Hands-On Automation of Day to Day Workflows
In the post-COVID era, CXOs around the world are thinking earnestly about comprehensive transformation strategies. However, if you have a half-decent engineering team, you don’t have to wait for them to get their act together. Just follow this generic approach that we use at Coseer for intelligent process automation at our clients.
Let’s say you already have a workflow in mind that you want to automate. Ask these questions:
1. Why do you want to automate this workflow?
The answers may be varied. The best reasons in my experience are about user convenience (internal or external) or to save time. The worst, to save costs. Automation takes some effort, and if the motivation is not inspiring enough, perhaps you should not start on it at all. This idea cuts across many problems in the field of digitalization. The first stage “Understand What You Really Need” in this article for chatbots published in The Startup is very pertinent.
2. What is the upstream journey of data in the workflow?
Always start from the output. If you have answered question 1 correctly, then you have a good idea by now. Start working backwards. What is absolutely necessary to generate this output? The answers would typically be in the form of a) data, b) processes, and c) judgment that is essentially human. The C part is usually a big pitfall and appears counter-intuitive. However as you map workflows you will realize that somethings only a human can do.
3. How can the data sources be integrated using APIs?
This is the dumbest and one of the most important parts of the automation journey. Sound data management is critical to every digital automation project, and is often woefully lagging. See the discussion under the fifth core value of digital transformation as presented in Chatbots Life. In short, if you cannot reliably and repeatably get the necessary data, it would be impossible to go about this automation.
4. How can the data processing be automated?
Processing the data automatically to get the right derivatives can be very easy or very hard. Take as simple an approach as can work. AI is very tempting, but if you do use AI, be wary of its costs. Some of them are mentioned in this KDNuggets article. Broadly, it takes a lot to get it to work, it needs a lot of data to train, and it can be unreliable unless controlled appropriately. Dark Reading has an interesting discussion about the last one. Even within AI, simpler algos applied correctly are far more effective than complex, little understood algos.
Another aspect here is to not reinvent the wheel. If minor modifications or minor compromises let you use an open source package, or some module that you programmed earlier, go for it. Done is far better than perfectly begun.
5. Can the workflow be redesigned to be more automatable?
For the first workflow you do this analysis on, you are bound to face hiccups. There may be things that must have humans. There may be data sources that are simply intractable. Or there may be processing steps that cannot be automated with resources that you have available. The answer is not to give up, but to see if you can redesign the workflow around these hiccups. Perhaps break it, or use alternative sources, or stage it to fix the easier parts now and worry about the rest later. Even partially automated workflows go a long way.
Only after you have thought through each of these five questions would you be ready to start on your automation project. By now, you must know of the best tools, best packages and the best approach to do it. The only other thing you should think about is to do everything in a modular, productized and repetitive way so that the next workflow you take up is easier, and the next even more.
