1st edition | July 8-11, 2019
Company Creation
Technology 2 Client (T2C) was
founded in December 2003 and
started operating in January 2004
establishing the first office in the
centre of Barcelona.
2003
New HQ
In 2014, T2C moves to Avda.
Diagonal, ‘prime’ zone for tech
companies in Barcelona
consolidating employees in a
single office.
2014
México Subsidiary
Where we expect to develop and
provide our Advanced Analytics,
development and SAP services.
2019
Málaga Office
Through an agreement with the
Tech hub in Málaga (PTA) and
the Málaga University (UMA), we
facilitate our entrance in the hub.
2018
Our Goals
We are a company built by people that
aims to help people. We care less about
numbers than we do about building trust.
TALENT
DNA
We help our clients to improve their processes and
consolidate projects using the best talent available.
CONSISTENCY
15 years
Since our creation we have never stopped growing
in a organic manner that allows us to maintain the
core values of the company.
INNOVATION
Future
We are always looking to anticipate what’s next by
constantly scanning the market looking for new
products and trends.
EFICIENCY
Maximum
Obsessed with efficiency, we thrive in projects with
maximum impact and added value.
ML for Logistics:
Predicting expedition outcome
Most Changes Are About Learning
T2C
Global logistics market is anticipated to register a CAGR of 3.48% from
2016 to 2022 to attain a market size of around $12,256 billion by 2022.
Allied Market Research
T2C
Logistics impact on sales
5
3.5
3.3
Warehousing
Other
Transportation
1
3
2
T2C
Logistics is a network based business that generates
huge quantities of data.
T2C
Word cloud
T2C
Operational inefficiencies can often lead to potential
revenue losses, increasing costs and poor customer
service.
T2C
24
32Clean Sheet
FTL Costs
Variable
Costs
Fixed
Costs
Fuel
Labor costs
Transportation costs
T2C
Use Case: Luís Simões
T2C
Project Definition
QUESTION
Define the question we are going to answer.
“Is an expedition likely to fail?”
LABEL
Define which label we are going to predict (Supervised Learning).
“Expedition outcome: Binary classification SUCCESS/FAILURE”
AMOUNT
Is there enough data to answer that question? Are there enough positive instances?
“Do our expeditions fail that often? Do we properly record these failures?”
GOALS
Define precisely what a training instance is, the goal and the evaluation method.
“An expedition can contain several movements.”
“ Improve service level.”
“We will measure the ROI of the project based on average failures reduction.”
T2C
Distribution
Smart
Plant
Warehouse
DB2
DB2
SQL ML DS
MySQL
Staging Area7 Systems Product
KPI DS
ERP
SQL
1
2
3
4 5
Architecture
T2C
Iterative process
Data
Acquisition
T2C
Integrating heterogenous systems
T2C
Iterating
T2C
Iterative process
Data
Acquisition
Dealing
with data
T2C
1. Data exploration
2. Data cleaning
3. Data transformations
4. Feature engineering
5. Feature selection
T2C
1. Data exploration
T2C
2. Data Cleaning
3. Data Transformation
Transformations need to be applied to raw data to obtain Machine Learning ready data.
• Types of missing values:
• Meaningful missing: The fact that a value is missing adds information.
• Meaningless missing: The fact that a value is missing is accidental.
• Strategies:
• Drop rows with missing values if there is a small percentage.
• Drop features with a high percentage of missing values.
• Impute missing values with static content: median, mode, mean…
• Use Machine Learning techniques to impute missing values.
T2C
4. Feature Engineering
T2C
Selected Features
Feature Description
date Date of order generation.
Planned_delivery_date Planned delivery date of the order.
Final_Destination Arrival destination code.
destination_name Destination name.
destination_location Destination location.
destination_zip_code Destination zip code.
destination_province Destination province.
destination_region Destination province.
origin Origin code.
origin_name Name of the origin of the order.
num_stops Number of stops made by the driver.
diff_hours Number of hours between order generation and estimated
time of delivery.
