Ai Use Cases Ebook
Ai Use Cases Ebook
Ai Use Cases Ebook
You-can-do-this-today
artificial intelligence
use cases
AI can help. Solutions are available now that can transcribe and summarize spoken text,
distinguish between legitimate and malicious activity, and even scan through millions of
social media posts to spot trends and perform sentiment analysis in real time.
Why is AI important?.................................................................................................. 06
1 Digital assistants apply deep semantic parsing using natural language processing and
understanding as well as custom algorithms to derive accurate intent and context from
conversations. With digital assistants that understand their business-specific vocabulary,
employees and customers can communicate with applications using voice commands.
2 Speech recognition systems create accurate, real-time transcripts from natural conversations
or convert recorded audio conversations to textual data to analyze with AI services. Speech
recognition is increasingly multilingual, with the ability to translate between multiple languages
instantaneously.
3 Language services detect and translate languages in written text and automatically recognize
specific types of content, including the names of people, locations, products, and organizations.
These services can also analyze the mood or tone of the text as positive, negative, or neutral with
a confidence score and automatically classify text based on key words and phrases.
4 Vision recognition extracts textual or visual information from still and moving images and
uses that data to drive analytics applications. These AI tools can enrich image-based files with
metadata tags, such as descriptions of objects; those tagged images can be indexed for uses such
as intelligent search and retrieval. With the right training, the system can also detect anomalies in
photographs and videos, which is why vision recognition is currently being used in manufacturing
applications such as automatic quality control.
5 Document understanding automatically extracts text and key-value pairs from a variety of
digital or scanned files, including those that are rotated, tilted, or even deteriorated. This can help
with applications such as expense processing, by reading receipts, or logistics, by understanding
bills of lading. These tools can also identify and extract table structure from documents and sort
documents into common categories.
Generative AI is being embedded into business applications to work with and augment other
AI functionality, such as automatically generating a draft job or product description, drafting
a doctor’s discharge note, or concisely summarizing an article to answer a customer support
inquiry.
3 generative AI capabilities
• Assisted authoring
Using a short prompt, generative AI capabilities can quickly create content, such as job or new
product descriptions, for review, revision, and approval.
• Suggestions
Quickly guide users to better results based on natural language processing and best practices.
Generative AI can also offer recommendations for survey questions or professional development
tips for managers to provide to employees. For example, a customer service application can
suggest a how-to article based on the customer’s inquiry.
• Summarization
Increase efficiency by identifying key insights from one or more data sources. New generative
AI capabilities can explain the key elements of content for simple and impactful consumption.
For example, a customer service agent can ask for a summary of a knowledge base article to
resolve a customer service inquiry or deliver a concise text summary from a table or chart.
Thank AI.
In fact, many of the best AI products for businesses aren’t products at all. They are features
and functions integrated into mission-critical applications, such as enterprise resource
planning (ERP), logistics, accounting, or human resources management.
These new features have one thing in common: sophisticated pattern-matching algorithms
that, once trained, apply patterns to help humans do tasks or solve problems using
capabilities such as speech recognition, image classification, and document understanding.
The data sets used by AI to train accurate models can be huge, and the amount of computing
horsepower required to train AI systems can be substantial. Organizations may choose to use
their own data—either structured, such as a database, or semistructured/unstructured, such
as text, graphics, audio, or video—to create custom models.
1 Retail
Supply chain optimization
The business’s enterprise resource planning system records historical sales data, which includes external
factors such as seasonality, weather patterns, and results of marketing campaigns and promotions.
Advanced data management systems integrate this information into a massive data set, which is then
modeled by data scientists and inputted into a forecasting model that can make predictions, such as demand
increase due to a holiday or planned marketing promotion, and determine the time required to acquire and
distribute the extra inventory.
The AI model can be proactive: For example, it can alert the retailer that the sales of certain items are
exceeding the ability to restock in some locations and suggest that promotional offers be used to redirect
customers to other products.
Most ERP-based forecasting systems can produce sophisticated reports. A generative AI report writer goes
further, creating custom, plain-language summaries of these reports tailored for each store, instructing
managers about how to maximize sales of well-stocked items while mitigating possible shortages.
Oracle solutions
Customer: A hotel chain that depends on positive ratings on social media and review websites.
