Unlocking Value with AI | WSO2Con USA 2024

Summary

The speaker emphasizes that the value of AI lies not in its complexity but in its practical applications to create tangible outcomes. They recount a story from a fishing company to illustrate the importance of showing positive results and discuss various use cases where AI enhances products, customer support, and operational efficiency. The talk also addresses the challenges of AI implementation, advocating for a clear understanding of the problem, the availability of simpler solutions, and the need for ongoing monitoring and risk management.

Takeaways

• 📈 The importance of applying technology to create value rather than just using it for the sake of it was emphasized, illustrated by a fishing company's data science team focusing on showing charts that went up.
• 🔍 AI's value is not inherent but depends on how it's applied to create outcomes that matter, rather than the complexity of the technology itself.
• 👕 A real-world example of AI's value was provided in the apparel manufacturing industry, where image processing technology helped improve quality control significantly.
• 🛍️ AI can be used to make products or websites seamless by anticipating user needs and personalizing experiences, enhancing user satisfaction.
• 💬 AI has shown success in customer support, with LLM-based solutions handling a large portion of queries and improving efficiency.
• 🚀 AI can simulate physical systems for analysis and troubleshooting, as demonstrated by NASA's use of physical models to solve problems during space missions.
• 🛒 In retail, AI can be used to optimize store layouts and checkout processes, improving customer experience and operational efficiency.
• 🔓 AI can remove friction from user experiences, such as with biometric authentication, making interactions more natural and less intrusive.
• 🔮 When considering AI, it's crucial to first determine if the problem can be solved without it, as simpler solutions are often more cost-effective.
• 📝 A checklist for applying AI models was discussed, including verifying the value, determining if AI is necessary, assessing data availability and legality, and planning for implementation and monitoring.
• 🛑 The speaker highlighted the risks and challenges of AI, such as the potential for biased predictions and the need for careful management to avoid unintended consequences.

Q & A

What was the fishing company's approach to data science and insights?

The fishing company's manager instructed the data science team to only show charts that went up, indicating a positive trend. The team tried hard to find such insights, eventually discovering a chart that showed the weight of fish increasing with age, which was a positive trend they could report.

What is the key takeaway from the fishing company's story about applying technology?

The key takeaway is that not every application of technology creates value. It's important to focus on how technology is applied and the value it creates rather than just the effort or investment put into the technology itself.

What is the speaker's main focus for the discussion on AI?

The speaker's main focus is not on how great AI is or how it will evolve, but rather on how to create value with AI. They aim to discuss practical use cases and the importance of applying AI effectively.

Can you provide an example of a successful AI application in the apparel manufacturing industry?

An example of a successful AI application is in the quality control of apparel manufacturing, where image processing technology is used to check if garments are made in the correct proportions. This reduces the risk of mistakes in quality control and helps factories meet the standards of top brands.

What is the concept of digital transformation as discussed in the script?

Digital transformation refers to the idea of using technology to make a significant difference in various aspects of a business. It includes using AI in various use cases to improve products, services, and operational efficiency.

What are some examples of AI applications in improving customer support?

AI can be used to handle a significant portion of customer queries through LLM-based solutions, improving the efficiency of internal support teams, or providing direct assistance to users. It can also be used for churn prediction to align with business efforts to retain customers.

How does the speaker describe the use of AI in creating a seamless user experience?

The speaker describes using AI to make products or websites seamless by anticipating user needs, guiding them, and even performing actions on their behalf without requiring explicit input, thus creating a smooth and intuitive user experience.

What is the importance of verifying the effectiveness of an AI model?

Verifying the effectiveness of an AI model is crucial to ensure it is solving the intended problem and providing value. It involves checking if the model works with the available data, understanding the implications of its predictions, and managing potential risks.

What is the speaker's advice when considering an AI use case?

The speaker advises to first consider if there is a simpler solution that can achieve the desired outcome without AI, as AI can be expensive and complex. If AI is necessary, it's important to clarify the value, verify its effectiveness, and manage the risks associated with its application.

What is the significance of the toothpaste manufacturing company story in the context of AI application?

The toothpaste manufacturing company story illustrates the importance of checking if an AI solution is actually needed and if it's working as intended. It highlights the potential for simpler, more effective solutions and the need to verify the impact of AI implementations.

How does the speaker suggest managing the risks associated with AI applications?

