Malaysian Airlines

A.I. History

A.I. Dilemma

ARTIFICIAL INTELLIGENCE

A program that can sense, reason, act, and adapt

MACHINE LEARNING

Only 1-in-3 A.I. projects are successful and it takes more than 6 months to go from development to production.

Algorithms whose performance improve as they are exposed to more data over time.

A.I.

DEEP

LEARNING

Subset of machine learning in which multi-layered neural networks learn from vast amounts of data.

Reference: Allam Sai Madhav

What should happen?

PRESCRIPTIVE

ANALYTICS

Prescriptive analytics suggests decision options on how to take advantage of a future opportunity or mitigate a future risk and shows the implication of each decision option. Prescriptive analytics can continually take in new data to re-predict and re-prescribe, thus automatically improving prediction accuracy and prescribing better decision options."

Using algorithms to optimize such things as TRO, fuel pricing optimization, commercial and freight airline routing, and labor utilization.

What could happen?

PREDICTIVE

ANALYTICS

Using algorithms and data to

Predict which customers are most likely to use a particular marketing offer.

Why did it happen?

4 Types of Analytics

4 types of Analytics

DIAGNOSTIC

ANALYTICS

Mining data to determine what caused a spike in the airline's web traffic over the past month.

What happened?

DESCRIPTIVE

ANALYTICS

Using Google Analytics to track an airline's website traffic, such as page views and numbers of visitors.

Only 1-in-7 A/B tests results in a positive outcome, making it a resource- and time-intensive strategy.

AI is able to dynamically change the look and feel of a website in real time, as travelers engage with it, to dramatically boost conversions, whether it’s to sell a seat upgrade, a more direct flight or special offers for their trip. And it can do this where airlines need it most, for return visitors or customers enrolled in loyalty programs.

MAL can use AI to test thousands or even millions of designs (be it text, icon, image or button color changes) in the same amount of time and see a 40% to 50% increase in conversions.

Rate of involuntary 'bumps' per 100,000 passengers

AI could predict the likelihood that certain passengers won’t show up or will swap to another flight. The AI could then give ground staff up-to-the-minute information on how many people are likely to board. It could even predict which flyers typically request upgrades or how many employees are likely to be flying standby. This could help the problem of having to remove passengers from planes that have already boarded.

The challenge airlines face today is that their models and staff can’t accurately predict how many people will not show up for a flight.

Flight Delays: Collaborative Intelligence

AI can help humans look at passenger revenue and value and quickly re-book high revenue and value individuals first if a flight is canceled. By accelerating historical data value analysis using AI, we can create a list according to priority – at speeds and with levels of accuracy simply impossible by humans.

A.I. Marketing

5000/Day

• Reason why you left a site.
• How you left.
• Whether you bought anything or not.
• System will make informed decision on whether we want to serve you that ad or something else.

“Technology has changed marketing and market research into something less like golf and more like a multi-player first-person-shooter game. Crouched behind a hut, the stealthy marketers, dressed in business-casual camouflage, assess their weapons for sending outbound messages. Email campaigns, events, blogging, tweeting, PR, ebooks, white papers, apps, banner ads, Google Ad Words, social media outreach, search engine optimization. The brave marketers rise up and blast away, using weapons not to kill consumers but to attract them to their sites, to their offers, to their communities. If the weapons work, you get incoming traffic.”

Dan Woods

300/Day

Fuel

Human + AI: Customer Service

Words are transformed into numbers to extract meaning, context, and the nuances of the customers.

• Agent sees suggested answer, but can ignore it.
• If the AI is confident about the answer, it will answer automatically.

A deep neural network is trained on the customer's data.

When a question comes in, the system routes it to the correct department

Answers are provided on a confidence threshold

If the answer is above the threshold, it is automated.

In the next few years, we’ll see chabots get much smarter as they start to bring in other forms of AI to help sell products or meet customer service needs. One example we could see is chatbots incorporating computer vision to help people discover lookalike trips at cheaper but equally appealing destinations.

Data virtualization

Monitoring

Governance

Metadata

Security

Design tools

Optimizer

Cache

Scheduler

Publish - Real-time, right time data services

Combine - Transform, Improve quality, Integrate

Connect - Normalized views of disparate data

Allows an application to retrieve and manipulate data without requiring technical details about the data, such as how it is formatted or where it is physically located.

