Introduction to DocAI and Contact Center AI

Key Takeaways

foreign [Music] engineers at Google Cloud welcome back to the technical guide for startups the grow Series where we create a series of videos for technical enablement we are thrilled to help you start build and grow your businesses successfully and sustainably on Google Cloud in our last episode sanskriti and Elise walked us through how you could architect a blockchain startup on Google Cloud today in the seventh video of the grow series we pie and I we take you through how you can start innovating with ML Solutions on Google file we will be covering the various ml Solutions available on Google Cloud what is ccai what you can bring for your businesses and how it helps other customers how Doc AI can help with your document processing needs what use cases it can solve and how other customers are using it finally a list of resources to get you started on innovating Google cloud has an umbrella of machine learning products and solutions that your businesses can leverage we have done a deep type of a few of these in our previous videos so let's do a quick recap of what we've covered so far we can broadly classify ml Solutions under three categories on Google Cloud use as is customize and build under build we have our text AI to help solve business problems using data and increase productivity with aim to end mL of services check out our video getting started with vertex AI to know more under customized we have automl which help businesses with limited ml expertise to start building their own high quality custom models under use as is we have two vertical Solutions Talent solution and recommendation AI Talent solution leverages ml technology to help better Understand Job content and job Seeker intent use as is also encompasses ml apis which we have covered in the video how to get started with ML apis we also have two horizontal Solutions doc Ai and ccai which we will be covering in detail in this video conversation is a fundamental building block of customer interactions and contact centers we want you to fully enable and Revelation the very same technology that we have integrated and built for Google at the core of our conversational AI is technology that helps enable a conversation with a machine that enables natural interactions with virtual agents understand what customers are saying talk to the customer in a natural Manner and interact with a customer in a conversation we have stt or speech to text TTS or text to speech and dialogflow as a building blocks to the conversational AI for text and voice nccai is built on top of this building blocks the virtual agent powered by the low flow with speech AI provides both a text and voice channel for call containment for common core topics for more complex topics that require Human Service agent we have agent assist the listen into ongoing conversation and promise assistance to the Human Service agents through knowledge base lookup and recommended resolution steps this will challenge an average handling time and provides more consistent call Quality across call center agents and finally we have insights where all the conversations are stored in a single place you can do a lot of analysis and modeling and you can also find a graph and chart to see what is the top reason that the customer contact you and generally we were working with the telephony partner like a wire and Genesis to help our customers to utilize this technology and drive this into their own business in 2023 we are offering a new product the ccai platform to give our customers more options when deploying ccai customer can show ccai platform for contact center as a service in our cloud or continue with the existing investment in their call Center Software like genesis Avaya lines and Cisco and so on today we don't want Mill ccai platform we give the choices for our customers to choose to use ccai either from Google file or from our partner ccai platform gives us integration and really brings this unified platform it can consolidate the Della flow agent assist and insights together to cover many different needs that the customers might have is really a turnkey solution for ccni to make deploying and using this conversational AI easier and quicker contact center AI provides our customers the orchestration capability to fully leverage Google cloud and our ecosystem from workspace Chrome all the way down to the building blocks of AI and infrastructure now let's take deep dive into each ccai building block well dialogflow helped with your agent designers and developers build faster by providing many pre-built agent templates and components and it doesn't require deprogramming or machine learning skills it can also improve customer experience with frequently asked questions info box searing and transactions you can also engage more efficiently with Advanced natural language processing and maximize reach by languages and platform Integrations supported by dialogflow secondly agencies Focus entirely on giving agents the right data at the right time ensuring that both agent and the customer have better experience whether it's contacts us they begin the conversation or helpful directions with frequent artificial and knowledge based answers to summarizing the interaction when it's done easing their wrap-up process and that's where the inside comes in sights can turn your customer interactions into insights through our built-in conversational AI understanding and topic modeling capabilities that are extendable by your call center analysis the inside scan can help make a Better Business decisions those business decisions are going to help Drive operational efficiency across a variety of your company ccai insights integrates with all other contacts under AI solution products allowing you to import conversations from dialogflow and agent assist contact center AI Insight helps user detect and visualize pattern in the contact center data understanding conversational data helps Thrive business value improve operational efficiency and provide a voice for customer feedback you can import your contact center interaction data into insights in order to run machine learning analytics or to automatically identify interesting interactions and reveal the conversation all data include transcribe highlights topic sentiment and entities gets exported into bigquery data set and insight table