Predict future business outcomes
Transcript of Predict future business outcomes
![Page 1: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/1.jpg)
![Page 2: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/2.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Predict future business outcomes using Amazon Forecast
A I M 3 1 2
Rohit Menon
Sr. Product ManagerAmazon Forecast
Amazon Web Services
Chris King
Sr. Solutions ArchitectVertical AI Services
Amazon Web Services
Sean Patrick McCrary, CPIM
Director, Supply Chain Data Analytics, LSG
Thermo Fisher Scientific
![Page 3: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/3.jpg)
Agenda
• The case for forecasting
• Overview of Amazon Forecast
• Behind the scenes: The algorithms powering Amazon Forecast
• Evaluating your Forecast
• Case study: Reducing excess inventory at Thermo Fisher Scientific
• Wrap up and Q&A
![Page 4: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/4.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 5: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/5.jpg)
The case for forecastingForecasting is the science of predicting the future
Product demand Actual demand vs. forecasted demand ($ Millions)
Actual Demand Forecast Demand
Over-forecasting leads
to wasted resources
Under-forecasting leads
to lost opportunity
![Page 6: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/6.jpg)
The case for Forecasting
Inventoryplanning
Workforce planning
Capacity planning
Excess inventory Unutilized laborUncapitalized infrastructure
Lost sale Overtime costs Unmet demand
Over-forecasting
Under-forecasting
Financial planning
Depleted cash reserves
Undercutting
Impact of under and over forecasting
![Page 7: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/7.jpg)
Product
availabilityFast
delivery
Lower
price
Demand forecasting for
over 400 million
products every day
Inventory and fulfillment
cost reduction to provide
low prices to customers
12 shipping options
with free same-day
delivery
Forecasting at Amazon.com
Accurate forecasting is critical for delivering on customer promises
![Page 8: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/8.jpg)
Forecasting processThree standard steps involved in forecasting
• Looking backward• We always begin with historical data; ideally, that contains a timestamp, an item, a value
• Larger datasets contain potentially millions of rows over thousands of items, but the construct is the same
• This provides the baseline
• Identifying trends• Using statistical, deep learning, or other approaches, you look over the historical data to determine
trends within your data—trends that hopefully continue into the future
• Projecting forward• Given the trends identified, take each item and predict in increments the expected future values
![Page 9: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/9.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
ForecastAmazon
![Page 10: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/10.jpg)
FRAMEWORKS INTERFACES INFRASTRUCTURE
AI Services
Broadest and deepest set of capabilities
The AWS ML Stack
VISION SPEECH LANGUAGE CHATBOTS FORECASTING RECOMMENDATIONS
ML Services
ML frameworks + infrastructure
A m a z o n
P o l l y
A m a z o n
T r a n s c r i b eA m a z o n
T r a n s l a t e
A m a z o n C o m p r e h e n d
& A m a z o n
C o m p r e h e n d M e d i c a l
A m a z o n L e x A m a z o n
F o r e c a s t
A m a z o n
R e k o g n i t i o n
I m a g e
A m a z o n
R e k o g n i t i o n
V i d e o
A m a z o n
T e x t r a c tA m a z o n P e r s o n a l i z e
Ground Truth Notebooks Algorithms + Marketplace Reinforcement Learning Training Optimization Deployment HostingAmazon SageMaker
F P G A SA m a z o n
E C 2 P 3
& P 3 D N
A m a z o n
E C 2 G 4
A m a z o n
E C 2 C 5
A W S
I n f e r e n t i aA W S I o T
G r e e n g r a s s
A m a z o n
E l a s t i c
I n f e r e n c e
D L C o n t a i n e r s
& A M I s
![Page 11: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/11.jpg)
The AWS ML StackThe forecast view
![Page 12: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/12.jpg)
Amazon Forecast
Fully managed
service
Highly accurate Easy to use Your data,
your models
Automatically sets
up data pipeline,
training, and
prediction
50% improvement
in accuracy over
traditional
methods
No deep learning
experience
required
Encrypted with
customer keys
through Amazon Key
Management Service
(AWS KMS)
Automated machine learning service for accurate forecasting
![Page 13: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/13.jpg)
Use of historical data
to predict future values
Target time-series
dataset
The primary variable to predict with its
historical values
(demand, sales)
Datasets used for forecasting
Use of related
attributes and
categorical data
Item metadata
(non-time-varying)