The problem: The hotel chain can’t keep track of its online reviews. There are too many postings, comments,
and hash-tagged mentions for customer service staff to track and analyze, leading to missed opportunities to
engage unhappy customers complaining on social media. Corporate executives are unable to track sentiment
trends. Hotel managers don’t know what’s being said fast enough to address problems in real time.
The AI solution: The hotel can create a large data set from the tens of thousands of previously published online
reviews on sites such as Yelp, TripAdvisor, its own Facebook page, booking websites as well as comments scraped
from social media, by looking at mentions and hashtags. A textual language AI system can perform a sentiment
analysis across that data set to determine a baseline that can be periodically reevaluated to spot trends.
Taking it up a notch, data scientists could build a model that correlates those textual messages—and their
sentiments—against specific hotel locations, their occupancy levels, staffing on specific dates, and other factors
such as weather. AI can determine if there are signals that could warn individual hotels of situations that may
lead to complaints so that remedial action can be taken in advance.
Generative AI can extract valuable suggestions and insights from both positive and negative comments.
Perhaps customers in one location have problems with parking. Out-of-date guest rooms and inconvenient
breakfast hours are generating buzz at another property. The buffet at another location gets lots of enthusiastic
shout-outs. AI can spot those trends, even if the phrasing isn’t identical. This use of AI can go beyond sentiment
analysis to offer proactive guidance to individual managers and the corporate headquarters, enabling it to
continually improve its guest experiences.
Oracle solutions
The problem: The call-center manager wants to give every customer a great experience while minimizing the
number of calls placed on hold or transferred to supervisors. But calls keep coming in. People call the hotline for
any number of reasons: to make reservations; ask for changes, cancellations, or a better price; or for help finding
pickup/drop-off locations. Customers call when they are having a crisis, such as a flat tire, or when they simply
can’t figure out how to open the trunk.
The AI solution: At the call center, various AI technologies can be combined to act as an invisible assistant for
agents, transcribing each call in real time, performing sentiment analysis to discover meaning, and advising
agents through onscreen information displays and chatbot-like windows. For example, speech recognition AI
can detect essential information, such as the customer’s name, the rental car city, the account number, and the
rental policy number, and display this information to the agent.
A generative AI system, underpinned by a large language model (LLM) and a knowledge database, can
understand many common questions and can feed answers back to the agent. Can’t find the rental desk? The AI
can provide the agent with a map and create directions. In an accident? The AI can trigger automatic responses
while prompting the agent on the essential questions to ask.
AI technology can also assist the agent by assessing the caller’s emotional responses, whether happy, stressed,
angry, or confused. It may offer suggestions for de-escalating a confrontational situation and prompt the agent
to offer discounts, upgrades, bonus frequent-renter points, or other accommodations. After the call, the AI
system can help assess the call quality and customer satisfaction, with feedback for the agent and the manager.
Oracle solutions
OCI Data Science OCI Language OCI Speech Oracle Digital OCI Generative AI
Assistant (coming soon)
Customer: A large warehouse and shipping center that hires hundreds of full-time, part-time, and seasonal
employees each year.
The problem: The process of forecasting hiring needs and timelines and generating the job descriptions can
be time consuming for HR, creating a bottleneck in the hiring process. Managers need to know when to begin
the recruiting process because some positions are harder to fill than others. And while many of the jobs seem
similar, there are sufficient differences to affect job descriptions and pay scale.
The AI solution: Using its database of current and past employees, HR can train an AI model to predict how
long workers are likely to stay in a position. The AI model takes into account the details of each job category
as well as the age of employees, starting pay, history of raises and bonuses, and typical seasons when workers
may leave, such as right before the beginning of a school year. Based on that information, the AI system can
make reasonable predictions to estimate the number of openings and when those openings may occur. When
coupled with historical data regarding advertising and start dates, the AI model can advise HR on when to begin
a search for those workers.
The HR department can feed its job description database into a data model (a visual representation of data
elements and their connections) to generate checklists for hiring managers and HR staff to use when preparing
to hire specific roles. Generative AI can then take those checklists and create textual drafts of job descriptions
and matching job ads formatted to the styles of specific job boards and other posting sites.
The result: Using these predictive models, HR can more accurately plan for hiring requirements throughout the
year, minimizing both staff shortfalls and the time spent preparing job descriptions and ads for open positions.
