Automation using Razor Sprint
Many businesses start out with manual processes. Some of these processes are complex and require human interaction to understand and complete.
As a business scales up, their processes scale up too. Everything is amplified and this includes processes that are inefficient and time-consuming; the good gets great and profitable, while the bad gets worse and becomes a larger drain.
A common example of this problem is managing incoming communication, specifically, purchase orders. In the beginning, this was fairly easy to cope with. An order would come in, a person would read the order, understand it, make a few decisions and process it. The task was fairly time consuming but far too complex to automate.
Over time, the number of orders rises (and this is great, the business is growing!). Each customer sends their purchase orders in different formats. Maybe they refer to a product or quantity in a slightly different way. It doesn’t matter, as humans, we can easily work out what they mean.
At scale, this process is becoming costly and slow. It is now an area where a marginal gain would make a big improvement to the bottom line. But historically, this would not be an easy process to automate.
It still isn’t easy but it is achievable with data. Data doesn’t just have to be something we use to create beautiful dashboards and predictions for selection or decision making. There are many other forms of data.
Purchase orders are essentially images, and using computer vision (CV) we can dissect an image into known areas and then extract meaningful data from them. Then with additional processes and natural language understanding (NLU), this meaningful data can be turned into actionable information.
This is exactly what we did. Training a model using historical purchase orders, the system can understand, process and take action and place orders within the backend systems. This unlocks people's time to do more valuable tasks.
One of the great side effects from the initial exploration was around an element that humans find incredibly difficult but computers are brilliant at - long, tedious and repetitive tasks.
An example of this? Compliance. It’s important and usually wrapped in a lot of words. Each purchase order had plenty of terms and conditions associated with them. Do you think that people actually read all of those terms every time? And if they did, do you think they would have spotted a change? Probably not, right?
This automation solution does. It will highlight any changes in the terms and flag the order for review. What it won’t do is know what action to take and this is left for the person to review, understand, accept or take action. Let the computers do what they are good at and unlock people to add real human value.
Something that can be taken from examples like this is that it can be hard to know when to automate complex tasks. When is too early? When is it too late? Overall, data suggests that being too early is far less expensive than being too late.