The speaker suggests managing risks by understanding the AI model's predictions, setting guardrails for its operation, and being vigilant about its performance. It's also important to consider the potential for models to be gamed and to monitor for unintended consequences.

Outlines

00:00

📈 The Value of AI in Data Science Insights

The first paragraph discusses the importance of creating value with AI rather than focusing solely on the technology itself. It narrates an anecdote about a fishing company that sought to apply data science to improve its business. The manager instructed the team to only present charts that showed positive trends, leading them to find a chart correlating the weight of fish with age, signifying growth. The speaker emphasizes that the application of AI should be about the value it brings rather than the complexity of the technology. The paragraph also touches on the evolution of AI, mentioning the advent of large language models (LLMs) and the importance of using technology to make a difference, as illustrated by a use case in apparel manufacturing where image processing improved quality control.

05:01

🔍 AI Use Cases for Seamless Experiences and Customer Support

The second paragraph explores various use cases where AI can create value, such as enhancing user experience by making products or websites more intuitive and personalized. It mentions the success of AI in customer support, with one company able to handle two-thirds of all queries through LLM-based solutions. The paragraph also discusses the use of AI in simulating physical systems for analysis and troubleshooting, as exemplified by NASA's practices. Another use case highlighted is the reduction of friction in user interactions, using the example of the author's personal experience with an old iPhone and the convenience introduced by new technology like Face ID. The paragraph concludes by stating that the true test of successful technology application is when it becomes so integrated that it seems to disappear.

10:02

🤖 Challenges and Considerations in AI Implementation

The third paragraph delves into the challenges of implementing AI and the importance of considering simpler solutions before resorting to AI due to its complexity and cost. It suggests asking whether a problem can be solved without AI and emphasizes the need to understand the value being provided by AI applications. The speaker shares a humorous story about a toothpaste manufacturing company that built a complex system to detect empty tubes, only to find out that a simple fan was used to solve the issue. The paragraph advises on a checklist for applying AI models, including verifying their effectiveness, understanding the implications of their use, and managing potential risks.

15:03

📉 Addressing Churn with AI: A Step-by-Step Approach

The fourth paragraph uses customer churn prediction as an example to illustrate the process of applying AI to solve a business problem. It outlines the clear value of reducing churn, the possibility of solving the issue without AI, and the importance of having data to verify the AI model's effectiveness. The paragraph discusses the need for understanding the data usage terms and conditions, the actions that can be taken based on AI predictions, and the potential risks involved. It also touches on the ethical considerations and the potential for AI to exacerbate biases if not carefully managed.

20:07

🛡️ Managing Risks and Monitoring AI Models

The final paragraph emphasizes the importance of managing risks associated with AI models, especially given that they often operate in the background without clear visibility into their decision-making processes. It discusses the need for limited testing, tweaking, and continuous monitoring to ensure AI models perform as expected. The paragraph also highlights the unpredictability of AI models, drawing a parallel to having a fox as a pet, suggesting that while they may be beneficial, they require careful oversight to prevent negative outcomes. The speaker concludes by reiterating that the focus should always be on the value AI can provide, and that careful management of AI applications is crucial.

Keywords

💡Data Science

Data Science is an interdisciplinary field that uses scientific methods, processes, and algorithms to extract knowledge and insights from structured and unstructured data. In the video's narrative, it is mentioned in the context of a fishing company attempting to apply data science to find insights, which underscores the theme of applying technology to create value.

💡Big Data

Big Data refers to data sets that are so large and complex that traditional data processing software is inadequate to deal with them. The video script alludes to the era of data science and Big Data, indicating a time when the volume of data became a significant factor in business decision-making and the need for advanced analytics.

💡Insights

Insights in the context of the video refer to the valuable understanding or knowledge gained from data analysis. The fishing company's team was trying to find insights, which is a central theme of the video, emphasizing the pursuit of meaningful outcomes from data.

💡Technology Application

The term 'Technology Application' in the script refers to the implementation of technological solutions to address specific problems or improve processes. The video discusses that not every application of technology creates value, highlighting the importance of the right application rather than just the use of technology.

💡Artificial Intelligence (AI)

Artificial Intelligence, or AI, is the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. The video emphasizes that the value of AI lies in the outcomes it creates, not just the technology itself.

💡Quality Control

Quality Control is a process by which entities review the quality of all factors involved in production, from raw materials to the final product. The script mentions an apparel manufacturing example where AI was used for quality control, reducing the risk of mistakes and enhancing the value of the technology applied.