Any data or content

Web automation

Library of wrappers

Read and Write

More structured less structured

Streaming Analytics

IoT

Supply chain control

• Inventory optimization

• NFC payment

• Intelligent shopping applications

• Smart product management

• Smartphone detection

• Facial recognition

• Logistics

• PAM

“Stream processing is designed to analyze and act on real-time streaming data, using ‘continuous queries’ (i.e. SQL-type queries that operate over time and buffer windows). Essential to stream processing is Streaming Analytics, or the ability to continuously calculate mathematical or statistical analytics on the fly within the stream. Stream processing solutions are designed to handle high volume in real time with a scalable, highly available and fault tolerant architecture. This enables analysis of data in motion.”

Kal Wähner

SSOT & MVOT

An SSOT is the source from which multiple versions of the truth are developed. MVOTs result from the business-specific transformation of data into information—data imbued with “relevance and purpose.” Thus, as various groups within units or functions transform, label, and report data, they create distinct, controlled versions of the truth that, when queried, yield consistent, customized responses according to the groups’ predetermined requirements

The SSOT is a logical, often virtual and cloud-based repository that contains one authoritative copy of all crucial data, such as customer, supplier, and product details. It must have robust data provenance and governance controls to ensure that the data can be relied on in defensive and offensive activities, and it must use a common language—not one that is specific to a particular business unit or function. Thus, for example, revenue is reported, customers are defined, and products are classified in a single, unchanging, agreed upon way within the SSOT.

• The single source of truth (SSOT) and multiple versions of the truth (MVOTs).
• The SSOT works at the data level; MVOTs support the management of information

SSOT & MVOT for Mkt. & Finance

What’s critical is that single sources of the truth remain unique and valid, and that multiple versions of the truth diverge from the original source only in carefully controlled ways. A data lake can house the SSOT, extracting, storing, and providing access to the organization’s most granular data down to the level of individual transactions. And it can support the aggregation of SSOT data in nearly infinite ways in MVOTs that also reside in the lake.

The marketing and finance departments both produced monthly reports on TV ad spend—MVOTs derived from a common SSOT. Marketing, interested in analyzing advertising effectiveness, reported on spending after ads had aired. Finance, focusing on cash flow, captured spending when invoices were paid. The reports therefore contained different numbers, but each represented an accurate version of the truth.

SSOT & MVOT

The SSOT allowed managers to identify suppliers that were selling to multiple business units within the company and to negotiate discounts.

$75M

Inventory Optimization

Spare parts, an essential component of the availability of any system, have intermittent consumption patterns and usually have only one specific use, and organizations can often source them only from the system manufacturer. For these reasons, many organizations overstock spare parts to avoid costly system downtime.

Airlines can use inventory analytics to identify items that are trending toward being out of stock, providing a means of stock management more reliable than supplier data. In addition, research has shown that monitoring technology can reduce the need for spare part inventory

Deep Learning

Developed by Amazon, Google, Facebook, and Microsoft, these companies are opening their technology to any and all in an attempt to increase AI + ML knowledge. Large and active communities are growing around these solutions.

In essence, TensorFlow removes the need to create a neural network from scratch. Instead, you can train TensorFlow with your data-set and use the results however you wish.

• Text-based applications, i.e., language detection
• Image recognition
• Time series, i.e., recommendation
• Video detection

• Voice recognition – mostly used in IoT, Automotive, Security and UX/UI
• Voice search – mostly used in Telecoms, Handset Manufacturers
• Sentiment Analysis – mostly used in CRM
• Flaw Detection (engine noise) – mostly used in Automotive and Aviation

Unified Analytics

The very thing that makes AI possible—data and all the secrets within it—is also what make the AI process so challenging.

According to Databricks, 90% of the respondents believe that unified analytics—the approach of unifying data processing with ML frameworks and facilitating data science and engineering collaboration across the ML lifecycle, will conquer the AI dilemma.”

Unified Analytics makes it easier for data engineers to build data pipelines across siloed systems and prepare labeled datasets for model building while enabling data scientists to explore and visualize data and build models collaboratively. A unified analytics platform can “unify data science and engineering across the ML lifecycle from data preparation to experimentation and deployment of ML applications—enabling companies to accelerate innovation with AI,

96%

of organizations say data-related challenges are the most common obstacle when moving AI projects to production.

Source: Databricks

AWP