AWS Summit SF 2022 - Solve common business problems with AWS AI/ML (AIM103)
AWS Summit SF 2022 - Solve common business problems with AWS AI/ML (AIM103)
In this session, learn how companies across all industries are using AI to address use cases that create measurable results. Results fall into four categories: enhancing customer experience, enabling employees and organizations to make better and faster decisions, improving business operations while reducing cost, and creating completely new products and services powered by AI and ML.
Learn more about AWS Summits at https://go.aws/3zwaA9T.
Subscribe:
More AWS videos http://bit.ly/2O3zS75
More AWS events videos http://bit.ly/316g9t4
ABOUT AWS
Amazon Web Services (AWS) hosts events, both online and in-person, bringing the cloud computing community together to connect, collaborate, and learn from AWS experts.
AWS is the world’s most comprehensive and broadly adopted cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—are using AWS to lower costs, become more agile, and innovate faster.
#AWSSFSummit2022 #AWSSummit #AWSAMERSummit #AWS #AmazonWebServices #CloudComputing
Content
0.239 -> (uptempo music)
8.49 -> Good afternoon.
9.36 -> Good afternoon, everyone.
10.53 -> Thank you very much for coming to this session.
13.57 -> My name is Albert Esplugas
15.26 -> and I'm the head of AI Solutions Marketing
17.39 -> at Amazon Web Service.
18.253 -> It's a pleasure to be here with you today.
20.97 -> As a reminder, this is a non-technical introductory session
24.55 -> in which we will cover what is AI
26.3 -> and what are the top AI/ML use case
28.213 -> that we see among our customers.
33.54 -> Not so long time ago,
35.28 -> machine learning was basically used by
37.3 -> very high tech companies
39.03 -> and some academic researches,
41.14 -> but with introduction of cloud computing,
43.7 -> machine learning went from an aspirational technology
46.33 -> to mainstream adoption,
48.56 -> and we're seeing now the impact of AI
50.3 -> across all the industries.
51.88 -> These are some numbers from IDC,
53.46 -> a forecast of spending in 2024.
56.097 -> $110 billion in AI soar
59.13 -> and 75% of the companies right now
62.88 -> are just doing their proof of concepts
65.03 -> and trying the first use case
66.6 -> moving to operationalizing AI
68.83 -> and adopting more than one AI use case.
71.09 -> And in a survey that Deloitte conducting among CxOs,
74.43 -> 53% of the ones that were adopting AI
76.55 -> were saying that AI was really
78.01 -> transforming their organization,
79.8 -> not just giving them optimization but also transformation.
84.43 -> So let's start, settle back,
86.41 -> couple of minutes just to give a quick overview
88.38 -> of what is AI,
89.53 -> and I'm sure that you're all familiar with those terms
91.44 -> but I think it's good to just review them.
93.36 -> Artificial intelligence is any system
95.3 -> that is capable of reproducing a task
98.5 -> that previously required human intelligence.
101.11 -> That will be kind of very high level definition
102.83 -> of what is artificial intelligence.
104.66 -> Artificially intelligence as mentioned
106.04 -> has been used for more than 50 years
109.55 -> but is now with a cloud option that is going mainstream
112.23 -> and is being adopted by large organizations.
114.84 -> The main technique under artificial intelligence,
117.09 -> the one that today is used
119.1 -> in all the AI/ML projects is machine learning.
122.02 -> Machine learning takes a probably strict approach.
124.48 -> It uses a lot of data,
125.85 -> use that data to train the model,
127.97 -> and once you have that model that is trained,
129.59 -> you're capable of adding more data
131.73 -> and predicting the outcome
133.05 -> based on the initial model.
134.57 -> So it's a process that use large amount of data,
136.58 -> that data needs to be clean,
138.31 -> that data needs to be unbiased.
140.15 -> And then you use that data to train that model
142.12 -> and that's machine learning.
143.43 -> It's the main tool,
145.09 -> the main technology that is used today
146.95 -> for any artificial intelligence project.
149.86 -> There's two different techniques under machine learning
152.42 -> that are also very popular,
154.16 -> the first one is deep learning.
156 -> Deep learning works a little bit like the brain
157.86 -> creating like multilayer neural networks.
161.648 -> And it's basically used for things like
163.33 -> computer vision or speech recognition.
166.73 -> Using a large amount of data
168.21 -> to train those neural networks
169.64 -> and then being capable of doing that kind of
172.31 -> computer vision or speech recognition systems.
175.3 -> And then second system that is also used today
177.6 -> is something that we call reinforcement learning.
179.81 -> This takes a completely different approach.
181.75 -> It's more about a reward function
184.48 -> in which we reward the system
186.14 -> and train the system through those rewards.
188.67 -> That's the typical technology
190.49 -> that is used for autonomous vehicles,
192.71 -> for robots, for drones and so on
194.653 -> because you train them,
196.11 -> usually you train them by using simulation,
198.34 -> and with that simulation
199.5 -> you train that reinforcement learning,
201.3 -> and then the model learns
202.84 -> and then is capable of managing
204.047 -> and being completely autonomous.
205.95 -> So those are kind of four big terms
207.97 -> that we use a lot in the AI space.
210.26 -> Artificial intelligence big umbrella.
212.61 -> Machine learning, the main technology that we're using today
215.2 -> and two subtechnologies
216.47 -> which are deep learning and reinforcement learning.
220.67 -> How Amazon service is helping our companies to adopt AI?
224.59 -> The first way is making sure that we democratize AI,
227.04 -> that we make AI accessible to everyone.
229.55 -> So you basically have like three options.
