better identified from a statistical
point of view, after Phase 1 application)

Typically, we can give Kmin a value close to 0.

Kmax must generally beteween [0,5 ; 1,5]

INSERER PHOTO

Valable interval = dynamic interval.

INSERER PHOTO.

Phase 1: from “Historical Demand” series (blue) statistical no promo outliers are removed, obtaining
“(1) No promo Cleaning” (red) output series;

Phase 2: from “historical demand” (only promo periods), in union with “(1) No promo Cleaning” (red)
(promo period already cleaned), we obtain the final historical baseline series “(2) Promo cleaning”
(green)
Pour les 3 premières semaines, I apply the static approach but pour calculer le global average, j’utilise
le static approach. J’espère que sigma ne sera pas aussi affecter par le nombre de suppliers. On peut
se dire qu’à partir de la 4e semaine, on utilize la dynamic approach. Peut-être parce que les 3 premières
semaines, c’était des données plus ou moins stables.

En pratique, c’est un exercice d’un hystorical approach.

Forecastability: historical series

en janvier 60 et 70, le system reconnait a local system.

THE SALES FORECASTING EXPONENTIAL SMOOTHING MODELS

The mathematical models structure:

- Moving average models

- Exponential smoothing models (single, double, Holt-Winters)
- Time series decomposition models (ask an exercice on the multiplicative approach)
- Sporadic series models (we don’t search about sesonality)
- New item introduction models (patterns and L4L)
We suppose that in multiplical model:

Dt = Tt * St * Ct + εt because Ct = 1

INSERER PHOTO

In this case you have 12 seasonal indices, one for each seasonal cycle

Sjan, Sfeb, Smar, …., Sdec one for each month

It means that in November sales are 25,8% higher on average than the average value of monthly
sales

St = Dt / Mt

Si on calcule the average de 2002 et de 2003, nous avons 468 et 467. La premiere seasonal value, est
calculé par le ratio de la demande de janvier 2002: Djanv2002 / M2002 = 155/468

Sjanv = ½ * (Sjanv02 + Sjanv03) = ½ * (0,331 + 0,381) = 0,356

Plotting seasonal indices, you obtain a seasonality graph.

THE TREND COMPONENT

You have to evaluate the monthly time series below:

We want to find a trend line an historical series. TRENDt = a + b * t

The independent variable is the time axes (1, 2, 3, ….)

Every month my growth is 12%. It is a parabolic trend. It is more than linear. A parabolic is a fixed
percentage.

THE SERIES ANALYSIS (TREND)

With regression analysis it is possible to identify and evaluate the characteristic of trend.
It is necessary to identify the theoretical function y=f(t) (straight line, parabola, …) which better fits
historical data series.

Assuming a linear type of trend, it is necessary to determine the values of regression line’s coefficient.

In this case it results:

For excel:

INSERER PHOTO

The time bucket: t coincides with how often the plan in updated

The actual demand in period t: Dt

The forecast about period t+m mode at the end of the period t: Ft+m

THE SIMPLE MOVING AVERAGE

Given a time series of “N” values (Dt, Dt-1, Dt-2, …), at the end of the period t you can calculate the
value of moving average (“k” is the moving average time lag):

MAt(k) = Dt + Dt-1 + Dt-2 + … + Dt-k+1

k
 All past data are given an equal weight (1/k)

You apply a moving average (time-lag k=4) to the time series

- In the month t=4: MA4(4) = D4 + D3 + D2 + D1

4

- In the month t=5: MA5(4) = D5 + D4 + D3 + D2

4

- In the month t=6: MA6(4) = D6 + D5 + D4 + D3

4

SIMPLE MOVING AVERAGE

Ending to the last available period you obtain “N-k+1” values for the moving average

Alternatively, with k=6 moving average you obtain 19 values

The irregularity of the time series can be filtered by moving average

Numerically you obtain:

Note: if time-lag “k” of the moving average increases, filtering increases, but you lose a large number
of data.

If the time series is stationary and non-seasonal, moving average can be used to forecast the future
demand.