Feature Description
distance Distance (in a straight line) between origin and destination.
num_pallets Number of pallets in the shipment.
num_volumes Number of boxes in the shipment.
weight weight (kg).
urgent Indicates if a shipment is urgent.
is_workingday_delivery Indicates if it is a workday the day of delivery.
is_workingday_delivery_d+1 Indicates if it is a workday the day after of delivery.
is_workingday_delivery_d+2 Indicates if it is a workday two day after of delivery.
is_workingday_delivery_d-1 Indicates if it is a workday the previous day of delivery.
is_workingday_delivery_d-2 Indicates if it is a workday two days previous day of delivery.
order_ok OBJECTIVE FIELD: Order completed with or without success.
T2C
Iterative Process
Data
Acquisition
Modelling
Dealing
with data
T2C
First Model: Decision Tree
• Comparative benchmark.
• First business insights.
T2C
Choose Your First Model
A B
T2C
First model: Decision tree
T2C
Chosen Model: Decision Forest
T2C
Final Model Evaluation
T2C
Results
• Possible savings $xxxx/year.
• Quality of service can be improved from x% up to y%.
T2C
Iterative process
Data Acquisition Consumption
Dealing
with data
Modelling
T2C
Overview
SERVICE LEVEL (ORDERS)
T2C
Incident Map
INCIDENTS IN DELIVERY
T2C
ARRIVAL TIME BAND
Deeper Analysis
T2C
Machine Learning
PLANNED ORDERS RISK
T2C
Project Numbers
Hours Afterworks
KPIsConsultants
Meeting hoursFeatures
Lines of code
Trained Models
870 11
335
4137
1.240174
Technology 2 Client S.L.
Finance Human Resources Marketing Sales
Other Machine Learning Use cases.
Fraud Transactions.
Rolling Forecast.
Capex Analysis.
Predictive
Client Churn.
Social Media Analysis.
Retention &
Engagement.
Performance Analysis.
Absenteeism Prediction.
Hiring Optimization
Customer Service
Legal R&D IS/IT Purchase Demand Planning
Pricing Optimization.
Sales Modeling.
Up-sell Opportunity
Analysis.
Customer Loyalty and
Retention .
Sentiment Analysis.
Recommendation
Engine.
Review Documents and
Legal Research.
Contract Review and
Management.
Predict Legal Outcomes.
Predict HW Failures.
User Behaviors.
Cybersecurity.
Success of Research.
Predict Hardness
Materials.
Trend Analysis and
Studies.
Pricing Strategy.
Stock-out Prediction.
TCO Analysis.
Demand Forecasting.
Predict Cannibalization.
New Products Impact.
T2C
THANK YOU
linkedin.com/company/5083052

DutchMLSchool. ML for Logistics

  • 1.
    1st edition |July 8-11, 2019
  • 2.
    Company Creation Technology 2Client (T2C) was founded in December 2003 and started operating in January 2004 establishing the first office in the centre of Barcelona. 2003 New HQ In 2014, T2C moves to Avda. Diagonal, ‘prime’ zone for tech companies in Barcelona consolidating employees in a single office. 2014 México Subsidiary Where we expect to develop and provide our Advanced Analytics, development and SAP services. 2019 Málaga Office Through an agreement with the Tech hub in Málaga (PTA) and the Málaga University (UMA), we facilitate our entrance in the hub. 2018
  • 3.
    Our Goals We area company built by people that aims to help people. We care less about numbers than we do about building trust. TALENT DNA We help our clients to improve their processes and consolidate projects using the best talent available. CONSISTENCY 15 years Since our creation we have never stopped growing in a organic manner that allows us to maintain the core values of the company. INNOVATION Future We are always looking to anticipate what’s next by constantly scanning the market looking for new products and trends. EFICIENCY Maximum Obsessed with efficiency, we thrive in projects with maximum impact and added value.
  • 4.
    ML for Logistics: Predictingexpedition outcome Most Changes Are About Learning
  • 5.
    T2C Global logistics marketis anticipated to register a CAGR of 3.48% from 2016 to 2022 to attain a market size of around $12,256 billion by 2022. Allied Market Research
  • 6.
    T2C Logistics impact onsales 5 3.5 3.3 Warehousing Other Transportation 1 3 2
  • 7.
    T2C Logistics is anetwork based business that generates huge quantities of data.
  • 8.
  • 9.