💡Digital Transformation

Digital Transformation involves the integration of digital technology into all areas of a business, fundamentally changing how an organization operates and delivers value to customers. The video discusses various use cases where digital transformation is applied to make a difference, including AI.

💡Personalization

Personalization refers to the strategy of customizing an individual's experience with a product or service. In the video, personalization is mentioned as a way to make a product or website seamless, anticipating user needs and enhancing their experience.

💡Customer Support

Customer Support is the provision of assistance to customers in need of information or help regarding a product or service. The script highlights the success of AI in customer support, particularly with LLM (Large Language Models), which can handle a significant portion of customer queries.

💡Churn Prediction

Churn Prediction is the process of forecasting which customers are likely to leave or stop using a company's product or service. The video uses churn prediction as an example of a clear value proposition for AI, where reducing churn by even a small percentage can significantly impact a business's bottom line.

💡Risk Management

Risk Management is the identification, evaluation, and prioritization of risks followed by coordinated efforts to minimize, monitor, and control the probability or impact of unfortunate events. The video script stresses the importance of understanding and managing risks when applying AI, such as the potential for wrong forecasts and the need for careful monitoring.

Highlights

A fishing company's data science team was instructed to only show charts that went up, emphasizing the importance of finding positive insights.

The value of AI is not in the technology itself but in the outcomes it creates, which depend on its application.

The speaker introduces the fourth generation of AI, mentioning the inclusion of LLMs (Large Language Models) in the current toolbox.

A use case from a garment manufacturing factory shows how technology can significantly improve quality control and reduce errors.

Digital transformation involves using technology to make a difference, with AI appearing in various use cases.

AI can be used to make products or websites seamless, anticipating user needs and actions.

LLM-based solutions have seen success in customer support, handling a significant portion of queries efficiently.

Digital twins are used to analyze and troubleshoot physical systems, with a case study involving supermarket layout optimization.

Removing friction in user experiences, such as with Face ID on smartphones, can greatly enhance user satisfaction.

The real test of technology is when it works so well that it becomes invisible to the user.

Biometric systems and adaptive authentication are examples of technologies that can work in the background without user awareness.

Creating value with AI involves considering whether there is a simpler solution that could be more cost-effective.

A checklist for applying AI models includes verifying the value, determining if AI is necessary, and assessing the risks and operational aspects.

Churn prediction is highlighted as a clear value proposition for businesses, as reducing customer churn can significantly impact the bottom line.

The importance of understanding the limitations and potential biases of AI models is emphasized, using the example of university admissions.

The complexity of AI systems requires careful monitoring and management to ensure they are working as intended and not causing unintended consequences.

The final comment stresses that the focus should always be on the value created by AI, rather than the technology itself.