231.64 -> Those companies that have ML practitioners,
234.21 -> data scientists,
235.28 -> usually what they do is they use
236.61 -> one of our products called Amazon SageMaker
239.21 -> and they use SageMaker to ingest the data,
241.95 -> clean the data, process the data,
243.7 -> train the model, deploy the model,
245.5 -> and then do all the inference.
246.85 -> So it's an end-to-end tool
248.08 -> that is used for building your model.
249.93 -> You need to have people that are used to building ML models
253.23 -> that's the data scientists.
255.31 -> If you don't have those data scientists
257.69 -> but you want to add AI capabilities to your application
260.16 -> or maybe you have them
261.48 -> but you have some time to market constrain,
263.81 -> a very good option is what we call
265.5 -> our pre-trained managed AI service.
268.06 -> Those are pre-trained models
269.49 -> that we use for things like personalization,
273.07 -> voice recognition, text to voice,
276.35 -> visual recognition,
277.45 -> and you'll see later there's more than
278.74 -> 25 different AI service
280.9 -> that you can add to your applications
282.54 -> by just using an API.
284.97 -> And then third you have new solutions
287.15 -> like the one that we presented at re:Invent last year
289.76 -> which are what we called low code/no code tools
292.34 -> which are more meant for business analysts.
294.22 -> People that understand their data,
295.67 -> understand their domain,
296.97 -> they are domain experts.
298.23 -> They are not ML practitioners,
300.06 -> they are not developers
301.75 -> but they are capable of using that tool,
303.56 -> that tool is called Canvas.
305.55 -> They are capable of using that tool
306.96 -> to train their own models,
308.42 -> get the output,
309.26 -> and then use it within the organization.
311 -> So, three different approach
312.26 -> depending on where you are in your ML journey,
314.65 -> depending on what is the level of ML skills
316.99 -> that you have in your organization.
318.33 -> But also depending on
319.62 -> if it makes sense to build your own model
321.5 -> or you can use a model that is already being pre-trained
324.54 -> and you can easily add to your application.
328.33 -> This is our stack,
329.22 -> this is all our AI/ML products in a layered system
332.88 -> that I'm gonna explain.
333.72 -> The lower level is the one that
335.51 -> companies that already have their frameworks
337.63 -> and their ML models
339.08 -> and want to bring their models to the cloud platform,
341.98 -> they can run it on these infrastructure.
343.9 -> And you have specific CPUs and GPUs
347.83 -> and systems to be able to run those model super efficiently.
352.26 -> Second you have SageMaker, the one that we mentioned.
354.68 -> If you want to build your own model
356.21 -> you use SageMaker
357.49 -> and that's the end-to-end tool that I was mentioning before.
360.43 -> And then at the third level
361.52 -> you have all those AI service
363.05 -> and you see that we have the basic ones,
365.59 -> text and document processing,
367.57 -> chatbots, speech recognition, vision,
371.27 -> and then a more complex, more specific
373.28 -> like personalization,
375.1 -> like forecasting,
376.49 -> or like more in industrial space,
380.42 -> predictive maintenance and other solutions
382.02 -> are very specific to those industries.
384.15 -> And then finally we have that tool that I was mentioning,
386.28 -> Canvas.
387.113 -> This is a tool that is in the SageMaker level
389.1 -> because it's very connected to SageMaker
391.323 -> but it's meant for people
392.17 -> that are not data scientists
393.98 -> but is kind of more a BI tool, a no code tool
396.31 -> that you're capable of ingesting that data and training.
399.36 -> So, this is the Amazon stack
401.1 -> covering all different kind of levels
403.6 -> and layers in that stack
404.76 -> for different kind of audience
406.35 -> and with all those pre-trained models
408.04 -> that I mentioned before.
411.28 -> Amazon has been using machine learning
412.75 -> for more than 20 years.
414.5 -> We're using machine learning in the amazon.com store
417.01 -> for older personalization and product recommendations.
420.18 -> We're using it in the fulfillment centers
422.24 -> for managing all the inventory
423.97 -> and all the robots that are going to fulfillment centers.
426.89 -> We're using it with Alexa,
428.87 -> Alexa is using heavily AI
430.41 -> and is capable of understanding
432.45 -> who's speaking, what is the intention,
434.58 -> and then helping that user.
436.2 -> And then finally, another example would be
438.1 -> the drones you have,
439.25 -> that we have a delivery system using drones.
441.44 -> Well, those drones are also using reinforcement learning
443.82 -> as I was mentioning before.
445.16 -> These are just four examples
446.9 -> that Amazon is applying internally.
448.73 -> There's many others
449.89 -> and all that experience is being brought
451.56 -> to the products that we offer
453.01 -> in the Amazon Web Service stack that I shown you previously.
456.66 -> But beyond Amazon, there's more than 100,000 customers
460.13 -> are using and running their AI/ML workloads
463.05 -> on Amazon Web Service.
464.76 -> And what I'm gonna do today is
466.52 -> to explain you some examples
467.92 -> of how they are using those technology
470.14 -> and that AI/ML within the organizations
472.17 -> and what is the kind of
473.003 -> the business value that you'll get.
476.17 -> Usually when you adopt AI in organization
478.23 -> you have like four main business outcomes.
480.94 -> The first one could be enhancing your customer experience.
484.1 -> Making that your experience is better,
485.92 -> it's frictionless,
487.06 -> it's more innovative,
487.92 -> and it gives more engagement at the user level.
491.41 -> The second category is more about
492.96 -> what we call augmentation.
494.11 -> It's more helping your employees or your customers
497.63 -> to make better and faster decisions
500.5 -> and I'll show you later some examples.
