Presentazione dello speech tenuto da Claudia Beldon
(VP - Fashion & Luxury Industry at ACT Operations Research) dal titolo "Fashion and Luxury - From sell through to risk-based management ", durante il Decision Science Forum 2019, il più importante evento italiano sulla Scienza delle Decisioni.
Fashion and Luxury - From sell through to risk-based management
1. Fashion and Luxury - From
Sell Trough to risk based
management
Claudia Beldon ACT OR – risk management in Fashion
Carlo Filippi UNIBS – an optimization model
SMASH - SMArt platform to enabling high performance
processes in faSHion's supply chains
3. AI & FASHION
The fashion and luxury industry provides a
particularly challenging environment:
• High seasonality
• High uncertainty
• High complexity
• Long lead times vs short life cycles
Therefore, the key to success is based on:
• Making better forecasts
• Taking correct decisions
• Minimizing risks
• Optimizing the processes More efficiency =
sustainability!!
4. OUR APPROACH TO
FORECASTING
Forecasting is already happening: merchandising
plans, purchasing, seasonal buying, pricing…
Not all the forecasting processes can be
engineered, but …
• What data underpin the forecast?
• Is the risk known and managed?
• Who owns the knowledge?
Process
implicit described
Risk
unknown managed
Knowledge
individual shared
5. A COMPLEX CHALLENGE
Demand forecasting is critical to businesses across almost all industries. It can
seem easy, because there are easy ways to build simple models. But in
practice, building a demand forecasting model that is accurate and useful is a
complex challenge.
6. DIFFERENT TYPES OF FORECASTING
MODELS
Each use case need a specific forecasting approach in order to obtain
the highest possible accuracy.
1. Use last month’s or last year’s demand to predict short-
term demand
2. Fit a time series on the demand signal, using trend,
seasonality, etc…
3. Include external drivers such as weather patterns, stock
market index changes, web and social network infos,...
4. Take into account complex demand interactions as
substitutions, complementary products, the reactions of
competitors, promotional campaigns, etc.…
Increasing
complexity
7. A PERSONALIZED SOLUTION
A demand forecast model is not something one
can build blindly.
The data scientist should first determine what it
will be used for: buying optimization, store
allocation, pricing optimization, etc...
A rigorous definition of the business case at
stake is needed to make sure the right
granularity, cost function, and expected accuracy
will be applied!
There is no single
demand forecast
model that will work
across different use
cases!
Where and how they
differ?
Which demand forecast
model should be used
for each?
8. FASHION RETAILERS
3/6 months is the
lead time needed
to manufacture
and ship out
goods
Seasonal product
has a planned
lifetime of
between three to
12 months
Omnichannel
distribution is the
leading pattern
S/T demand forecast
to optimize allocation
of goods and
replanishment
L/T demand forecast to
optimize order quantities
of goods
L/T demand forecast to
optimize order quantities
of goods
S/T demand forecast
to optimize allocation
of goods and
replanishment
Qualità del dato
importante;
esempio pre-sales
9. Short term Vs Long term forecasts
Season Sales Forecast
Forecast by season, by delivery window, style
(aggregation of items), geographical area, etc..
Usefull to help budget, sales campaigns and purchase
process
Output one single value that encapsulates the total
demand
Rolling Trend Forecast
Usefull for stock reallocation by sku/size/store during
the season (by week, month)
Rolling forecast as soon as new data of sales are
available
Intercept probable changes vs past trends
includes Sell-through trends in
order to account overstock and
stockout
can either use the existing
product categories/stores
classification or build a clustering
model
can build an overall trend
prediction model based on
external data and use it as
another feature in the final model
for features that are unknown,
such as weather, advertising or
promotions, can either build
individual scenarios or use
averages.
10. IMPLEMENTATION STEPS
Input data:
past sales and stock time series,
assortment dates, rich set of sku
attributes, similarities, rich POS
data, local events, promotions,
advertising campaigns, prices,
relevant social network and internet
data
Implementation complexity:
medium – high
Typical time to implement:
6 – 10 months for implementation
6 – 8 months for learning and
adjusting
Critical steps: need to spend
some time in data cleaning
and advanced analytics to
implement the best
solutions
11. Choosing the right algorithms
time-series algorithms make use of signal
extrapolation whereby trends, seasonality, and
cycles that occurred in the past
feature-based algorithms rely on an abundance
of data
The demand will not be perfectly forecasted!!!
It needs to either account for a specific
strategy or a stochastic optimization model to
decide how much to purchase or deliver based
on the cost of under- and over- stock (risk
based approach)
12. ACTOR’S BLOOMY DECISION PLATFORM
Bloomy is ACTOR’s integrated technology platform built to
support supply chain strategic decisions.
- Demand forecasting
- Distribution optimization
- Production and purchasing planning
- Price and promotion optimization
- What if scenarios simulation
- Big data analytics
- Collaboration tools
are included in the platform, which aims to improve
efficiency and manage risk in decision making.