1 of 43

Bots in Robotic Process Automation

by Carlos Toxtli

2 of 43

Experience

3 of 43

Human-Computer Interaction

4 of 43

The research side of life ...

5 of 43

Let's start with Machine Learning

6 of 43

All that we need is data!

7 of 43

Feed the machine learning model with the data ...

8 of 43

Training the model ...

9 of 43

Now we have a model to predict new data!

10 of 43

But an AI model is a single piece of a solution

11 of 43

Robotic Process Automation is the missing link ...

12 of 43

Robotic Process Automation = Software Automation

13 of 43

Bots (agents) are the building blocks of RPA

Bots simulate & understand human interactions such as:

Vision
Audio
Text
Interfaces interaction
Location

14 of 43

Tools are reactive, Bots are proactive & reactive

You use an app when you need it

A bot contacts you when it needs you.

15 of 43

RPA + AI = Intelligent Automation / Cognitive RPA

16 of 43

RPA for AI

Bots for

Data collection

Bots for monitoring the input channels

Bots for training the AI model

17 of 43

Data come to me ...

But getting the right amount of data in the format that we need is not always easy ...

18 of 43

Web crawlers / Spiders / Scrapers

19 of 43

Different sources

20 of 43

Scraping platforms

21 of 43

Data cleaning ...

22 of 43

Then we can train a model in an old fashion way

23 of 43

Or with RPA automate the machine learning process ...

24 of 43

AutoML platforms

25 of 43

Then we need to have an interface for the model

Bots are implemented in channels or applications that supports bidirectional communication such as Notifications, Webhooks, or Endpoints.

26 of 43

Bots on messaging channels

27 of 43

Messaging bot platforms

28 of 43

RPA Frameworks

29 of 43

Examples of RPA that implements Bots & AI

30 of 43

FineBot - RPA (chatbot,crawler) only

Parking

Speed limit

...

8:15pm 🤖:

Hey David there is a fine charged to your car. The concept was illegal parked.

8:15pm 🤖:

David I’m sorry to let you know that your son got a ticket.

The concept was illegal parked

🤖

8:15pm David:

My plate is: ABC123

31 of 43

BobbaBot - RPA (chatbot) + NLP

8:00am Mark:

Hey bob about, today I gonna get a taro tea

8:00am 🤖:

Got it!

8:15pm Lupe:

Bob please bring me a Matcha

8:15am 🤖:

Got it!

8:45pm Alice:

Today I want a Honey Dew tea

8:00am 🤖:

Got it!

9:00am 🤖:

Hey Mike this is the Ford’s order for today: Honey dew, Matcha, and Taro

9:15am 🤖:

Hey group, your beverages are at the reception.

8:00am

9:00am

32 of 43

ApprovalBot - RPA (email) + AI

The bot gets a document approval request for 3 people

The bot sends the document approval request and tracks its completion

Every time that a document is accepted or rejected the bot learns the criteria

The bot learns the times of each actor and their criteria to give suggestions in the future

Suggestion

33 of 43

TaskBot - RPA (voice) + NLP

From: luis@...

To: carlos@..: john@...

Cc: taskbot@...

Hey taskbot,

Please remind Carlos yo send the proposal by Friday and Luis to send the signed contract.

Task management tool

John

Carlos

Luis

🤖

34 of 43

MATT - RPA (crawler) + AI

Student sends an assignment

A bot recruits experts on Linkedin

The assignment is sent to the most relevant and available expert

The experts evaluate the assignments and provide feedback

35 of 43

Failure Predictor Bot - RPA (sensors) + AI

The bot monitors the sensors (physical or visual) and failure alerts (even audio)

Every time that a failure happens, the bot learns from the previous sensors information

The bot is able to prevent failures when early symptoms are raising.

🤖

The bot sends a preventive message to operators when there is a potential failure.

36 of 43

Security - Liveness test (RPA + Machine Vision)

The system requires to ensure that it’s you and activate the verification bot

The bot starts a video call or request the use of the webcam

The bot process the video on real time and verifies that the person matches with an ID and is not a photo or a phone screen

The bot retrieves the decision to the system.

37 of 43

ImporterBot - RPA + ML - All autonomous

The bot searches for trending products on Chinese marketplaces

The bot stores the found products and assigns a competitive price for the local marketplace

The bot publishes the products in the marketplace (MercadoLibre)

The bot waits for purchased products

The bot confirms the payment

The bot purchase the product in the Chinese store and set the customer shipping address

The bot track the package and update customer of any delay

The bot gives feedback to the customer and reminds the customer to give feedback

38 of 43

And many many more ...

39 of 43

Lessons learned

Automated recruitment systems are effective over work channels (Linkedin)
Automation helps to scaling solutions.
Social friction is reduced in mediated communication.
Even that all the selling workflow can be automated. The vendors and the delivery services should be properly selected in advance.
Mitigation strategies must be always set on any AI solution.

40 of 43

How to go from the idea to the implementation?

Structure your idea as an autonomous system that can act whenever it is needed without necessarily having human intervention.
Diagram a workflow of your solution. BPM diagrams are commonly used.
Identify what processes requires simulated interaction and processes can predict values from data.
Implement the elements by using RPA frameworks and AutoML.
Test and deploy using continuous integration.

41 of 43

Tutorials for getting started

Collection of curated RPA content

http://bit.ly/awesome-rpa

Collection of curated AI content

http://bit.ly/awesome-ai

Main AI technologies explained

http://bit.ly/siselai

UiPath video tutorials

https://www.uipath.com/developers/video-tutorials

42 of 43

Conclusions

AI often relies on RPA to be implemented
RPA solutions are more useful when these are not only rule based
Coordinating human teams and orchestrated bots lead to boost productivity

43 of 43

Thanks

@ctoxtli

carlostoxtli.com