Ft+1 = MAt(k)

FORECASTING MODEL

With “k=6” moving average you obtain:

k = 1 -> Ft+1 = Dt
- Forecasts agree with the moving average values, but with a month time lag
- Forecast only for the first following period (m=1)
- If k=1 forecast agrees with the last available value
BROWN’S MODEL

Given demand time series D1, D2, …, Dt, t+1 period forecast is:

Ft+1 = α * Dt + (1-α) * Ft

α : smoothing coefficient (0 < α < 1)

forecast is obtained by the weighted average of actual Dt and previous forecast Ft

Example: calculate sales forecast from the time series below:

Using Brown’s model, with both alternatives: α = 0,3 and α = 0,5 adopting starting forecast value:
Finitial = 140

Proceed as below:

Carrying on the calculations it’s possible to simulate the forecasts for each month in the past 2 years
(24 months)
Proceeding in this way it is possible to generate forecast for both 2 years and for january 2004.

INSERER PHOTO (plus loin qu’elles sont)

ITERATION PROCESS

At period t: Ft+1 = α * Dt + (1-α) * Ft

At period t-1: Ft = α * Dt-1 + (1-α) * Ft-1

At period t-2: Ft-1 = α * Dt-2 + (1-α) * Ft-2

Replacing: Ft+1 = α * Dt + (1-α) * [α * Dt-1 + (1-α) * Ft-1]

Ft+1 = α * Dt + α * (1-α) * Dt-1 + (1-α)2 * Ft-1

Ft+1 = α * Dt + α * (1-α) * Dt-1 + α * (1-α)2 * Dt-2 …. + α * (1-α)t-i * Dt-i

INSERER PHOTO

INSERER PHOTO

SINGLE EXPONENTIAL SMOOTHING

Brown’s model (simple smoothing) is suitable for data with no trend or seasonal patters

Belongs to interval {Dt ; Ft}

Needs only two data f {Dt ; Ft}

Forecast Ft+1
Holds all historical data (Dt, Dt-1, …, D1) weighted with weighted with
decreasing values following a negative exponential function

α value influences model’s reactivity

SMOOTHING COEFFICIENT

α high: reactive model (> weight to new data)

α low: static model (> weight to past)

SERIES DECOMPOSITION MULTIPLICATIVE MODEL

INSERER PHOTO

INSERER PHOTO
3. DEMAND PLANNING AND FORECASTING

MULTIPLICATIVE SERIES DECOMPOSITION: THE ALGORITHM

Decompose an historical series into trend / seasonality / cyclicity components (multiplicative model):
a) Calculate the conjoint component of trend T and cyclicity C by applying a centered moving
average (CMA) algorithm, given L step equal to seasonality lenght:
Tt * Ct = CMAt
b) Calculate the seasonal S component using multiplicative seasonality coefficients:

c) Calculate the mean seasonality coefficients: removal from historical series of random
fluctuations component et to isolate seasonality, averaging St  et values over homologue
seasonal periods → output: SMt

INSERER PHOTO

d) Deseasonalize the historical series values by dividing each demand value by the
corresponding seasonal coefficient SMt:

e) Calculate the linear trend component identifying a linear regression trend line interpolating
deseasonalized values across time buckets t (t=1, 2, …, T):

f) Calculate Sales Forecast over the past and the future horizon (multiplicative model):
seasonality (seasonality vector SMt, step 3) multiplied by the trend component (trend line Tt,
step 5)

Example: apply multiplicative series decomposition model (MSD) to the quarterly sales of an item, as
shown in the following table

A’) Apply a CMA algorithm of step k=4 (to remove seasonality one annual basis)
 INSERER PHOTO

B’ + C’) Calculate seasonal coefficients for each quarter (mean values to eliminate irregular white
noise)

INSERER PHOTO

It is a very long vector. How many data points ? 4 years minus one

D’) Calculate deseasonalized historical series values

INSERER PHOTO

E’) Calculate the trend line on deseasonalized demand values

INSERER PHOTO

Sales forecast & non linear trends

- Non linear trend models

Increasing trend, less than linear

Decreasing trend, less than linear

Demand planning for Slow Moving items

Sales patterns:

Sales events are rare and completely casual, not correlated among them. Sales orders are for unit
volumes/few quantities; they occur rarely in the planning horizon.