Categorical data that provide more context
about items
(color, city, channel)
Use of known time-
varying data specific to
your business
Related time-
series
dataset
Time-varying related features that may impact
the target value
(price, promotion, weather)
Amazon Forecast
![Page 14: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/14.jpg)
Private forecasting API
Amazon Forecast
The technology that powers the world’s largest ecommerce business
Get started with the console or APIPoint Amazon Forecast to your data
stored in Amazon Simple Storage Service
(Amazon S3)
Automatically train your custom ML
modelLet Amazon Forecast auto select the best one for
your data through AutoML
Generate accurate forecastsRetrieve forecasts through the console or
private API
Historical data
Related data
Sales, call volume, inventory,
resource demand
Price, promotions, weather
data, custom events
Item metadataColor, city, country, category,
author, album name
Built-in dataset(holiday, weekends)
Amazon Forecast
![Page 15: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/15.jpg)
Customized forecasting API
Inspect data
Identifyfeatures
Select most accurate model from multiple algorithms
SelectHyper-
parameters
Hostmodels
Loaddata
Trainmodels using
multiple algorithms
Optimizemodels
Amazon Forecast
Behind the scenes
Fully managed by Amazon Forecast
Historical data
Related data
sales, call volume, inventory,
resource demand.
Price, promotions, weather
data, custom events
Item metadataColor, city, country, category,
author, album name
Create datasetCreate predictor (train,
inference, metrics)Create Forecast
Create Forecast export
Query Forecast
Amazon Forecast
![Page 16: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/16.jpg)
Visualize the distribution of forecasted values
View probabilistic forecasts at
any quantile in the console
Retrieve forecasts
through your private API
Export forecasts to .csv
Amazon Forecast
![Page 17: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/17.jpg)
Handles tricky forecasting scenarios
Missing values
Cold start
(new product introduction)
Irregular seasonality
Product discontinuation
Highly spiky data
Sensitivity analysis
(future price change)
Amazon Forecast
![Page 18: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/18.jpg)
Cost type Pricing
Generated forecasts $0.60 per 1,000 forecasts (rounded to the
nearest 1,000)
Data storage $0.088 per GB
Training hours $0.24 per hour
Pay-as-you-go pricing model
Pricing for Amazon Forecast
![Page 19: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/19.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 20: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/20.jpg)
Algorithms powering Amazon ForecastAmazon Forecast includes a gallery of statistical and machine learning forecasting algorithms
ARIMA
Auto-regressive integrated moving average
De facto statistical method
NN
Flexible Local
Baseline algorithms
ARIMA
![Page 21: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/21.jpg)
Algorithms powering Amazon ForecastAmazon Forecast includes a gallery of statistical and machine learning forecasting algorithms
ETS
Error trend seasonalityStatistical algorithm that uses exponential
smoothing
NN
Flexible Local
Baseline algorithms
ARIMA ETS
![Page 22: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/22.jpg)
Algorithms powering Amazon ForecastAmazon Forecast includes a gallery of statistical and machine learning forecasting algorithms
NPTS
Non-parametric time series
Jan 06
2014
Apr 07
2014
Jul 07
2014
Oct 06
2014
Jan 05
2015
Apr 06
2015
Jul 06
2015
Oct 05
2015
Jan 04
2016
Apr 04
2016
02
46
810
A Typical Time Series in Large Inventories
NN
Flexible Local
Baseline algorithms
ARIMA ETS NPTS
![Page 23: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/23.jpg)
Algorithms powering Amazon ForecastAmazon Forecast includes a gallery of statistical and machine learning forecasting algorithms
Prophet
Additive model with Gaussian likelihood
NN
Flexible Local
Baseline algorithms
ARIMA ETS NPTS
Flexible local algorithms
Prophet
![Page 24: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/24.jpg)
DeepAR+Global model that can use related time
series and attributes
NN
Flexible Local
Baseline algorithms
ARIMA ETS NPTS
Flexible local algorithms
Prophet
DeepAR+
Neural networks
Algorithms powering Amazon ForecastAmazon Forecast includes a gallery of statistical and machine learning forecasting algorithms
![Page 25: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/25.jpg)
How is DeepAR+ different
• The DeepAR+ forecasting algorithm has been used internally in Amazon for mission-critical decisions
• Classical forecasting techniques such as ARIMA and ETS fit one model to an individual time series. However, in many situations, a set of related time series have been or can be collected.