Transcripts

okay so few years

back uh when the at the time of data

science and Big

Data uh there was a fishing

company trying to do data

science and apparently the manager told

them

so there was a team they were trying to

figure out find some insights

Etc manager told them only to show

charts that goes

up so the team get together

tried nothing was going

up tried really really hard and they

find a one

chart excess hor time

Y axis shows the weight of the age

x-axis shows age Y axis shows the weight

of the

fish when the time goes fish

grows

so we can apply

technology but not every application of

Technology creates value

so when we try to do AI actually it's

not about

AI because it's the value that AI

creates outcome very often depend on how

you apply

the your technology AI the

force rather than how hard you are

trying how much you invest how

complicated the

techologies so

today I'm not going to talk about how

great AI

is I'm not talk about going to talk

about how it

happened I'm not going to talk about how

the future will look like because I'm

pretty sure you have heard it

enough so let's try to go and talk how

we can create value with

a so if we look we this is in like

fourth gener fourth incarnation of AI

in we have llms we have the older now we

our toolbox is bigger we had the this

old options and now we have llms rag Etc

bit more

too so let me start with the one use

case I saw this like 10 years

ago which I saw as a real value so I saw

this in a uh uh apparel manufacturing

flow they have the factory flow they are

creating

garments the last step is to the

person a person would take the

Garment there's a board it has certain

Dimension Mark they put the Garment in

there's a camera on

top it'll do the basically image

processing and there's a but button goes

green or

red it's just making sure that the

Garment is done

in the ex right

proportions the technology

wise reasonably simple I mean if you

have done image processing it's pretty

simple

nothing nothing huge but the apparently

it make a huge difference because the

now the quality control the risk of

making a mistake on the cality control

goes way

down and apparently it help also a lot

because like these factories they are

they are creating garments for the top

brands and that question is how do you

make sure it won't like won't get

rejected but the moment this solution

come in it was a easy

cell right so it's again it's not that

technology it is how it is applied made

the

difference so these are some of the

examples of use cases what we call

digital

transformation the idea that you use

technology to make a difference

right so I won't go through everything

so I won't I won't go through every line

in the slides but we'll share the slides

right so you will see that lot of these

use

cases these are example kind of groups

of use cases where you can use

technology to make the difference you

can see the AI appearing in many places

so I'll give I'll go through few

examples the first

one is to use AI to make your product

or website

seamless so the goal is to make the user

feel like

that anticipate him show him show him

where he

is help him if possible even do the

right thing without asking

him this has been for happening for a

long time right personalization

recommendations

Etc okay so that's the first

one the second one customer support this

is a topic that we saw like lot of

success with llm based Solutions

recently for example this

company they were able to handle about

2/3 of all

queries so you could use so there are

many ways to use this you could use it

internally to improve the used by your

internal support team to make them

efficient or you may put it outside to

the user

directly or you may use churn prediction

and align it with your

efforts another

one so okay how many of you have watched

either Martian or Appo 13

okay so you might remember that in both

they had this physical TN right when the

the shuttle was in

trouble the they basically brainstorm

with the physical model to figure out

the answer so the NASA has been doing

that for a long time the idea is that

whatever the system physical system you

have you create a digital copy that you

would use

to analyze understand

and sometime when you troubleshoot you

can actually deal with the exact model

right one one case study is that

basically they were changing how the in

the supermarket how they lay out the

Isles and the checkout process so they

could

simulate the older version I mean this

the cooler version older version is

called simulations right so you simulate

you basically model the system you

simulate there are a lot of use cas

around

this so another variation is the

removing

friction I had a personal experience of

this so

I I used to use a very old phone very

old iPhone I'm like usually phone is a

phone I buy a new one when it fall

apart I had to buy new one recently

because the Ws has a app to submit Leu

to open the doors and I'm like I because

my phone is very old I'm like the guy

who debugging everything because nothing

works so I decide to buy a new one I

brought the

one and actually the face ID thing blew

me away because it's so natural and then

I realized oh the that one button they

had that's G

too so it's like

I mean we think we are like

logical beings but we are so sensitive

to this emotional things and if we are

so same for the users right so if you

can remove this

friction it can make a big difference so

there are lot of use cases around that

so the the the real real test

is if we really using the if this really

working technology disappears you don't

see it anymore

right so now that whatever the button is

gone everything going you look at it oh

it turn up and

works the next

one sec

systems if you want to get into a subset

of in computer science or in

uh computer related that never goes away

this the

topic so

use cases go from

biometric face ID these kind of things

to trying to look at what's going

on and do adaptive authentication

authorization look risky ask for more

things I mean he logged in yesterday he

came back for same machine same

fingerprints fine let him in

okay okay

so so

far we talk about how do how do you

create so the doing AI means creating

value right so if you if if there's one

thing I want you to remember that's the

most important

thing so and we went through several use

cases and discuss so let's get to okay

how to make this happen what are the

challenges so

when somebody come and told tell me

about AI use

case my first reaction almost always is

that is there a way to do it without

Ai and it's actually throw some people

off because it's like so but the thing

is I know how expensive AI is I know how

much work if there is a simpler

