Using Einstein Analytics To Empower WSP Expense Management
Overview
After the integration of Optical Character Recognition (OCR), Workforce Success Platform (WSP) added a new feature to create an expense record just by scanning the receipts.
A user can choose to take a photo of the receipts through their smartphone or upload them through PC. This new feature greatly enhances the user experience. It eases their life of manual input of receipt amount and other required fields, as well as the life of approver to approve it without cross-checking the fields.
Once the implementation of OCR was released, we realised it’s just the start of making our system easier and smarter than our competitors.
My team then started looking for the problems customers face in daily life and decided to begin some research on data analytics.
Workforce Success Platform (WSP)
While going through the overview, you might be wondering what is WSP.
Let me explain it briefly, as you already have seen, WSP stands for Workforce Success Platform. Its main aim is to improve employees' overall productivity.
In short, it is a combination of Time Tracking, Attendance, Expense and Planner into one system to reduce all indirect hours an employee wastes by switching in between different systems.
More follow-ups for WSP will be added soon.
Using Salesforce Einstein AI in daily life
You can get the basic information about Salesforce research on Einstein through its official website (https://einstein.ai/).
The usage of Einstein can be classified into three segments:
- Einstein Vision: This uses image classification and object detection algorithms. It can classify images and helps in a variety of daily operations like visual search through the phone camera, visual filters of related products, classifying objects and characters from images, etc.
- Einstein Language: Commonly knows as sentiment analysis, it uses a natural language processing(NLP) algorithm to predict the intent from a stream of unstructured data. This segment is used quite frequently in chatbots which are trained to answer as per the questions asked.
- Einstein Voice: This is the latest in the series and can be thought of in a way you interact with Google, Alexa or Siri. It uses NLP for voice recognition and can do a real-time response. In my opinion, this can be trained to use as a smart screen reader too for the employees who are visually challenged or have a tough time going through the whole application and a lot of tabs.
To get more knowledge for the above, you can read on their developers' platform https://metamind.readme.io/
Integration with WSP Expense Subsystem
Have you ever submitted any expense to your Finance Department or HR before?
How hard was it?
You may say it was easy in case you have claimed it on paper.
Because in that case, your HR is the person who has updated the remaining required information for you. These additional pieces of information can be Expense Types, Report Types, Accounting Period, Cost Center, Job or some additional information which your finance department needs in order to process your request.
I will not talk more about such manual paper submission here as I highly recommend you to go digital. (And I am not saying this just because I am some environment-friendly developer!)
Recently I found out that a company having just 500 employees can have up to 5000 expense types. This can be used as a rough estimation of how complicated data can be.
Let's increase the complexity of the problem little bit more.
In case these 500 employees are working at multiple (say ten) locations, then those 5000 expense types can be ten copies of 500 expense types with different codes. Sometimes these expense types are grouped under certain categories (Expense Type Groups) and their parent hierarchy can go up to ten parents too.
In the event that you are the only person claiming for all 500 employees, the job is easier for you (all you need to do is to memorise those 5000 expense types, can you?).
I will tell you one easier solution: That is, leave it on us, the developers.
Let me share with you how we fixed this issue in our last kaizen (improvement) development.
Here goes the integration part:
Our integration consists of three simple steps performed in a loop:
- Scan receipt to capture amount, date and receipt text.
- Create record from amount + date and expense type suggested from previously trained receipt data.
- Add receipt text and selected expense type to train the dataset to get a better prediction in the next record creation.
After this integration, we start getting Expense Type Suggestion after receipt selection:
In the above screenshots, I have selected a smartphone-camera captured Starbucks receipt, and the top expense type suggestions we got were:
- “Client Meeting”
- “Team Event”
- “Office Snacks”
Imagine out of 5000 expense types your company has, you are getting the expense types you need in front of you in the top 3 suggested expense types. It happened because of Einstein (and some developers who understand customers' problems).
We tested on receipts from restaurants, cafes, transports and convenient stores in English and Japanese. The top suggestion in English was accurate in all cases, while in Japanese it was giving an accuracy of around 75% which we consider is also quite good as this accuracy is going to increase once we have more receipt data to train.
The screenshots are in Japanese as I was asked to present this in Japan as a part of the Salesforce World Tour Tokyo ’19. In case you want to know more about this or want to check the English version of the product, please ask by adding a comment.
Future of WSP
With the above solution, and if scaling it to add suggestions for Job and Cost Center, we can help individuals to create their expense reports in seconds even when using the system for the first time.
Subsequently, our future plans involve a more robust approach for fraud detection using AI which I will write in my next blog.
At TeamSpirit, we are bound to provide the most innovative solution to the most common problems.
In the end, I want to ask my readers a question. If you are given enough resources, how you think you can proceed with, to utilise it in a way different from others. Do add comments for your answers.
Well, since this was my first article, thank you for going through it, and show your support by claps and subscribe.
Any suggestions are much appreciated.