High-tech manufacturing context:

Maintenance Repair and Overhaul (MRO) and spare parts

Methods:

Croston -> forecasting algorithm

Poisson -> replenishment algorithm

Hypothesis on demand patterns

Demand patterns are irregular, having low values, in a timely sporadic way, without causal
relationships. If an item is sporadic in the past → it will remain sporadic also in the future planning
horizon (Forecasting hypothesis). The variability range for sporadic series in manufacturing retail is
from 0 to 10 pieces per period

Example of series: {0 0 0 1 0 0 0 0 1 1 0 0 0 0 2 0 0 1 1 0 1 0} → no trend and no seasonality can be

«identified»
CROSTON’S MODEL

Introduction

It is a method developed in 1972. It is used for time series forecasting model for sporadic items. It
provides answer to the following questions:

- Which is the next positive demand volume?

- When next positive sales will occur?

Algorithm

The time series is composed by two independent part, to be separately forecasted.

- Historical series of positive volumes

- Series of time interval between consecutive positive values of sales

How Croston’s algorithm works

Given the series of positive sales, we calculate the volume Q of the future demand (a unique value Q
for each sku – warehouse couple), applying forecasting algorithms such as simple moving average
(MA) or single exponential smoothing (SES) over the sequence of positive historical values only.

Given the series of time interval between 2 consecutive sales (integer number of time periods), we
calculate the forecast of the next «average» interval lenght T applying forecasting algorithms such as
simple moving average (MA) or single exponential smoothing (SES) over the sequence of delta
intervals collected into the historical series

Example: Historical demand: {10 0 0 0 4 0 0 2 0 0 0 0 0 3 0 2 0 0 1}

- Positive volume series: {10 4 2 3 2 1}; average: 22/6 = 3,7 = Q

- Delta interval series: {4, 3, 6, 2, 3}; average: 18/5 = 3,6 = T

We put the future fixed forecasting volume Q

- Rounded (→ if sales are discrete volumes)

- Not rounded (→ otherwise)

in a regular «semi-periodic» way (following the lenght of the average interval T), both over historical
periods and over future periods of the planning horizon:

- Starting from the oldest positive historical value of demand (over which we put – by default –
the first future demand volume Q);
- Calculating – at every step – where positioning the next Q value in the time axis, using a
simple algorithm «rounded integer»

Example:
Historical monthly demand: {10 0 0 0 4 0 0 2 0 0 0 0 0 3 0 2 0 0 1}
Delta interval series: {4, 3, 6, 2, 3}; average : 18/5 = T = 3,6 months
Demand forecast: {0 0 Q 0 0 0 Q 0 0 0 0 Q …}; in the example: Q = 22/6 = 3,7 units (rounded at 4
units)
Where to place Q volumes:

- Step 0 : Q is placed on historical volume 10 (the oldest historical value)

- Step 1 : int(3,6) = 4 months (→ Q is placed on historical volume 4)
- Step 2 : int(3,6–0,4) = int(3,2) = 3 (→ placed on historical volume 2);
- Step 3 : int(3,6+0,2) = int(3,8)= 4 (→ placed on ninth 0)
- Step 4 : int(3,6-0,2) = int(3,4) = 3 (→ placed on eleventh 0)
- Step 5 : int(3,6+0,4) = int(4) = 4 (→ placed on historical volume 1)
We proceed this way also for all future months

int(…) = nearest integer

Example:

Historical volumes: October 2004 to September 2005

Future months: October 2005 to September 2006

Inventory planning for slow Movers items: Poisson’s model

For solving the manufacturing retail planning problem for spare parts we use the statistical
distribution of Poisson

Pk = Probability that the future demand during one lead time (LT) will be equal to k pieces (k = 0, 1, 2,
...)
DLT = Average historical demand during one lead time (LT) : DLT > 0

Poisson’s distribution: numerical histograms

Lead Time: is the delivery time to the «target» location (Wartsila central warehouse), that is: the time
interval between two consecutive deliveries

D(LT) is calculated as average of the historical demand (after aggregating historical sales for each
historical LT), including into the average calculation also periods having zero sales

Poisson’s distribution for demand forecasting?

- Using it as a Sales Forecasting model, the Poisson’s distribution provides only a probabilistic
range about future sales quantity (ex: the probability to sell 3 pieces in the next LT = 2 weeks
is 87%)
- This info is only useful for strategic planning scenario analysis (long term planning models)

Poisson methodology steps

We assign a target service level SL for each SKU – customer or SKU – warehouse combination, via the
Service Level matrix grid (ex: 99,5%, 98%, 95%, etc.)