503.21 -> The third category is more about optimization.
505.9 -> Optimization and automatization
507.47 -> of things that happen more in the backend
509.46 -> that can be optimized and can be automated using AI.
512.83 -> And finally there's a fourth one
514.33 -> which is more the cutting edge innovation
517 -> which is those things that we're previously
519.19 -> completely impossible without AI.
521.05 -> New service and new problems
522.81 -> that exist because they are using AI
524.84 -> and were completely impossible in the past.
526.85 -> Let's see some examples for each of those categories.
529.61 -> For customer experience you have contact center intelligence
533.03 -> and I'll give you more details later about what it is
534.91 -> but it's basically being able
536.52 -> to add intelligence to your contact center
539.35 -> so you can understand what is the voice of the customer,
541.7 -> you can help the agent and so on.
544.22 -> Chatbots and virtual assistants,
545.76 -> kind of the way to also have that
547.43 -> cell service mode with your customers
549.37 -> and help them through chatbots or virtual assistants.
552.62 -> Personalization, personalization in your UI,
555.49 -> in your application, in your store,
557.36 -> personalization in your marketing message.
559.53 -> Everything is target to your specific audience.
562.67 -> Content moderation,
563.56 -> so when users are bringing you content,
565.75 -> this system is capable of doing content moderation
567.86 -> and identifying if there is something
569.55 -> that shouldn't be there and will be removed.
571.85 -> Fraud prevention, that's more about
573.59 -> avoiding online fraud on your platform,
576.2 -> on your store.
578.33 -> Moving more for the decision-making.
580.41 -> You have intelligent search
581.63 -> and I'll show you later what is intelligent search
583.48 -> but is adding more accuracy
585.21 -> and natural language understanding when you do that search.
588.86 -> Forecasting, this is a very popular AI use case.
591.7 -> It's probably the number one
593.18 -> that is being adopted among companies.
595.16 -> Being able to predict for some specific time series values
598.63 -> what's gonna be the value.
599.86 -> And I'll give you more details later.
601.67 -> Anomaly detection, finding in a time series
604.27 -> in an information when suddenly you have an outlier.
607.22 -> What happens? What's the reason?
608.97 -> And then help you address that specific problem
611.85 -> or that specific opportunity
613.17 -> because maybe that outlier
614.73 -> is an opportunity for your business.
617.14 -> On the process optimization and automatization
619.45 -> you have a very important one
621.44 -> which is intelligent document processing.
623.8 -> Extracting information from documents
625.76 -> and processing that information within your organization.
628.99 -> Demand planning which is connected also to forecasting
631.42 -> but more on the backend.
633.13 -> Predictive maintenance super important in industrial space.
636.2 -> Quality control using vision recognition.
639.15 -> And also using AI for DevOps.
640.92 -> So, how you can apply AI
642.84 -> to have a better software engineering
644.46 -> within your organization.
646.06 -> The last one is more that innovation
648.74 -> that I was saying before.
649.64 -> Those use case don't exist yet,
651.22 -> those use case are being built by our customers
653.68 -> by using this machine learning techniques
655.52 -> to build completely new different products and services.
660.87 -> Depending on the industry,
662.27 -> some use cases will be more relevant and less relevant.
664.38 -> I'm giving you here some examples
666.36 -> for the healthcare industry,
668 -> everything that is diagnostic,
669.75 -> everything that is helping with computer vision
671.82 -> in identifying a cancer in a specific medical image.
675.95 -> Everything that is more about
677.36 -> interacting with a patient
678.68 -> and being able to extract notes
680.27 -> and insights from those conversations.
682.36 -> And relations with your patients and patient engagement.
685.96 -> Those are typical solutions in the healthcare industry.
689.28 -> For manufacturing and industrial
690.86 -> the main one is predictive maintenance.
692.78 -> Being able to predict one specific equipment will fail
695.88 -> and then be able to change it before it fails
697.85 -> so you would reduce the downtime.
699.64 -> Or other space like workplace safety
703.03 -> or quality control of the products using vision recognition.
706.92 -> Financial service a lot about document processing,
710.22 -> there's a lot of documents in the financial industry.
712.12 -> Forecasting, credit scoring
714.92 -> and all that kind of more financial peer process
717.51 -> done with AI/ML.
719.31 -> Retail, a lot of personalization, recommendations,
722.22 -> what are the products that cross-sell, upsell
724.33 -> that you can do with that kind of recommendations.
727.14 -> And then finally in the media and entertainment industry
729.61 -> is also a lot about personalization,
731.293 -> it's about content moderation,
733.42 -> it's about content monetization,
735.53 -> and is about analyzing those videos,
737.56 -> those assets that you have,
738.79 -> and identifying specific brands
740.44 -> with that kind of content.
741.42 -> So, there's a lot of very specific solutions
744.3 -> for the media and entertainment industry.
748.35 -> Let's go in deeper detail in some of them.
751.83 -> Contact center.
752.663 -> I mentioned the contact center is adding AI capabilities
754.97 -> to your existing contact center
756.117 -> and there's kind of three big sub-use case.
758.94 -> The first one self-service virtual agents.
761.69 -> A system that is capable is a chatbot
763.46 -> or a virtual assistant that is capable of
765.74 -> attending your customers,
767.02 -> answering the basic questions.
768.69 -> You train those systems with your FAQs
771.08 -> and with all the documents
771.99 -> that you have from your organization.
773.51 -> You will train that chatbot
775.02 -> and that chatbot will be able to attend your customer
777.5 -> and solve the basic problems,
779.28 -> and then route that customer to a personal agent
782.72 -> if they need more help.