• DeepAR+ can train a model over such a set of related time series for additional insights and increased predictive power
• Requires minimal feature engineering and can produce forecasts that are either point (amount sold was X) or probabilistic (amount sold was between X and Y with Z probability).
![Page 26: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/26.jpg)
DeepAR+ for time series forecasting
• Modeling many related times series: Over 100,000 at one time
• Cold-start forecasting: Based on data from related time series
![Page 27: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/27.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 28: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/28.jpg)
QuantilesExamples
A P90 quantile predicts that, 90% of the time, the true value will be less than the predicted value
A P50 quantile predicts that, 50% of the time, the true value will be less than the predicted value
Select the quantile most appropriate for your use case
Evaluating your forecast
![Page 29: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/29.jpg)
Error/loss functions
Weighted quantile loss (wQuantileLoss) calculates how far off the forecast a certain quantile is from actual demand
RMSE oot mean square error) calculates the difference between the actual target value and the predicted (forecasted) mean value
Many more: MAPE, MASE, log loss
Understanding error and loss metrics
Evaluating your forecast
![Page 30: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/30.jpg)
How would your model perform when deployed at different points in time
Evaluating your forecast: Back testing
![Page 31: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/31.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 32: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/32.jpg)
32 Proprietary & Confidential
The world leader in serving science
invested in R&D
$1BR&D scientists/engineers
5,000employees
70,000in revenues
$24B
![Page 33: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/33.jpg)
33 Proprietary & Confidential
Organizational structure
The world leader in serving science
Analytical precision and
diagnostics excellence
Inspiring meaningful
genetic analysis
Accelerating discovery
research
One-stop access for
scientific products
Instrument and
enterprise services
![Page 34: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/34.jpg)
158,000 SKUs from highly valuable devices to perishable consumables
Highly complex and diverse inventory to manage
![Page 35: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/35.jpg)
Manufacturing Forecast inaccuracy is costly
![Page 36: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/36.jpg)
Traditional forecast models
![Page 37: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/37.jpg)
Our approach to experimenting with and adopting Amazon Forecast• Input file: 3 years of sales data for 158,000 SKUs for all sites
• Inspected and preprocessed data feed to remove:
• Financial adjustments and internal transfers
• SKUs with less than 3 transactions per year
• SKUs with aggregate demand or price equal to or less than 0
• Used sales price as related time series and other item metadata information
• Trained and optimized models using various algorithms and parameters, including DeepAR+
• Focus group: 20,000 SKUs where we generated forecasts
• Conducted back-testing to evaluate accuracy and compared with our legacy system by looking
at MAPE, WAPE, and mean bias metrics for various SKU groups
![Page 38: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/38.jpg)
Amazon Forecast vs. legacy system results
CategoriesOther
(legacy system)Amazon Forecast Difference
W-Runner-High * GB01 78.78% 89.21% 10.43%
W-Runner-High * US02 61.25% 82.08% 20.83%
W-Runner-High * US03 68.55% 80.81% 12.26%
W-Runner-High * MTS Manuf 70.74% 79.87% 9.13%
W-Runner-High * MTS Purch 69.26% 84.55% 15.29%
W-Runner-High * Prod Line FCR 72.75% 83.44% 10.69%
Amazon Forecast Deep AR+ showed better results in all the SKU groups evaluated
Forecast MAPE accuracy results – Set A
![Page 39: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/39.jpg)
Amazon Forecast vs. legacy system results
CategoriesOther
(legacy system)
Amazon
ForecastDifference