We calculate the INVMIN minimum stock level to hold in warehouse to satisfy the required service
level SL, supposing that the combination follows (in the past sales but also in the future horizon) a
distribution very close to the Poisson’s one

Which is the minimum stock level of pieces (INVMIN) to hold in stock to have a «success
probability» (→ not to have stock-out at all during the lead time LT) at least equal to the target
service level?

The INVMIN calculation is done by cumulating each success probability / event. Example: if I hold
INVMIN = 2 pieces on hand, there’s no stock-out is the actual demand is either 0, or 1, or 2.

If we supposed to have, at the beginning of each delivery lead time, a future demand equal to
INVMIN level,

- for each lead time interval in which the planning horizon can be divided, starting from
«today» (time now) (ex: 6 different lead times of 3 weeks, in a planning horizon lasting 18
weeks)
and if we replenish, at the beginning of each consecutive lead time, the stock needed to satisfy the
INVMIN demand,
we will guarantee the target service level SL, regardless the specific time period (day, week) internal
to the lead time during which the sporadic demand will fall.

From the classification of Replenishment methods point of view (warehouse replenishment), the
Poisson’s algorithm can be considered as min-max replenishment methodology: that is,

- if the current stock is lower than INVMIN,

- we need to reorder a variable quantity to cover a target stock equal to INVMAX

where: INVMIN = INVMAX = minimum quantity output of the Poisson’s algorithm

- Minimum stock level INVmin → if current stock is lower than INVMIN we launch a new
purchase order;
- Maximum stock level INVmax → stock level to reach when preparing a new purchase order

(INV is the on hand stock, at the beginning of each lead time)

Example 1

- if INV = 2 and INVMIN = 3 → replenish 1 piece

- if INV = 0 and INVMIN = 3 → replenish 3 pieces
- if INV = 5 and INVMIN = 3 → not to replenish (and recalculate the stock projection at the end
of the next lead time)

Example 2

- if INV = 0 and INVMIN = 2 → replenish 2 pieces

- if INV = 1 and INVMIN = 2 → replenish 1 piece
- if INV = 2, 3, 4, … and INVMIN = 2 → not to replenish

Safety stock calculation in Poisson’s algorithm

The safety stock calculation is «implicit» inside the Poisson’s method for calculating the INVMIN
stock level. The safety stock is the difference between the two values:

- INVMIN Poisson’s level (both target stock level and reorder point)
- Average historical demand during the delivery lead time

Example

- if average demand D(LT) = 2,4 pieces and INVMIN = 3 pieces

- then the safety stock is equal to: SS = INVMIN – D(LT) = (3 – 2,4) = 0,6 pieces

Exercice:

Based on historical sales data of the last years, we calculate that the average demand for the item
“Alpha 0101R65” into the 0101 warehouse has been equal to 1,2 pieces/week. Supposing that the
delivery lead time is equal to 2 weeks, and the required SL is 75%, calculate:

- The probability that the item “Alpha 0101R65” will be sold into the next 2 weeks
- The probability not to have stock-out supposing to have on stock (on hand) 3 pieces of the
item
 - The expected number of stock out pieces in 1 year (supposing that the order lot size Q is
equal to 3 pieces)
- Supposing an annual inventory holding cost equal to 25% and a unit stock-out cost equal to
the 40% of the product selling price (250 €), calculate, the inventory management global
costs for the replenishment configuration

Step 1: we need to reconduct the average weekly demand value to the average demand during the
lead time (lasting 2 weeks): 1,2 pieces/week  2,4 pieces/2 weeks

Step 2: we calculate the Poisson’s distribution values for different future possible demand values k =
0, 1, 2, 3, …:

Poisson algorithm is equivalent where min is equal to max if it is equal to Q

Si nous arrivons à la quantité minimum, il faut recommander du stock et e pas attendre d’arriver à
It !!