784.85 -> Second one is what we call real-time call analytics.
787.44 -> That's during the conversation with your customer
789.95 -> the system is listening to the conversation
791.82 -> and identifying problems.
793.42 -> It's first helping the agent
795.11 -> by providing them answers to all the questions
797.26 -> that the customer is asking.
798.75 -> And guiding that agent
800.17 -> and training that agent at the same time,
802.18 -> and also at the same time if there is
803.56 -> a problem in that conversation
804.85 -> that will scale to that supervisor
807.53 -> that will go and try to help that specific agent
809.83 -> with that customer.
810.91 -> So that's real-time process of those conversations
813.54 -> and helping the agent.
815.27 -> The third one is what we call post-call analytics.
818.16 -> That's while you have all those recordings
820.44 -> about all those conversations with your customers,
822.507 -> you're capable of extracting insights,
825.42 -> extracting sentiment.
826.86 -> Understanding what are the main questions they are asking.
829.46 -> Understanding what are the main problems
831.53 -> and the main friction.
832.57 -> So there's a lot of value from a voice of customer
834.78 -> in taking all those audios that have in your call center,
837.83 -> analyzing them, and extracting those insights
840.28 -> that will help you to learn
841.24 -> a little bit about what is your consideration
843.52 -> from a customer perspective,
844.56 -> what are the problems.
846.95 -> Example, Maximus is a company that provides,
849.33 -> manage problems like Medicare and Medicaid.
851.97 -> They use one of our partners, SuccessKPI,
855.75 -> and they implemented the one that I was saying,
857.34 -> this call analytics.
859.74 -> So listening to all the interactions
861.26 -> that they have with their customers,
862.67 -> learning from those interactions,
864.32 -> and then being able to help and scale through our agents,
867.55 -> and providing them the right information
869.24 -> so they could help their customers
870.67 -> based on those learnings.
873.28 -> Another example will be Ryanair.
874.83 -> Ryanair is created a chatbot in six different language
878.15 -> capable of listening or sorry,
879.78 -> helping their customers in a 24 dot seven.
882.55 -> No need for any human in that first conversation.
885.72 -> Helping the customer and if necessary,
888.25 -> they route it to the corresponding agent.
890.82 -> And they've been doing this since 2018
892.62 -> and more than three million conversations
895.07 -> through this chatbot developed by this partner.
899.65 -> Many other customers across many different industries.
902.68 -> I would say that chatbot is by far
904.5 -> one of the main use case adopted by most companies
907.41 -> because it solves kind of the basic problem
909.52 -> in that 24/7 self-service service.
913.98 -> But also virtual assistants a little bit more advance
916.91 -> and that kind of call analytics
918.23 -> is something that we're seeing also across the board.
922.58 -> Second big use case
924.05 -> what we call intelligent document processing.
926.62 -> All companies are using documents
928.43 -> and they process the information in those documents.
930.043 -> Some of those documents are PDF,
931.83 -> some of those documents are scanned documents.
934.35 -> And they need to be processed within the organization
936.7 -> and we need to extract the information
938.19 -> into those documents.
939.34 -> Intelligent document processing does this exactly.
941.71 -> It's capable of looking at the document
944.04 -> even if it's new match,
945.05 -> not doing a normal OCR process
947.93 -> but extracting the insights.
949.65 -> If there is a signature,
950.69 -> if there is a photo ID,
952.26 -> if there is a table,
953.31 -> if there is some specific paragraph that we need to analyze,
956.66 -> it's capable of understanding the structure of the document,
959.41 -> extracting those insights,
960.91 -> understanding what this document is saying,
963.81 -> and then processing within the organization.
966.27 -> Have three big industries
967.7 -> that are using a lot these IDP solutions.
970.04 -> The first one is the financial industry.
972.01 -> Mortgage processing is a good example.
974.19 -> You need to provide a lot of documents
975.8 -> for a mortgage to process
977.2 -> and this system is capable of understanding those documents,
980.15 -> extracting the main data that needs to be extracted,
982.66 -> and enter that data in your workflow
984.88 -> instead of having a human doing that kind of process
987.73 -> and entering that information manually.
989.5 -> Or instead of asking your customer
991.39 -> to do that process and enter that information.
993.94 -> Healthcare industry, all the insurance claims,
997.16 -> that's another big area,
998.5 -> a lot of documents that need to be processed,
1000.22 -> a lot of documents containing
1001.91 -> confidential and medical information
1003.54 -> that needs to be treated sacredly.
1005.42 -> That's another big area in which IDP is being used.
1008.29 -> And then across all organizations
1010.08 -> all across all organizations are using documents.
1012.51 -> A good example will be manufacturing or retail.
1015.38 -> There's many different,
1016.42 -> this is a very horizontal use case
1018.27 -> that is being used across all industries.
1022.05 -> One example Intuit.
1023.73 -> Quite familiar those days.
1025.29 -> If you've submitted your taxes using TurboTax
1027.89 -> you know that there is a process
1028.93 -> in which you can option one,
1030.22 -> connect to your financial institution
1032.15 -> and it will pull the data.
1033.51 -> Option two, if you have your W-2
1035.65 -> or your 1098 document,
1038.49 -> you just upload the document
1040.14 -> and then TurboTax is capable of extracting that information.
1043.11 -> If you've seen multiple of those documents
1044.62 -> you'll see that those documents
1045.5 -> are not exactly always the same.
1047.62 -> So, AI is being used to understand a document
1050.32 -> and extract that data
1051.47 -> even if the format is changing.
