Read on Twitter
If your are a small business owner who has people double-checking, QCing, or verifying things a lot in your business, I want to share how you can automate that (and how it works).
Maybe:
Invoices : packing slips
payments : invoices
schedules : billed
parts : orders
orders : plan
Maybe:
Invoices : packing slips
payments : invoices
schedules : billed
parts : orders
orders : plan
2/ First, let's understand what is being done.
A person has existing data or knowledge that allows them to compare incoming information to that existing reference.
So you get an invoice, you compare to existing packing slip to make sure you are billed right. This is to data.
A person has existing data or knowledge that allows them to compare incoming information to that existing reference.
So you get an invoice, you compare to existing packing slip to make sure you are billed right. This is to data.
3/ You get a list of items to quote, and use your industry knowledge to price them. This is comparing incoming information to knowldge.
Now lets take a break here, and discuss something else.
Deep Learning
When you have a history of items that came in...
Now lets take a break here, and discuss something else.
Deep Learning
When you have a history of items that came in...
4/ ... what they were checked against, and what decision was made, you have an extremely valuable resource. A history of human judgements, adjustments, or QC checks.
This is known as a "training set" in the AI world.
This is known as a "training set" in the AI world.
5/ It would be like tracking all historical packing lines, invoice lines, the decision made in a giant spreadsheet.
The decision you made is called a "label". You "labeled" that particular circumstance with what you wanted.
Approve
Doesnt Match
Reject
Adjust numbers
etc
The decision you made is called a "label". You "labeled" that particular circumstance with what you wanted.
Approve
Doesnt Match
Reject
Adjust numbers
etc
6/ With that history, you can "train" a very simple program to see what decisions you made, and then keep making them in the future.
"oh, everything this supplier sends this type of bill, we put it against Supplies GL account, not Inventory"
It will now make similar decisions.
"oh, everything this supplier sends this type of bill, we put it against Supplies GL account, not Inventory"
It will now make similar decisions.
7/ Back to your person spending time every day checking, verifying, etc.
They are "labeling" in real time. So rather than keep a history, you can have a program watch what they do as they do it, and get better and better at helping.
It will soon catch all the same errors.
They are "labeling" in real time. So rather than keep a history, you can have a program watch what they do as they do it, and get better and better at helping.
It will soon catch all the same errors.
8/ This is basic models, available for free to those that know how to use/build them.
The idea of tracking real-time labeling, using it in deep learning to build assistants that help with auditing your work in realtime is a massive opportunity for small business!
The idea of tracking real-time labeling, using it in deep learning to build assistants that help with auditing your work in realtime is a massive opportunity for small business!
This lowers the risk of “knowledge walking out the door” and makes it easier to scale experience based positions.
You can augment or suggest decisions for less experienced personnel based on the actions of your best people!
You can augment or suggest decisions for less experienced personnel based on the actions of your best people!