We obtain in output the following Poisson’s probability distribution for the future demand values k =
0, 1, 2, …

We can also evaluate the stock coverage demand level: if we keep on hand 3 pieces of leather goods
item “Alpha 0101R65”, the probability not to have stock-out during the next lead time (2 weeks in
the planning horizon) is equal to 77,8%:

SS75 = Qp – DLT = 0.6

SS80 = Qp – DLT = 1.6

SS93 = Qp – DLT = 2.6

Week 1 2 3 4 5 6 7 8 9 10
= RP
Sales 2 0 1 2 0 1 0 1 0 0
1. P(0)=
2. No more than 2 pieces
P0 + P1 + P2 = 96.6%
Faire avec la formule pour 0 au minimum
P0 = 49.6%
P1 = 34.7%
P2 = 12.2%
3. At least 2 pieces
P2 + P3 + P4 + P5 + … Pinfini = 100% - [P0 – P1] = 100% - 84.4% = 15.6%
4. To have stock out
I0 = 1 piece
Same question as the previous one. 15.6%
5. 92% that guarantee the customer
CSL = 92% -> Qp = TSL = ?
0 = 49.6%
1 = 34.7%
2 = 12.2%
P0 = 49.6% < 92%
P1 + P0 = 84.4% < 92%
P0 + P1 + P2 = 96.6% > 92%
 2 pieces
6. Qp = 2 pieces/RP <- CSL = 92%
SS% = ?
SS% = SS / Qp = (Qp – DLT)/ Qp = (2 – 0.7) / 2 = 65%

Example

Historical demand: {10 0 0 0 100 0 0 2 0 0 0 0 0 250 0 50}

Delivery lot size (or production lot): 50 units

The manufacturing company delivers:

- to some customers: multiple lots

- to other customers: any volumes (also small / unit lots) -> small

Solution modeling:

- Distinguish sales between «big» and «small / medium» customers

- «Small / medium» customers are managed following the standard Croston’s / Poisson’s
algorithm, without pre-processing the numerical volumes of historical demand
- For «big» customers, after nullifying small quantities eventually included into the historical
series (i.e. whose value is strongly less than multiple lot size), we divide the total «big»
quantities by the multiple delivery lot size, obtaining a new historical series that is sporadic
with small positive (integer) values → in the example: {0 0 0 0 2 (100/50) 0 0 0 0 0 0 0 0
5(250/50) 0 1(50/50)}

NEW ITEM PLANNING

Problem definition
- To forecast the future sales of new items, not having for them – by definition – historical
sales available
- To forecast rapid changes in demand patterns

Application fields

- Pre-season forecasting in the sector fashion / retail (sales of the first launch of a brand new
product)
- New models introduction in high-tech (consumer electronics, information technology),
becoming rapidly obsolete
- In general: all the items having a limited lifecycle (at maximum, some months of lifecycle,
only one sales “season”)
- Changes in market behavior (crisis, consumer spending review, fashion)

Product lifecycle curves / patterns

- One-seasonal product
Phase 1: launch; Phase 2: increasing sales
Phase 3: maturity; Phase 4: decline
 - Multi-seasonal product
Phase 1: launch; Phase 2: increasing sales
Phase 3: stable sales (permanent product, «never out of stock»)

Lifecycle curves (percentage patterns)

Pattern matching algorithm

- Initially we build a pattern library (percentage or absolute series of numerical values with
time flag identifier: sales percentage of the first week, of the second week, etc.) based on
historical values of new item introduction occurred in the past
Example: curve XYZ – product 123 – market ijk – percentage values for the first 7 weeks of
the sales cycle: {15%, 15%, 20%, 25%, 15%, 5%, 5%}.
- The initial sales pattern for the new segment or couple product – market is chosen, manually
or using automatic criteria (ex: highest similarity, same family, same channel, …), inside the
pattern library previously built inside the Sales DataMart.

Example:

In-season budget (defined by Marketing / Sales): 15000 pieces

Seasonal campaign duration for the item NEW into the store STORE_A: four months (March to June)

Analogy forecasting (product links)

The forecast for the new couple product – client is calculated using classic Time Series algorithms,
playing as the new couple had an historical pattern of sales.
The historical sales patterns of the new couple is “created” by linking the historical series of an old
product/client to the new one.

The underlying hypothesis of this method is that the customers will perceive the new product sold in
a given market as “substitutional” (totally or partially) of an obsolete one, progressively removed
from the market offer of the company.