Primary Site: U50 16.40% 43.49% 27.09%
Primary Site: UM -69.78% 20.15% 89.93%
Prod Line Group: 1C7 5.06% 36.99% 31.93%
Prod Line Group: ANT 40.60% 53.20% 12.60%
Prod Line Group: ELI 17.90% 42.94% 25.04%
Item Group Code: INV250 19.45% 40.45% 21.00%
Stocking Type:
MTS Purch Catalog-Vendor14.32% 48.45% 34.13%
Forecast MAPE Accuracy Results – Set B
![Page 40: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/40.jpg)
Visualization - Example 1
Amazon Forecast inventory accuracy visualization
![Page 41: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/41.jpg)
Amazon Forecast inventory accuracy visualization
Visualization - Example 2
![Page 42: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/42.jpg)
Increase in forecast accuracy translates to millions of dollars of reduction of
inventory and/or improvement in managing backorders
Amazon Forecast inventory accuracy visualization
Visualization - Example 3
![Page 43: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/43.jpg)
Amazon Forecast vs. legacy system TCO
Other
(legacy system)Amazon
Forecast
Installation time 2–8 months 6–8 weeks
Pricing model License fee Pay-as-you-go
model
Maintenance costs (3 years) Up to a $1,000,000 $6,000*
Professional services Up to $150K As needed
*Includes training, model maintenance for monthly forecast for 35,000 SKUs, and
two quantiles
![Page 44: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/44.jpg)
Our AWS AI/ML Journey
leap-frog to better results
![Page 45: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/45.jpg)
• Brainstorming
• Custom modeling
• Training
• Work side-by-side with Amazon experts
AWS Partner & ML Solutions Lab
• Practical education on ML for new and experienced practitioners
• Based on the same material used to train Amazon developers
Machine learning training and certification
Thermo Fisher Scientific and AWS collaboration
![Page 46: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/46.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 47: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/47.jpg)
Amazon Forecast in production
• Automate into a repeating process
• Define your integration points andcreate data pipelines for input data sources
• Continually infer new predictions
• Train models on an as need basis
• Automate evaluation and trigger based on error thresholds
Ingest
Train
Predict
Evaluate
![Page 48: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/48.jpg)
Amazon Forecast in production
Manages a set of EC2 instances
Maintains cache pools for EC2 instances
Needs to forecast cache pool size
Demand
publisher
Cluster
scaling
EC2 Cache Pool
Amazon
Redshift
Demand
Forecast
table
ETL Lambda
Amazon
Forecast
Demand
Forecasts
Demand
history
1
2
34
1. EC2 Cache pool demand changes are published to S3 bucket
2. New data is ingested into Amazon Forecast and new forecast predictions are stored in S3 bucket
3. A Lambda function copies new forecasts to an Amazon DynamoDB table
4. The cluster scaling logic reads forecasts and adjust the cache pool size based on projected demand
Amazon Redshift Cluster Management
![Page 49: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/49.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 50: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/50.jpg)
Partners with capabilities in Amazon Forecast
![Page 51: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/51.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Role-based ML learning paths for developers, data scientists, data platform engineers, and business decision makers
Learn ML with AWS Training and Certification
Visit https://aws.training/machinelearning
The same training that our own developers use, now available on demand
70+ free digital ML courses from AWS experts let you learn from real-world challenges tackled at AWS
Validate expertise with the
AWS Certified Machine Learning - Specialty exam
![Page 52: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/52.jpg)
Thank you!
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.
![Page 53: Predict future business outcomes](https://reader031.fdocuments.us/reader031/viewer/2022011923/61d8acbc8316c1572247b2d5/html5/thumbnails/53.jpg)
© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.