Oracle solutions
1 Healthcare
The patient’s journey
Generative AI can summarize transcripts from the visit, as well as lab results and the doctor’s diagnosis, and
prepare a draft of these materials for medical staff to edit and approve, thereby saving time. If the doctor
recommends a specific exercise regimen, for example, the generative AI model can prepare instructions that
consider factors such as a recent knee replacement or a preference for swimming over walking.
With generative AI, medical practices can deliver positive health outcomes while seeing as many people
as possible. And in the process, practices can reduce costs, improve earnings, and provide human-centric
experiences for patients who receive better treatment faster.
Oracle solutions
Customer: A regional bank needs to detect fraud more quickly and effectively.
The problem: Unfortunately, fraud can creep into any part of a bank’s retail operations. Branches may
encounter fake documents being used to open accounts and stolen checks being deposited. Fraud can happen
with online transactions from a phone or browser, at offsite ATMs, even among staff. Without trust, banks won’t
have customers—or shareholders. Excessive fraud, and delays in detecting it, can violate financial industry
regulations, cause insurance premiums to increase, invite closer scrutiny by regulatory agencies, and generate
negative press. That means a loss of confidence by customers, investors, and the broad market.
Together, these technologies feed into an anomaly detection model that calculates the statistical risk of
fraud in each attempted transaction within seconds. If the risk is too high, alerts trigger automated actions,
such as to place a hold on withdrawals, prompt the customer to provide additional forms of identification,
or otherwise temporarily delay the transaction pending a manual review by a risk assessment specialist.
Consider a request to open a credit card. The applicant might be using a real credit score, address,
identification, and other personal information; or it may be fraudulent, gained via identity theft. The AI
system could detect anomalies based on factors a human may not notice, such as an ID with a slightly
different physical or email address than the application, or an ID that is a photograph of an ID scanned by a
mobile device instead of the genuine item.
The result: Fraud detection that is fast and reliable, with minimal false positives or negatives. AI-enabled,
automated triggers and explanations can facilitate a rapid, manual risk-assessment review within the
timeframes needed to satisfy government banking regulations, without inconveniencing honest customers.
Oracle solutions
Customer: A biotech firm looking to find effective new treatments faster and more cost-effectively.
The problem: Researchers at biotech companies must integrate a tremendous quantity of information such as
prior research at their firms, papers in peer-reviewed journals, academic work still in progress, formal clinical
trials, patent applications, and more. Having a full grasp of prior work can suggest new approaches to solving
problems while saving the time and cost of inadvertently trying something that’s already failed. Anything
that can help the researcher find and interpret relevant information will speed the process, helping bring new
medications and medical devices to market more quickly.
The work doesn’t stop when a relevant paper, diagram, database, or other information is found. Generative
AI can work with the researcher to find the specific information, summarize it, and even interactively create
a contextual representation of how that outside data relates to the researcher’s current work.
The benefit goes beyond the obvious productivity boost for the researcher. Given the vast quantities
of pharmaceutical data available, it would be easy to overlook relevant information that might result
in breakthroughs or warn of potential areas of concern, such as side effects. Given the extreme cost of
this research and the time-consuming nature of reviewing prior material, AI assistance can be a game
changer—and a lifesaver.
Oracle solutions
1 Utilities
Predictive maintenance
Customer: An electricity grid operator with a service area of thousands of square miles.
The problem: The operator owns a vast array of equipment—pylons and substations, long-distance
transmission lines, underground cables, poles, and transformers. Preventive maintenance on this equipment
is both necessary and complex, and it must happen according to the manufacturers’ schedules as well as
when individual devices show signs of trouble. The objective is to perform maintenance as cost-effectively as
possible, reducing travel time and making use of technicians’ skill sets and availability while meeting service
uptime delivery requirements.
The algorithms can then not only detect anomalies based on real-time data from devices but determine
severity: Is it a case of “check this out next time a crew is in the area” or a “prepare to reroute power while
dispatching a team” situation? The forecasting software can make recommendations while also showing the
justifying data. It can also maximize field crew efficiency by identifying proactive maintenance that can be
performed while workers are fixing nearby equipment.
With intelligent scheduling, anomaly detection, and forecasting based on past data, the utility can anticipate
failures and get early warnings of potential issues, which ensures equipment uptime and helps manage
maintenance costs.
Oracle solutions
Customer: A factory that makes metal parts and other products used in shipbuilding, construction,
and other applications.