Single or multiple link

- Link 1:1 → 1 new product / 1 obsolete product

- Link 1:n → 1 new product / n obsolete products

Linearity of the link

- Linear: the historical sales of the old product are copied “as is” over the new item phantom
«historical sales»
- Not linear: to the historical sales of the old product, we apply multiplicative coefficients for
increasing (new product is better than old one, or better promoted) or decreasing historical
sales (the new item is only partially substitutional)

Example:

4. INVENTORY AND REPLENISHMENT PLANNING

Why to hold inventory?

a) To absorb demand uncertainty (“value’ = product availability)

 Safety stock (to face variability)
 Seasonal stock (to face lack of capacity)
Both are demand, supplier
b) To decouple different operations phases (“value” = supply efficiency)
 To keep stock is a way to allow source, make and delivery systems to work with
different rate, as it allows the different phases to be “asynchronous” and it absorbs
upstream variations.
c) To gain profit through speculation (“value” = profitability)

Inventory to gain Efficiency

Seasonal-related inventories: they aim at decoupling a strong seasonal phase (e.g. market) from a
more balanced one (e.g. production) (tradeoff: production casts – inventory carrying costs)

Chase or production

Speculation-related inventories: they aim at leveraging on low price for high quantity purchases
(trade-off: purchasing costs – inventory carrying costs)

Distribution-related inventories: they aim at leveraging on low prices for transportation (trade-off
transportation costs – inventory carrying costs)

The first input is the forecast plan (the blue lines)

Standards of measurement and calculation of values into the stock balance equation:

- Inventory INV(t) is conventionally projected at the end of each period t

- Inventory at the end of period t is equal to the inventory at the beginning of period t+1
- Replenishment R(t) – production or distribution volumes – is available to sell at the
beginning of delivery period t (“receipt” date of goods, first date of inventory holding)
- Demand Forecast F(t), composed by the mix of forecasts and orders portfolio, is due and
served during the period t, in a “uniform” shape in the sub periods forming the whole period
t
INV(t-1) + R(t) = F(t) + INV(t)

The standard stock balance equation, referred to the period t:

INV(t-1) + R(t) = F(t) + INV(t)

is not working properly if it happens the following condition:

INV(t-1) + R(t) < F(t)

that is: if the sum of global availability of item volumes is less than the sum of global «demand»,
thus less than global amount of resources required in the period t, depending on mix {forecast +
orders}.

Example: initial stock INV(t-1) = 100; replenishment R(t) = 20; forecast F(t) = 145

Two possible solutions to the issue of reduced stock availability:

a) Modeling of stockout L(t) → lost demand

INV(t-1) + R(t) = F(t) – L(t) + INV(t)
Example:
Ending stock INV(t) = 0 → physical stock is zero!
Lost demand L(t) = 145 – (100 + 20) = 25
Stock balance equation: 100 + 20 = 145 – 25 + 0

The lost demand L(t) = 25 is not subjected to backlog, thus generates stock-out, following the MTS
(Make to Stock) «pure» model: the customer demand not served on-time is definitely lost.

b) Modeling of a delivery delay B(t) -> backlog

INV(t-1) + R(t) + B(t) = F(t) + B(t-1) + INV(t)
Example:
Ending stock INV(t) = 0 -> physical stock is zero !
Backlog demand B(t) -> 145 – (100 + 20) = 25
Stock balance equation : 100 + 20 + 25 = 145 + 0

In the period t, the company has to serve two types of demand:

- The forecast F(t) [on-time customers]

- the backlog B(t-1) cumulated in the previous periods [delayed customers from previous
periods t] → backlogs can be solved in the next periods

The standard stock balance equation has to be modified:

INV(t-1) + R(t) = F(t) + INV(t) adding two components:

- the net requirements coming from safety stock SS(t), to replenish «by delta» (incremental or
decremental)
- the availability of scheduled orders receipts A(t)
INV(t-1) + A(t) + R(t) = F(t) + ΔSS(t) + INV(t)

R(t): planned orders receipt

A(t): scheduled orders receipt

The stock balance equation:

Ending inventory INV(t) represents the residual availability, not used during the period t. The ending
inventory INV(t) is composed by cycle stock, to be considered distinct from safety stock SS(t).
The final version of stock balance equation:

B(t): current «delay», generated over the forecast on period t

B(t-1): total delay, «cumulated» from previous periods
ΔSS(t): planned increment or reduction for the safety stock

An order (or line, or case) is filled if all the items ordered are available in the required quantity for the
shipment

The Fill Rate:

- is correlated with the number and typology of the items kept in the inventory (which part of
the product range) and the quantity held in stock per item
- is directly perceived by the customer if the required order cycle time is zero (e.g.:
supermarkets)
- otherwise influences the average order cycle time: the higher the fill rate the lower the
average cycle time

The emphasis on the idea of “customer served”

- orders fill rate = (fulfilled from stock / orders received) *100

the useful also for single measures:

- line fill rate = order LINES FULLFILLED FROM STOCK/ order lines received) * 100
- Case fill rate = cases fulfilled from stock/ cases received * 100

Customer service level example:

The inventory turnover ratio:

ITR = (Outgoing flow/Avg. inventory level (AIL))

 It shows how many times the inventory of an item “turns” in the time period (usually it is
calculated on yearly basis)
 The unit of measure is [Turns/period] or [1/period]
  Outgoing Flow and Average Inventory Level (AIL) can be measured as preferred: number of
pieces, cases, pallet loads, m2, m3, liters, kg, …

Some examples:

Days of supply (coverage period)

DOS = AIL / Avg. Daily Consumption

 It is the reciprocal of the ITR (=1/ITR)

 It measures the expected time of autonomy (coverage) of the warehouse in case it was not
replenished anymore
 It is equal to the average duration of stay of the goods in the warehouse

Exercises:

The inventory in the distribution system

Cycle stock= these inventories deal with the different operative rhythm of two following stages in the
supply chain

Safety stock: these inventories deal with the uncertainty of both the demand and the replenishment
lead times
In transit stock= these inventories are in transit between stocking or production points (mainly inside
the vehicles)

Cycle stock example:

- Warehouse replenishment rated different from the retrieval/consumption rates

EOQ – REORDER POINT model

a) Fixed order quantity system

How much? A fixed quantity is ordered, equal to Economic Order Quantity (EOQ)
When? Place an order just when stock level (availability) reaches a given order point (OP) ->
continuous control
Actual Stock + Qty on order – Qty allocated = Availability

The safety stock (SS) is useful to prevent uncertainty in demand and in lead time which could
generate stock-out during the Lead Time

The safety stock is calculated in order to guarantee that demand variability during LT doesn’t give a
stock-out.
The periodic review model

The uncertainty management

If the lead time and the demand are uncertain, safety stock is required in order to avoid the stock out
during T+LT

The availability target

The safety stocks

Safety stock is based on the stock cover level. It faces demand variability during LT and LT variability
as well

You have to ask your supplier the right amount of stock

The safety stocks formula

Considering a generic product, it’s possible to define the Safety Stock quantity to hold:

K = CSL. The k factor is related to the CSL. I have to

decide it.

We just have to analyze sigma. You don’t have to

choose one. It will calculate in pieces. E1, E2, E3.

How to calculate safety stock

Safety stock calculation formula:

- SS(t) = safety stock profile for a given SKU (depending on time t)

- K = target service level (ex: 95% probability not to have stock out during a replenishment
period) -> = [Link].S
- σ(t) = forecast error (depending on time t) + supplier reliability

How to calculate customer service level

Hypothesis: demand during lead time is a random variable normally distributed with mean equal to d
and standard deviation equal to σL

By defining safety stock level, the probability of stock coverage is univocally defined.

The safety stocks

a) demand system
if demand and lead time are uncertain variables following a normal distribution

You will have

An example:

The distribution requirements planning

Supply chain network for inventory & replenishment planning

The replenishment: terminology

Demand (gros requirements)

- independent demand at store level (consumers)

Net requirements

- independent demand, nettifed by on hand inventory at POS: {Gros Req - Inv} = Net Req

Replenishment

- Planned Order Receipts: quantity to send to POS, to cover the net requirements, with a Just
In Time delivery, to avoid stock out (receipts at their physical delivery date)
- Planned Order Releases: quantity to send from the “supplier” node (warehouse) to the
“customer” node (POS) (departure date)
 Release date = {Receipt Date – Distribution Lead Time}