1053.45 -> The format doesn't need to be always the same
1055.39 -> and that's the problem with the traditional
1056.88 -> document processing solutions,
1058.66 -> that once the format changes, the system breaks.
1061.21 -> This one is capable of looking at the document,
1063.58 -> understanding what has changed,
1065.14 -> understanding what is the data
1066.31 -> that needs to be extract.
1067.44 -> Extract that data and pull it in the system.
1070.39 -> So, good example will be Intuit.
1072.31 -> Another example is Anthem.
1073.87 -> This is more about processing claims.
1075.577 -> It has more medical information that is extracted.
1078.73 -> And that in the past they were at
1080.43 -> 80% automation of the process
1082.29 -> by using this technology.
1083.525 -> They moved to 90% optimization.
1085.7 -> There's always some part that will
1087.52 -> need some human intervention.
1089.62 -> That means that all those IDP processes
1092.03 -> include also the option of whenever there's a document
1094.4 -> that cannot be processed,
1096.21 -> it's sent to a person that will process that document
1099.1 -> and then fill it back to the system.
1100.96 -> But increasing that 80% to a 90%
1103.71 -> has a huge implication from a cost perspective
1105.99 -> and productivity.
1109.57 -> As you see, many different customers
1111.14 -> across different industries using IDP solutions
1114.85 -> for their very specific industry elements.
1119.08 -> Personalization, that's also a big one
1120.89 -> that we see across the board.
1122.8 -> Use personalization to acquire new customers,
1125.73 -> to retain your existing customers
1127.51 -> by engaging them and increase the engagement,
1129.64 -> by putting information that is more relevant
1131.74 -> for those customers.
1133.08 -> For increasing the discoverability
1134.65 -> of some specific products
1135.92 -> because you're suggesting products
1137.33 -> based on the preference of that user,
1139.09 -> and the profile of that user.
1140.8 -> And at the end that will increase efficiency,
1143.27 -> will increase conversion rates,
1145.2 -> and that will increase revenue.
1147.33 -> One of the industries that is using a lot personalization
1150.71 -> is the media and entertainment industry
1153.24 -> and I'm sure that you're all familiar
1154.68 -> with looking at your channels,
1156.52 -> and having all those videos being suggested
1159.22 -> based on the movies that you've seen in the past,
1161.52 -> based on your profile, your preference,
1163.27 -> it's suggesting from thousands and thousands
1165.56 -> of movies available.
1166.58 -> It's suggesting Pulse Live or Discovery Plus
1170.54 -> are examples of companies that are doing exactly this
1173.06 -> using our ML capabilities.
1175.39 -> And then with that you'll increase content views,
1177.91 -> you'll increase the retention of those users,
1181.09 -> you acquire new users,
1182.74 -> and you'll be able to basically
1184.96 -> increase the relevance of the content
1186.44 -> that you're putting in front of your customer.
1189.42 -> Another big industry, retail.
1191.55 -> And in retail that's more about
1193.8 -> discoverability of the products,
1195.03 -> making sure that you're giving the right recommendation
1197.35 -> that you do upsell and cross-sell
1199.52 -> across all your products
1200.93 -> based on who's the user in front of you.
1203.45 -> What is his previous history consuming products,
1205.87 -> buying products,
1206.98 -> and other users that are similar to this one,
1209.12 -> what are the products they were also buying?
1210.96 -> That increase heavily the number of sales per product
1213.7 -> and discoverability of those products.
1216.54 -> And it can be applied everywhere.
1217.91 -> So there we'll have for examples Bundesliga
1219.9 -> who's the German soccer association.
1222.12 -> They have an application for engaging with the fans
1225.07 -> while they're also using the product behind personalization
1227.97 -> which is called Amazon Personalize
1229.72 -> to suggest the most engaging content
1231.62 -> and increase the time that the users spend
1233.94 -> consuming the content on their phone app.
1239.77 -> Bundesliga 67% increase in the article reads per user.
1244.99 -> Marc O'Polo increase in the number of purchase
1247.41 -> by using personalization.
1249.13 -> And Lotte Mart 40% increase
1251.17 -> in the products that customers buy
1252.99 -> in the first interaction that they have with the system.
1256.3 -> All those systems are provided pre-trained.
1258.91 -> This is Amazon personalized.
1260.32 -> It already knows exactly how to do personalization
1262.34 -> in the retail scenarios,
1263.9 -> in the media and entertainment scenarios.
1265.69 -> What it does is it pulls your data
1267.18 -> to personalize for your specific data.
1268.917 -> But those sub-use case are already pre-trained.
1274.62 -> Another big scenario, big use case is fraud detection.
1278.5 -> And fraud detection is used across
1280.74 -> any online activity that requires a payment
1283.86 -> or that there is a loyalty program,
1285.66 -> or some promo codes that need to be processed.
1288.14 -> Then you need to use fraud detection.
1290.36 -> In the past it was more will-based.
1291.94 -> The problem is option one,
1294.66 -> your system is not good enough
1296.23 -> which means that you will lose money.
1297.76 -> Option two, your system is too good
1300.55 -> or maybe not good enough to detect false positives
1303.28 -> and then you have a very bad user experience
1305.37 -> because that transaction is denied
1307.34 -> when that was probably not a fraudulent transaction.
1311.39 -> By using AI you're capable of learning and understanding
1314.49 -> and seeing exactly what is the situation
1316.89 -> and the kind of fraud.
1318.16 -> For different scenarios like new account sign up,
1320.74 -> like online checkout,
1322.52 -> detecting fraudulent cards,
1324.3 -> people trying to enter multiple times for the same card.