The problem: Many manufacturers work under strict requirements and quality controls, often due to industry
regulations and/or contract terms. As a result, these factories use both visual inspections and electronic means
to monitor product quality. A part that fails to meet the requirements may be reworked or repurposed, or it may
need to be scrapped. The factory seeks to maximize profits and throughput by shipping as much good material
as possible while minimizing waste by detecting and handling defects early.
The AI solution: Factories generate a great deal of data about how each product was created, including furnace
temperatures and the quantity of metals and gases used for each stage in the manufacturing process, which
can be compared against requirements as well as best practices.
In addition, the quality assurance process creates X-ray images. This data can be interpreted by computer
vision, which can learn to identify cracks and other weak spots after being trained on a large data set. Such
defects can be categorized based on their size, shape, and quantity. In addition, problematic or ambiguous data
can be highlighted for human inspectors, who will be the final arbiters of the QA inspection.
In this way, machine learning based on large data models, real-time telemetry, and computer vision can assist
with analyzing data and make recommendations for approving or repurposing each product. For example, a
beam that’s not sufficient to hold 100 tons for one application can be certified to hold 10 tons for a different
application. It might be sold for a lower price, but that’s better than scrapping it entirely.
Oracle solutions
The problem: Security personnel at a minimum-security defense installation need to quickly determine which
individuals should be granted access, often using documents, such as passports, national ID cards, or (in the
United States) a driver’s license. These documents come in a wide range of sizes, shapes, designs, and with unique
authenticity markings making it difficult for security staff to spot fakes. It can also be challenging to automate the
process of extracting information from those IDs to check against databases and record in visitor logs.
The AI solution: Checkpoint scanners can quickly submit ID cards into a large AI model that has been trained to
not only recognize various types of identification documents, but to check them against both implicit security
controls and a large sample of forged or otherwise fraudulent documents, making it possible to determine
which documents are legitimate with a high degree of confidence.
Beyond the appearance of the ID itself, document-understanding algorithms can parse and extract data,
even when that data is unlabeled. For example, a field may be written in many forms, some numeric, some
alphanumeric; an ID with several items that appear to be dates may have an issue date, an expiration date, and
the individual’s date of birth—none of which are clearly labeled. Names, addresses, and other information may
also require interpretation by an AI algorithm trained with a large data set.
Once data has been extracted from the ID, and a fraud determination reached, the AI system assisted by
algorithmic prompts can compare that individual against those who are explicitly authorized to enter the facility;
those who may have implicit authorization, such as contract employees; and those who should be deferred for a
closer screening.
Oracle solutions
AI services
With OCI AI services, developers can custom-train models, deploy ready-made ML models, and enhance applications
and solutions with AI capabilities, without needing to be machine learning experts. Oracle AI comes with advanced
models that are pretrained on business data, which can be customized by domain and customer data. They include
OCI Document Understanding: Extract text, tables, and other key data
from document files through APIs and command-line interface tools.
Oracle’s machine learning services, geared toward data scientists, make it easier to
build, train, deploy, and manage custom machine learning models. These services
deliver data science capabilities with support from favorite open source frameworks
or through in-database machine learning and direct access to cleansed data.
OCI Data Labeling: A service for building labeled data sets to train AI
and ML models more accurately. It applies labels to text or images and
then uses those labels to customize models and build shared catalogs.
AI infrastructure
Use the examples presented here to spark your own conversations, and
see where your organization can leverage the benefits from today’s AI
technologies to help prepare for a more successful future.
We suggest the following Oracle resources to help guide you in this journey:
Artificial Intelligence on OCI and Generative AI with Oracle.
Learn more
Connect with us
Call +1.800.ORACLE1 or visit oracle.com
Outside North America, find your local office at oracle.com/contact
Copyright © 2023, Oracle and/or its affiliates. This document is provided for information purposes only, and the contents hereof are subject to
change without notice. This document is not warranted to be error-free, nor subject to any other warranties or conditions, whether expressed
orally or implied in law, including implied warranties and conditions of merchantability or fitness for a particular purpose. We specifically
disclaim any liability with respect to this document, and no contractual obligations are formed either directly or indirectly by this document.
This document may not be reproduced or transmitted in any form or by any means, electronic or mechanical, for any purpose, without our
prior written permission. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their
respective owners.