    T2C Operational inefficiencies canoften lead to potential revenue losses, increasing costs and poor customer service.
  • 10.
  • 11.
  • 12.
    T2C Project Definition QUESTION Define thequestion we are going to answer. “Is an expedition likely to fail?” LABEL Define which label we are going to predict (Supervised Learning). “Expedition outcome: Binary classification SUCCESS/FAILURE” AMOUNT Is there enough data to answer that question? Are there enough positive instances? “Do our expeditions fail that often? Do we properly record these failures?” GOALS Define precisely what a training instance is, the goal and the evaluation method. “An expedition can contain several movements.” “ Improve service level.” “We will measure the ROI of the project based on average failures reduction.”
  • 13.
    T2C Distribution Smart Plant Warehouse DB2 DB2 SQL ML DS MySQL StagingArea7 Systems Product KPI DS ERP SQL 1 2 3 4 5 Architecture
  • 14.
  • 15.
  • 16.
  • 17.
  • 18.
    T2C 1. Data exploration 2.Data cleaning 3. Data transformations 4. Feature engineering 5. Feature selection
  • 19.
  • 20.
    T2C 2. Data Cleaning 3.Data Transformation Transformations need to be applied to raw data to obtain Machine Learning ready data. • Types of missing values: • Meaningful missing: The fact that a value is missing adds information. • Meaningless missing: The fact that a value is missing is accidental. • Strategies: • Drop rows with missing values if there is a small percentage. • Drop features with a high percentage of missing values. • Impute missing values with static content: median, mode, mean… • Use Machine Learning techniques to impute missing values.
  • 21.
  • 22.
    T2C Selected Features Feature Description dateDate of order generation. Planned_delivery_date Planned delivery date of the order. Final_Destination Arrival destination code. destination_name Destination name. destination_location Destination location. destination_zip_code Destination zip code. destination_province Destination province. destination_region Destination province. origin Origin code. origin_name Name of the origin of the order. num_stops Number of stops made by the driver. diff_hours Number of hours between order generation and estimated time of delivery. Feature Description distance Distance (in a straight line) between origin and destination. num_pallets Number of pallets in the shipment. num_volumes Number of boxes in the shipment. weight weight (kg). urgent Indicates if a shipment is urgent. is_workingday_delivery Indicates if it is a workday the day of delivery. is_workingday_delivery_d+1 Indicates if it is a workday the day after of delivery. is_workingday_delivery_d+2 Indicates if it is a workday two day after of delivery. is_workingday_delivery_d-1 Indicates if it is a workday the previous day of delivery. is_workingday_delivery_d-2 Indicates if it is a workday two days previous day of delivery. order_ok OBJECTIVE FIELD: Order completed with or without success.
  • 23.
  • 24.
    T2C First Model: DecisionTree • Comparative benchmark. • First business insights.
  • 25.
  • 26.
  • 27.
  • 28.
  • 29.
    T2C Results • Possible savings$xxxx/year. • Quality of service can be improved from x% up to y%.
  • 30.
    T2C Iterative process Data AcquisitionConsumption Dealing with data Modelling
  • 31.
  • 32.
  • 33.
  • 34.
  • 35.
    T2C Project Numbers Hours Afterworks KPIsConsultants MeetinghoursFeatures Lines of code Trained Models 870 11 335 4137 1.240174
  • 36.
    Technology 2 ClientS.L. Finance Human Resources Marketing Sales Other Machine Learning Use cases. Fraud Transactions. Rolling Forecast. Capex Analysis. Predictive Client Churn. Social Media Analysis. Retention & Engagement. Performance Analysis. Absenteeism Prediction. Hiring Optimization Customer Service Legal R&D IS/IT Purchase Demand Planning Pricing Optimization. Sales Modeling. Up-sell Opportunity Analysis. Customer Loyalty and Retention . Sentiment Analysis. Recommendation Engine. Review Documents and Legal Research. Contract Review and Management. Predict Legal Outcomes. Predict HW Failures. User Behaviors. Cybersecurity. Success of Research. Predict Hardness Materials. Trend Analysis and Studies. Pricing Strategy. Stock-out Prediction. TCO Analysis. Demand Forecasting. Predict Cannibalization. New Products Impact.
  • 37.