1327.64 -> Fake accounts,
1328.79 -> accounts created by a bot,
1330.35 -> and everything that could happen on the online world
1332.22 -> from a fraud prevention point of view.
1336.25 -> This is a product called Amazon Fraud Detector
1338.39 -> that is used to do fraud detection scenarios.
1341.38 -> Example Omnyex.
1342.213 -> Omnyex is a company that manage different gaming platforms,
1346.22 -> one of them CDKeys.
1347.93 -> And they use Fraud Detector
1349.63 -> and with Fraud Detector they move the level from 10%
1353.12 -> that require manual intervention to just 1%.
1356.43 -> That means that 9% additional
1358.26 -> were being able to process using fraud detection,
1361.07 -> not being sent to humans.
1362.62 -> That means increasing the productivity,
1364.32 -> reducing the cost,
1365.43 -> and lowering the losses because of that fraud.
1369.6 -> Some additional companies using
1371.31 -> examples on different industries
1372.92 -> using fraud detection in their organization.
1377.62 -> Very connected to fraud detection is identity verification.
1381.11 -> That's something that is used for
1382.82 -> many different scenarios.
1383.95 -> The typical one is you use your license ID
1387.8 -> or your passport,
1389.18 -> and the system is capable of looking at that information,
1391.93 -> extracting the information,
1393.05 -> looking at your face in the passport,
1394.8 -> looking at your own face
1396.55 -> and checking and validating that you're the right person.
1399.24 -> And it can be applied for online proctoring,
1402.18 -> for gig economy work with the phone application,
1405.34 -> for transportation, customer onboarding,
1408.9 -> or even retail online e-commerce
1412.09 -> where you're validating the user identity.
1414.57 -> That's using facial recognition
1416.45 -> and also ID verification and comparing.
1419.66 -> Have a very nice example is a company called Aella Credit
1422.83 -> and they give credit for people that are,
1426.08 -> don't have a bank account.
1427.75 -> And with that what they're capable of doing
1429.38 -> is verifying the identity with the user
1431.57 -> and whatever passport or ID that they have,
1434.82 -> validate that is exactly the same person.
1436.8 -> And then being able to process that loan
1438.49 -> or being able to process that financial transaction
1441.06 -> without requiring them to have that account
1443.79 -> in that financing institution.
1448.3 -> Some other examples of identity verification
1451.05 -> in different industries and different examples.
1455.15 -> Forecasting.
1456.32 -> That's also a big one,
1457.47 -> as I mentioned before is a very popular use case.
1460.17 -> It allows you to predict
1461.36 -> what's gonna be the value of a time series
1463.5 -> within your organization.
1464.89 -> And it can be applied for planning your inventory
1468.26 -> and knowing exactly if you have enough
1469.77 -> of each product and what should be the product
1471.8 -> that you should be provisioning in your inventory.
1474.26 -> Workforce planning,
1475.32 -> knowing exactly how many people
1476.93 -> you will need a specific date
1478.38 -> based on what is the demand
1480.41 -> and the prediction of the demand for that specific date.
1484.4 -> We can also use it for capacity planning
1486.43 -> more from what is the capacity
1488.63 -> that I will need to produce based on that specific demand.
1491.64 -> And also for financial planning,
1493.05 -> that sales, that financial forecasting.
1496.09 -> You use a product called Amazon Forecast
1498.57 -> and with that you do the forecasting all those time series
1501.44 -> and apply it internally within your organization.
1504.41 -> And it combines this forecasting with other information.
1506.65 -> Same thing like when I was mentioning personalization.
1508.7 -> Personalization can also combine
1510.21 -> with other information like the weather
1512.43 -> because the weather is extremely relevant
1514.58 -> when you're doing prediction for a personalization,
1517.7 -> same thing for forecasting.
1519.13 -> That kind of additional information
1520.45 -> can be added to the system
1521.75 -> and considered when analyzing
1523.26 -> and doing the forecasting of those metrics.
1527.03 -> One example more is the grocery chain in India
1529.99 -> which reduced 30% the waste of their production
1533.85 -> because of the right forecasting
1535.53 -> and planning on what is the stock that they need,
1537.57 -> what is gonna be the demand,
1538.89 -> and based on that, they can optimize
1540.72 -> all their supply chain, all their inventory,
1542.64 -> and reduce the waste by 30%.
1546.89 -> Another would be more on the manufacturing industry.
1549.04 -> That good example would be Foxconn
1551.673 -> in which they are using also for optimizing inventory,
1554.96 -> for optimizing fulfillments,
1557.44 -> and for doing that 8% increase
1561.06 -> in demand forecast accuracy
1563.41 -> by using Amazon Forecast and our forecasting solution.
1570.31 -> Some additional examples in the forecasting,
1573.77 -> in different industries using forecast.
1577.2 -> Another one, we were mentioning that
1578.87 -> some were more for the customer experience.
1580.77 -> This is more about enhancing your employees,
1584.39 -> allowing them to make better decisions,
1586.83 -> and find information easier.
1589.3 -> And that will be on the left side
1590.81 -> you see a little bit the traditional search.
1592.2 -> In the traditional search what you'd use
1593.77 -> is you use some keywords
1595.2 -> and the system will provide you a list of documents
1597.79 -> and you'll have to click one by one those documents,
1599.74 -> and read if that document is relevant,
1601.19 -> and extract the insights.
1602.42 -> And intelligent search what it does is first,
1604.71 -> it understands natural language.
1606.45 -> When you're asking that question
1607.7 -> it will understand exactly what you're trying to search for
1610.8 -> and then it will provide you an answer.