ccai insights allows you to export your ctai insights conversation and Analysis data to bigquery so that you can perform your own route query allow you to build any custom analysis or visualization in any tools like looker Luca studio and so on we are happy to share our customer success story from Taylor's from your data in 2021 there's always 20 million customer conversations per year imagine if we could transcribe every customer call into a customer intent that we could take action on the ccai will give them the transcribed data asset of customer intents for every customer call this asset will be used to help Drive multiple commercial outcomes including the digital personalization use case and contact center analytic use cases the result is so impressive tailors can reduce the customer effort for 7 and 200 hours per year and reduce agent effort for 2.3 million US dollars per year now let's see a quick demo on ccai insights in this demo we are trying to improve the customer experience of the electronic shop that sell devices and tools and behind the scene this is running on the advanced natural language model on Google Cloud first thing we will do is to import a child transcript into Insight here we have 10 000 generated conversations which store in our bucket and you can either import by Json format or audio and the insights will handle the transcription for you I will use insight to identify the most common topics in our conversation data it uses topic modeling to group the conversation into relevant cluster we can click on the start training and you can take up the few hours we will come back after it completed to take a look on the result this is how the result look like we got seven different topics discover all the data and the best part is Insight did everything but as superwise learning we can deploy this model so that insights can use this to analyze the conversation now if I'm if I'm there for agent I noted my customer are likely to have these seven kind of problems so you can build the dialog flow or the virtual agent to help them solve the issues in addition to high level topic we can analyze the conversation in more details using the topic model we can configure the fraction of conversation that should be analyzed automatically here we can go through one of them in details in this conversation it seems like a customer has a problem with a streaming device it wasn't turning on and they're trying to use troubleshooting wizard this conversation is 6 minutes long and had 14 turns so now I can identify which conversation I can explore in more detail and improve the virtual agents another technique that we can use to learn our conversation data is conversation highlight smart highlights are built in the insights and can detect the important attributes in the conversation this also has some highlights where you can track your own tag like laptop phone or Internet it looks like most of our customers have pop and with the internet issues so we can add some intent or resources to our virtual agents in dialogflow finally now you understand how the inside fits in as a part of the contact center AI on Google Cloud thanks mile for showing us how businesses can leverage ccai for their call center needs with the recent shift to remote and hybrid work and an increasingly digital world we have adopted a paperless route and a lot of documents have now become digitized think of all the PDFs forms emails and contracts you have used to consume or relay information these documents essentially make up a gold mine of data that businesses can leverage to derive insights and drive business decisions enter document AI document AI is an end-to-end Cloud platform for document processing which turns unstructured content into structured data with data in a structured format you can begin to make it useful maybe you want to run analytics on customer feedback or you're processing large multi-page application forms the possibilities are endless document AI helps with document processing tasks you can use it to digitize text from documents this means you can extract text words paragraphs blocks symbols and even correct rotation document AI also helps in pre-processing documents with image quality detection and this queuing labeling and reviewing can also be done with document AI you can split and classify documents based on type this would be helpful when for example you have scanned a bunch of different procurement documents into a single PDF or file and you need to programmatically switch the documents based on logical boundaries you can also use it to process industry-specific documents document AI workbench can be used when you need to create a custom model specific to your business now I know I threw around a few processor names so let's understand the different processors document AI has to offer first we have General processors that are pre-built processors that are compatible with most document types under General processors we have document OCR and the form passer second we have specialized processors that are pre-built processors for specific document types this includes procurement documents which are used for payments and purchases such as receipts and invoices identity documents that are used for identity verification lending documents that are used for mortgage loans and business contracts lastly we have custom user creative processors if you have a unique type of document and want to get the same detailed information from it this can be achieved with the help of document AI workbench document AI offers the ability to obtain existing specialized processors or create new processors completely from scratch in order to get the most accurate information from any document type these work very similarly to automl with vertex AI so you don't have to write any tensorflow or Pi torch code to start training the model however there is a certain criteria to be met while working with document AI workbench a minimum of 20 documents is required to train a custom document extractor or uptrain a specialized processor you will reserve a minimum of 10 documents for training and 10 for testing you need a minimum of 10 annotations per schema label task the customer or a partner needs to be resourced to label the documents with the many processor types available it can get confusing to select which processor to use in a given situation this decision tree