solution it'll save you a lot of time so

it's in my opinion that's the first

question you want to

ask

so I mean there are a lot of lot of

stories but the I'll tell one toothpaste

Manufacturing Company nobody knows

whether this true because it's

everywhere and there are variations

toothpaste Manufacturing Company once in

a while toothpaste go out without the

toothpaste what do you do so they build

a very

complicated measuring

thing to detect and put a

warning they put it in

Great CEO checked after a few months no

complaints okay everything look good no

more complaints

working then he go and check has the

alert has been ever

fired it has only been fired

once so he was okay what's going on I

mean seem to be happening but the air

thing is doing nothing but how does it

get fixed so he go and go to the

production line and then somebody said

ah

oh okay I I can tell you so there was a

fan running next to

the the production line uh because the

guy when the alarm goes off the guy has

to come and remove it so he figured a

simple solution run the fan the thing

will get blown away problem

solved so this this a this a famous

engineering story nobody knows whether

it's actually true but anyway the moral

is true right I I to me it's a it's a

one thing to remember if you put the

model check actually what is

doing and sometimes I mean if it works

fine but the that'll give you the

understanding so this is a

checklist to try to kind of go through

if you're trying to apply a model

first is the value clear what's the

problem you are trying to solve if

there's no problem don't solve it the

second for that problem can you solve it

without

AI answer is yes it'll be

cheaper the Third how do you know that

it's working with llms actually you can

work with less data but still to make

whether know whether it's work for most

use cases you need some data to

verify and of course you need to know

whether can I use that data what's the

copyright what's the uh what's the

policy is it in the terms you want to

know next if the AI works if I get that

answer what I'm going to do with

it is that clear can we Implement that

what can go

wrong and finally how we are going to

operate

it so let me quickly take CH prediction

as a motivating

example

value value is clear because the the

churn so most businesses okay it depends

on what kind of businesses it is but

most companies most businesses would

spend lot of money to acquire a

customer it takes many years to recover

that

expense and a happy customer will keep

giving you money versus a

churn is a disaster because you had like

burned a lot of money to even get that

right so I mean you can you can put

simple numbers and it make a if you can

reduce J by 1% 2% the bottom lines move

significantly okay yes so clear value

can we Sol it without the I I mean it up

to the your Chief custom officer maybe

he has great

intuition you need whatever model you

build in this case you need to compare

against what you're doing right

now how do you know whether it's working

yeah most likely you'll have the data I

mean

if you if you are not tracking

the customers you won and lost I mean

it's a disaster so usually you have the

data um can you use that data here okay

depends on your terms go and check your

terms and conditions is it allowed to do

that what action can we can you take

give more attention give them offer Etc

you need to think it's true can they

game it but look

okay the

risks I mean now if you're going to give

them a huge

discount it there may be a risk but

generally if you go and so let's say you

get

a get a forecast that is 70%

accurate right

most probably it'll work out because the

there's no harm in go and be nice to the

other guys too so in this case it's

fine so with

that you we need to dig to the risks

right so we know we have heard about all

those risks

so that's a very important point to

understand each of these lot of these

especially buyers the fourth one humans

are less accurate than most

models but there's a catch though

because for example if human being being

biased each guy buys in little bit

different

way it's kind

of cancel it out let's say I mean let

let's say that that's a great model to

predict whether a student would finish

their graduate

program let's say it has great

accuracy what will

happen all the universities will use

it now what will happen

now if you are the that one person where

it is accurate you can get in you can't

get to any University

Now

versus maybe before that the admission

committee may be

biased maybe there are 10%

wrong but each one would be wrong

differently so you will get to one

if you are okay I mean so there there I

said that there's no chance but there's

a there's a gray area right so the so

these are the challenges right because

the moment you put the AI in usually it

goes to the background you don't see

what's going

on and then everything start to do the

same thing so these are the complexity

so these three are common kind of okay

types of use

cases the first one

risk of a wrong forecast

is fine Google Amazon recommendation if

you don't like it you don't pick it

fine the variation of that yes there may

be risk but the person can they know

they can select there's a reasonable

chance the third

one even critical cases sometime I can

work

because you can collect enough data you

can build a model you can take time to

train self driving Etc so usually these

three works and there are other cases

true but the it's it's kind of useful to

realize it and understand why it's

working okay so this is the

usual process Malik will talk through

this in lot of detail right so I I so

the the you build the model but you need

to verify usually you want to get it to

some limited testing mode there you

tweak you make sure it works

Etc and then continue to

monitor the

final

comment unfortunately most models we

don't know how they

works now yes AI is mimicking

us which is great the trick is they

don't fail like

us right so for example this flash

crises at stock market there's 100 so

there are $100,000 books in

Amazon

so so because we don't know exactly how

they works we had to be careful right I

say like it's little bit like having a

fox as having a fox as a pet in your

house I mean it may be okay but you need

to keep counting the chickens they may

be disappear and they're

crafty so you want to monor them put the

guard rails kind of ask okay what's the

craziest prediction it can give what

will

happen Okay to wrap up it's not about

AI it's the value right you always need

to look from the

value if you can realize that value from

other means it'll be

cheaper then think true how you going to

verify what can I do with the results

what can go

wrong and be carefully manage the risk