1612.88 -> It will look at all the documents,
1614.36 -> understand those documents,
1615.46 -> and provide you with the best answer
1617.25 -> for that specific question.
1618.9 -> And also, additional questions you may have
1621.84 -> which are related to this one.
1623.68 -> So it's a completely different experience.
1624.613 -> It's not just finding documents,
1626.36 -> it's understanding those documents
1627.89 -> and providing you answers to your questions.
1630.01 -> It can be applied across all the industries
1633.15 -> and that will be, it'll be the experience.
1634.87 -> So, you first type your question using natural language.
1637.95 -> I want to find what is the IT help desk
1641.04 -> and then the system will provide you with specific answer,
1644.29 -> in that case is first floor.
1645.86 -> That's where you need to go for IT.
1647.91 -> And then it will propose you additional questions
1650.52 -> which are related to this one
1651.75 -> and that other people that were asking
1653.2 -> the same question were also asking.
1655.48 -> And then it will index and analyze
1657.78 -> all the information in your organization,
1659.26 -> it has different connectors
1660.093 -> and it's capable of pulling that information,
1662.65 -> indexing that information,
1663.98 -> understanding those documents,
1665.36 -> and providing you with that intelligent search.
1668 -> As I was mentioning,
1668.833 -> this was used across different industries.
1671.72 -> A good example will be 3M.
1673.666 -> 3M from a research perspective
1675.09 -> every time that they need to do a new research
1677.56 -> they need to pull from previous research they've done,
1680.28 -> documents and finding those documents
1682.02 -> across thousands and thousands.
1683.73 -> And also based on who's doing that question,
1686.54 -> what is the keyword they are using,
1688.24 -> By using this solution they are capable of
1690.25 -> understanding exactly what is the user looking for
1692.76 -> and suggesting what is the best document.
1694.65 -> That increase productivity,
1696.3 -> that provides a shorter time
1699.05 -> to market let's say from a research perspective.
1702.84 -> Another example will be Magellan.
1704.45 -> That's more on the, for the contact center.
1706.66 -> Remember that I was mentioning the contact center
1708.46 -> and helping the agents.
1709.83 -> Well, those agents are having
1710.95 -> a conversation with a customer.
1712.44 -> They get a question.
1713.71 -> That question is understood by the system
1715.83 -> and immediately send to this system
1717.68 -> so it's capable of providing you
1719.043 -> what is the best answer for the question
1721.15 -> that this customer is asking.
1722.76 -> So that's called mining tool those use case,
1725.06 -> one helping the agent
1726.53 -> and also providing them
1727.363 -> with the best answer for that question
1729.43 -> by using this intelligent search solution.
1733.35 -> Some examples of Amazon Kendra.
1736.02 -> Amazon Kendra is the product that we use
1737.88 -> for intelligent search across different industries,
1740.96 -> the Wall Street Journal and so on.
1746.46 -> Industrial.
1747.293 -> In the industrial space,
1748.38 -> the main use case is predictive maintenance
1752.43 -> but we have others.
1753.31 -> Automate visual inspection and quality
1755.81 -> capable of looking at the products
1757.43 -> are coming out from the manufacturing line
1759.8 -> and identifying any problem
1761.57 -> that exist in that product from a quality perspective.
1764.61 -> Or also security,
1765.92 -> capable of looking at all the manufacturing plant
1768.08 -> and identifying if there is
1769.57 -> maybe someone that is not wearing
1771.58 -> the right protective gear,
1773.46 -> or if there is a problem with the robot
1775.103 -> that is too close to a person.
1777.25 -> And then the predictive maintenance
1778.6 -> that I was mentioning before.
1780.2 -> That's basically taking,
1782.94 -> analyzing all different sensors
1784.7 -> available throughout the manufacturing line.
1787.01 -> Capturing that data
1788.61 -> and once you have two months,
1790.35 -> three months of history of that data
1791.82 -> and what are different problems
1792.98 -> that came up with those specific equipment,
1795.46 -> the system is gonna be able to predict
1797.29 -> when a specific equipment will fail.
1800.03 -> And that's something that will prevent,
1801.35 -> you can change that equipment or fix that equipment
1804.76 -> one week before it fails
1806.29 -> which means there's not gonna be downtime,
1807.9 -> you can do that in maintenance whenever you want
1810.25 -> and it won't affect the production
1811.72 -> or it won't affect the quality
1813.17 -> of the product that is produced
1814.56 -> because of that defective manufacturing process.
1818.7 -> You can also do forecasting,
1820.08 -> forecasting is also used in the manufacturing space.
1823.16 -> Related to predictive maintenance,
1824.89 -> one of the products that we have is called Monitron.
1827.93 -> Monitron are those yellow devices.
1830.06 -> Those are things that you attach to your equipment
1832.67 -> and then they connect to the hub,
1834.24 -> the big, yellow device on the right.
1836.74 -> And they send and transmit the information
1838.56 -> about those sensors,
1839.45 -> and that's temperature, vibrations,
1841.19 -> and everything that they capture from that specific device.
1843.92 -> And this is connected with an application
1845.58 -> that is capable of then identifying alerts,
1848.33 -> identifying anomalies.
1849.59 -> Sending those anomalies to the operator
1852.03 -> with the phone application and letting them know
1854.4 -> that there is something going on
1856.07 -> with a specific equipment
1857.68 -> based on the learnings of
1860.01 -> all the information analyzed throughout the time,
1862.23 -> and any points on improving
1863.75 -> because those systems are constantly improving
1865.85 -> with additional information being brought by the system.
1871.79 -> Examples of companies using this,
1874.03 -> Koch will be a good example.