The replenishment: simple starting example

Gross requirement

- Forecast = 100 pieces in week w1

Net Requirement

- Initial stock = 18 pieces at the beginning of week w1

- Lead time = 1 week
- Net Requirement = (100 – 18) = 82 pieces

Replenishment

- Replenishment rule: multiple lot = 15 pieces

 - Replenishment (arriving at the beginning of week w1) = 90 pieces
- Stock at the end of week w1 = 0 + 18 + 90 – 100 = 8 pieces

Distribution requirements planning

Basic hypothesis and constraints for inventory & replenishment planning

a) Single-level BOM / BOD

Replenishment of package finished products (textile or fashion) @POS
b) Infinite capacity at the warehouse / supplier (distribution warehouse / plant warehouse)
Storage capacity @warehouse
Production capacity @plant
c) No alternative sourcing
Unique supplier / central warehouse for replenishment
d) Delivery logistic constraints (distribution / warehouse)
Lot sizing (deliveries with minimum, maximum, multiple lot size)
e) Delivery lead time
The length of supply lead time determines the length of frozen period (first lead time)
f) No mix constrains (multi-item)
Minimum supplier Quantity / minimum Order Quantity
Assortment mix

Example of textile / fashion constraints related to the product mix

a) Minimum per order (per supplier / period)

 Value constraint: replenish at least 2.000€ of merchandise (item mix is free)
 Logistic constraint: replenish at least 33 pallets (1 standard vehicle) of merchandise
(item mix is free)
b) Assortment mix
 Assortment Planning: if you order a quantity Q > 0 of item A, you have also to order
at least 10% of Q of item B
 Replenish at least 1 item for each item of a family brand / group (complete mix of
colours, sizes, styles): not only best seller but also slow movers

Distribution requirements planning: steps

Project the system of inventory & replenishment planning

a) Input parameters definition

b) Replenishment algorithms selection
c) Safety stock calculation formula selection
d) Output KPIs definition (for all periods in the planning horizon)
 Replenishment plan
 Inventory plan
 Expected stock out profile
 Delayed demand profile (backlog)

Replenishment KPI: basic report

Output KPIs

- Replenishment plan in the planning horizon (Planned Orders Receipts)

  Also Planned Orders Releases
- Stock projection / profile (inventory Plan)
- Stock out profile (lost demand, blocklog)

(1) Input parameters

- Lead time, review period
- Frozen plan (past replenishment during the first lead time)
- Scheduled orders receipts (schedulati)
- Initial stock @store (on hand)
- Demand (mix forecast / customer orders portfolio)
- Inbound calendar @store (opening days)
- Transportation lot sizes (min, max, multiple)
- Demand shelf – life (backlog)

(2) Replenishment policies definition

- Pull Stock control: Min – Ma
- Push demand forecasting
 Fixed lot (EOQ), variable interval
 Fixed interval (“stock coverage days”)
 MSQ: Minimum Supplier Quantity
 Assortment mix control: replenishment multi-item and/or multi-site

Basic rule: replenishment quantities come to stores always “just in time” (at the least day), for
minimizing average projected stock profile (consequence of the replenishment algorithm used)

Pull replenishment algorithms

Minimum – maximum algorithm

When the current stock level INV(t) is lower than a minimum MIN, replenish a variable quantity, to
reach a fixed target quantity MAX (maximum stock level)

- When to replenish?
Variable interval, if INV(t) < Min;
- How much to replenish?
The variable quantity {MAX – INV(t)}
This roughly simple stock control algorithm does not take into account the future forecast demand.
This may cause stock out in future periods of the planning horizon.

Hypothesis: it is not allowed to have any replenished volume

- Calculate customer service level

The push replenishment algorithms: activation rule

How to determine the reorder point?

Replenishment quantities are requested:

- (fixed quantity: EOQ)

- (variable quantity: to serve forecast into coverage interval)

In a “just in time” way, that is when the total stock availability (total supply), composed by:

- Inventory on hand INV(t) at the end of period t

- Scheduled orders receipt at period t+1

Is less than the sum of requirements (total demand):

- Safety stock delta SS from period t to period t+1

- Demand at period t+1
 See exercises that we have done