gives us a quick overview of which processor to use let's go through it in a bit more detail if your requirements are to only extract the text and the layout then the OCR processor is a way to go it lets you identify and extract text including handwritten text the form passer is more useful in cases when you want to extract General key value pairs for example entity and check boxes tables and other generic entities which are present in documents apart from OCR text if you need to process documents with a specific layout and a specialized processor is already available you can then leverage these pre-trained processors or uptrain them by up training we mean that you begin with a pre-made model and then train this model with your own data to improve its accuracy if our available specialized processors do not cover a document type that you would like to process then you can use document AI workbench to create a custom processor provided that it meets the criteria outlined in the previous section in case a criteria can't be met you can offer OCR processor and automl with the help of automl natural language you can create and deploy custom machine learning models that can analyze documents categorize them and identify entities let's go through a sample dock AI architecture you can either use the batch or the online API calls to process documents using document AI with batch or asynchronous calls these use documents which are stored in Google cloud storage and can write the results back to it online or synchronous calls allow you to send a document within a request and receive the results synchronously you can then use a serverless processing pattern via Cloud functions Cloud functions are good for actions that you might want to keep separate from your main app code and deploy Monitor and maintain separately just like Doc AI if it's a separate integration point to your system you could wrap your doc AI API calls in a cloud function and then attach it via an HTTP Trigger or you could create a background Cloud function that gets triggered by events that happen in your cloud storage bucket let's move on to the demo so we can get a glimpse of the functionality doc AI has to offer if you head over to the document AI documentation on the left pane you will see your try it option which essentially provides a drag and drop feature to let you try out the different processors as you can see you have the option to test the various processors either with a document on your local machine or with one of the readily available samples for the purpose of this demo I'll be using a sample form to test the form passer if I zoom into the form you will see a health intake form with various Fields such as name address phone number medical concerns Etc on the left pane you can see the list of key value Pairs and you can even filter the results based on the field if I type in state I can access the value associated with the field on the top pane if we select the Json tab we can view the request URL the request and the response in Json format in case you need to integrate doc AI into your application and make a call to the API the Json response may be easier to work with to extract the identified values to create a specialized processor on the console head over to the document AI platform overview click create processor and scroll down to specialized and select invoice parser give it a name and select the closest region on the list click create to create your processor foreign document and select an invoice to pass as we can see the various Fields along with their values have been extracted in a key value pair format we can also classify and split documents in this example we'll be using the procurement processor let's create a python file that will use the online processing API to classify and detect logical split points for a multi-page document I will create a file called classification.py and replace processor underscore ID with the ID of the procurement splitter processor I created earlier replace project underscore ID and location with the cloud project ID and the processor location respectively after running the file the procurement splitter classifier correctly identified the document Types on pages 0 to 1 and 3 to 4. page 2 contains a generic medical intake form so the classifier correctly identified it as other the code for the python file is available in the code lab Linked In the description after witnessing the capabilities of document Ai and how we can transform document processing many customers have started leveraging it one such customer is Senex senix is an Australian natural gas producer using document AI Senex was able to unlock data that could be used to optimize well design and operations as a result Cenex can also look for Trends such as safety issues across their assets before determining if the issue needs to be addressed as stated by Timothy Corcoran the general manager for digital Senex Google Cloud demonstrated impressive built-in OCR capability it took minimal efforts by our team to pull the Trap data from handwritten notes and PDF documents and turn it into useful information that could be used to run our business if you would like to know more about CCA and Doc AI there are a ton of resources to help you get started here are field links you can go through in the description you'll be able to read up more on ccai and Dot AI check out our YouTube playlist experience our Hands-On tutorials and lastly please stay connected to learn more our next video if you are interested in using document AI you can start with the document AI documentation which has sample code examples and an extensive list of the available parcels we also have a document AI playlist called the future of documents which includes a deep dive into each of the document AI processors and demos illustrating how you can use them finally you can get Hands-On with the cloud skills boost Labs or the code Labs that provide step-by-step tutorials that brings us to the end of this video in our next video we will go over learning the fundamentals of ml arts in detail and how you can achieve it on Google Cloud we will then go through a demo and inspiring customer story and that's a wrap as always don't forget to like And subscribe to our YouTube channel click on the Bell icon to get notified each time a new video is posted thank you so much and stay tuned foreign [Music]