1875.61 -> They are using Monitron, Monitron is the device,
1878.95 -> and Lookout for Equipment is the service
1881.57 -> that is doing the analysis of that information.
1883.82 -> So if you already have sensors,
1885.15 -> you will use Lookout for Equipment.
1886.9 -> If you don't have sensors,
1888.09 -> you can use Monitron as sensors connected to your devices.
1892.39 -> And they're basically using this for
1894.54 -> optimizing their predictive maintenance
1896.46 -> across the different devices.
1899.27 -> ENGIE, that's a french company
1900.69 -> also in the manufacturing space.
1902.3 -> They are using SageMaker
1903.42 -> so instead of using Monitron or Lookout for Equipment,
1906.13 -> they build their own model using SageMaker.
1908.91 -> And they improve,
1911.21 -> they have more than 1,000 predictive models
1913.33 -> being built across all their manufacturing lines
1915.56 -> and all their equipments
1916.91 -> capable of doing predictive maintenance
1918.67 -> across all the different devices.
1924.15 -> So how we make it real?
1925.42 -> How we help our customers deploy those AI/ML solutions?
1930.29 -> Well, the first option is just where with Amazon,
1932.87 -> Amazon has professional service
1934.58 -> and we can help our customers
1936.6 -> by developing the process and helping them,
1938.59 -> and working with them.
1940.08 -> We also have what we call the Amazon ML Solutions Lab.
1943.02 -> This is a group within Amazon
1944.86 -> that helps customer ideate new ideas
1947.27 -> and come up with a proof of concept
1949.66 -> about how we could apply AI/ML in that specific company,
1953.31 -> and what will be that next use case.
1955.67 -> Could be one of the ones I presented,
1957.09 -> could be a completely new one
1958.72 -> and that's kind of an ideation process
1960.34 -> that works with solutions lab,
1962.06 -> working close with the customer,
1963.73 -> and then deploying a proof of concept
1965.86 -> that will be deploying in the organization.
1968.46 -> Second option is to work with our partners
1970.06 -> who have extensive network of partners
1973.58 -> that are certified with machine learning
1975.6 -> and are using our machine learning technologies,
1977.64 -> and they're on development on top of them.
1979.48 -> That can help customers
1980.93 -> for very specific solutions and industries
1983.24 -> and we have a broad catalog of third party solutions
1986.39 -> for AI/ML.
1988.12 -> And then finally, you can also do it yourself.
1990.04 -> You have like two options as I was mentioning before.
1992.51 -> The first one is by using artificial AI services.
1996.03 -> Those are the pre-trained models
1997.37 -> that I mentioned before.
1998.61 -> The ones for voice, for vision, for text.
2002.39 -> Option two would be to build your own model
2004.64 -> that will be using SageMaker
2006.53 -> and that's a tool that allows you
2007.6 -> to train your own models.
2009.2 -> And then to help our customers
2010.99 -> in the adoption of those specific use case,
2013.24 -> we've created a set of solutions library
2015.52 -> that use those AI service and machine learning
2017.78 -> that our customers can download for free.
2019.89 -> And you have all the different service,
2021.5 -> how to stitch together those service,
2023.38 -> and something that you can use as a starting point
2025.89 -> for your own deployment.
2027.21 -> Same thing for SageMaker,
2028.57 -> we have what we call Amazon SageMaker JumpStart
2031.82 -> which is something that helps
2033.45 -> whoever is working with SageMaker with specific models.
2036.33 -> And so, they don't have to start from scratch,
2038.22 -> they all had some templates they can use
2040.08 -> to develop their own machine learning models.
2044.12 -> These are some examples of those solutions
2046.72 -> and below you have the URL
2047.857 -> if you are interested in checking for additional ones.
2050.48 -> And you have for most of the ones I have mentioned,
2052.96 -> for intelligent document processing,
2054.76 -> for chatbots,
2056.02 -> for contact center intelligence.
2057.82 -> For all of them you have solutions,
2059.14 -> you can download those solutions
2060.73 -> and they will show you exactly
2062.17 -> how to stitch together different service
2063.96 -> if you decide that you want to build it yourself.
2066.22 -> As I was mentioning, this is one option,
2067.81 -> option two, working with partners or working with Amazon.
2072.48 -> That's an overview of the main AI use case
2075.58 -> that I wanted to present today.
2076.95 -> It took me less than I was expecting.
2079.3 -> So, that's it.
2081.52 -> Basically I wanted to give you an overview of what is AI,
2084.75 -> what are the main AI tools
2086.81 -> that we use internally at Amazon
2088.17 -> and the ones that we're offering
2089.24 -> in those multiple layers that I was mentioning.
2091.67 -> What are the main use case
2092.94 -> across the different industries?
2094.49 -> What are some examples of some of the customers
2096.78 -> and the value that they bring?
2098.35 -> If you're interested in any of those use case,
2100.62 -> please check our website.
2102.23 -> You'll see some additional examples and customers,
2104.26 -> you'll see the solutions that I was mentioning.
2106.38 -> You see how to engage with us.
2108.1 -> And there's additional sessions here
2109.95 -> that you can go and attend for the specific use case
2112.7 -> like IDP or like predictive maintenance.
2115.28 -> I know that there are many different sessions
2116.82 -> that go deeper from the technical perspective
2119.19 -> of how these things works
2120.68 -> and how to deploy it.
2122.67 -> Okay.
2123.503 -> Thank you very much for listening.
2124.6 -> Thank you.
2125.754 -> (uptempo music)
Source: https://www.youtube.com/watch?v=KnTya5qpWUs