4 steps to find processes AI can automate

Adrian Goergen

Automation is everywhere – and it matters: Slow and labor-intensive processes are a heavy burden for companies that face global competition. Artificial Intelligence is a proven method to streamline business processes at modern companies.

However, finding the first use-case is never easy. Since AI has such a vast range of possible business applications, it can become overwhelming to filter them. To help you to find the ideal project to kickstart AI process optimization within your company we created this 4-Step Guide.

We have also created a Google Sheet that helps you list and evaluates the processes happening inside your company.

But first:

Step 1: Know what you need to know

To get started, you and your team need to get a good understanding of what AI can do for you. Equally important, you also need to know its requirements.

Therefore, make sure you understand:

  1. How AI algorithms learn
  2. What data is needed
  3. What problems can be addressed (eg. using an image classifier)

Artificial intelligence can help you with a lot of things - but it is not a magic fix-all, catch-all solution. In practice, you should think about involving AI in the process whenever you have a task which answers to these four criteria:

1. Is it repetitive?

You should use AI to get rid of tedious daily tasks and not to make difficult, one-time decisions. Intelligently replying to a common refund request is a task a trained AI can handle. Reaching out to a customer with a unique, creative email is not.

2. Are the rules of the task fixed - or NOT fixed?

If you can describe the task by logical rules that a machine can understand, you don't need an Artificial Intelligence for it. Go for *robotic process automation* (RPA) or integration Platform as a Service (iPaaS) instead.

For example, a logical rule that a machine can understand could be:

If the email contains "job application" (Trigger) → forward to HR (Task).

The words "job application" are a clear trigger to forward an email to the HR department.

However, all applications that don't contain these exact words, will not be forwarded. This is where RPA ends and intelligent automation begins. An AI could look for patterns to identify a job application and make a decision based on judgment. RPA and iPaaS only work with clearly defined triggers.

3. Can a skilled human decide on it in a few seconds?

AI excels at automating decisions with clear input and related output. If a trained employee of the company needs to think hard about each decision, don't try to automate it - at least for now.

For example, a hiring decision takes into account qualitative and quantitative criteria as well as your gut feeling - this is the core of why it is not attractive for intelligent automation. On the other hand, filtering a database for relevant CVs might only take a human seconds per document and could likely be done by a machine.

4. Is there is enough underlying data for a machine to learn (or can this data be easily collected)?

A machine learning model is not a magic super-brain that can find answers to all of your questions by itself from the get-go. Even if a task matches the criteria, the model still needs to be trained on a large set of data to work properly. You need to give the algorithm examples of possible input (think images of dogs) and output it should predict (think dog breeds).

The amount of labeled data required varies by the task you are trying to solve and the desired accuracy you require. A (very general!) rule of thumb is that you should have at least 100 pictures per class (eg. dog breeds) you want to predict - but of course, the more, the better.

Step 2: List all possible use cases

You and your team are now aware of which applications you are searching for. Next, you can start identifying possible use cases. To find the perfect starter use case, you need to first get an overview of all the repetitive processes that happen within your company. Don't worry about the other criteria mentioned in step 1, we will address them later.

Get your people together from different departments and start brainstorming. Be creative. Also, think about "shortcuts" that do not currently exist but could be enabled by AI and break processes down to tasks.

At this stage, you shouldn't worry much about the practicability of solutions but rather imagine a perfect-world scenario. Imagine you have all the tools and resources you need and employees will easily adapt to changes.

List all repetitive processes that spring to your mind irrespective of their complexity. You can do so in the "Process Overview" sheet.

To get started, think about:

Operations

  • How does information travel within your company?
  • How is information stored (eg. Email attachments)?

Research

  • How is new information generated?
  • How is this new info filtered and analyzed to generate insights?

Customer engagement & service

  • Which communication is repetitive?
  • How do you handle customer input & learn from customer interaction?

Supply & Logistics

  • How do you source materials & pick suppliers?
  • How do you track the flow of goods and avoid mistakes?

Quality Assurance on your Platform/Product

  • How do you check for product/service defects
  • How do you ensure online content respects your guidelines
  • How could you improve your offering? (make it faster, more accurate...)

Finance & Accounting

  • How are reports created and updated for invoices, expense reports, accounts payable and receivables and other documents
  • How do you ensure compliance and detect fraud or suspicious activities

Once you listed the processes, you can proceed to think about how you want to improve them. Think about where AI could speed up decisions, for example by giving suggestions, and where it might be able to automate or cut out entire processes or tasks. Be creative about potential workaround solutions that AI automation could unlock.

As mentioned, it might be helpful to break down the initial set of processes into smaller pieces. While doing so, focus especially on the links between the subprocesses where (small) decisions have to be made. Make sure you detect each decision where manual labor has to be carried out. Forwarding an email is labor, so is updating a document or renaming a file.

Step 3: Evaluate automation potential

After you brainstormed a set of possible use cases, it is time to rank them. You can do this in the "Potential Calculator" of the google sheet. The goal is to map use cases according to their feasibility and impact, so you can easily see where to start.

First of all, you should determine the impact of the chosen (sub-)processes for your company. For us, the overall impact is a combined measure of direct financial impact (through time savings) and criticality (alignment to company strategy, improvement of product/service...).

What is the direct financial impact? Quantitative impact

Our approach to calculating the direct financial impact is estimating how many times a task is carried out per day and multiplying this by the time it usually requires. If we multiply this value by the number of people that conduct the task, we get total daily costs which can then be translated to annual costs.

After we calculated all direct financial impacts, we compare them to each other and rank them on a 1-5 scale.

What is the criticality? Qualitative impact

This is where your qualitative judgment comes into play. How much value does the process provide for your company? For reference, we use the following scale:

5 = could become a competitive edge
4 = will increase customer contentment
3 = follows an industry trend
2 = will slightly improve product/service quality
1 = changes will not be noticed

Second, you should determine the feasibility of implementing cognitive automation. Now is the time to rule out those sub-processes that:

  • Require deep thought or human touch/creativity (If possible, split the task and check if sub-processes don't require long thought. If that doesn't work, keep doing it manually. It is likely a core part of your job doing that task.)
  • Can be clearly defined by rules (If this is the case, you should implement an RPA or iPaaS solution. However, make sure you are prepared to handle cases that are not picked up by the RPA.

Determine the practical feasibility (each point counts 1 on a 1-5 scale)

  1. If data is required: The data is accessible (Do you have the required data on hand or can you collect it easily?)
  2. Automation could fit current workflows or the redesign could be done quickly (Will the integration be seamless or will you need to change current processes in a way that would require extensive training of employees?)
  3. Automation could fit within current systems (Is your current soft- & hardware able to integrate with the automation you imagine?)
  4. You have experience with similar automation or know how it should work (Can you imaging what the final solution should look like? Do you know which types of tools to use and how they can be connected?)
  5. A solution could likely be implemented in < 5 months (Think about all the stakeholders involved. Are they ready to switch? How much effort will it take to create a solution?)

Step 4: Develop a plan of action

Now that you selected a set of realistic AI implementations in your company, it is time to develop a plan of action. The goal is to establish a first showcase solution that can pave the way for further AI developments. You should start with the projects that are the easiest to implement (Feasibility < 3) to generate quick wins.

Keep synergies in mind while assessing which use case you want to start with. If your company already has automated processes that draw from the same data source or are in other ways similar to your new use-case, you should start with them. If that is not the case and you need to start from scratch, you should start with the least complex solution that will be the quickest to implement.

You should only start looking at which solution could provide the most value once you have multiple candidates that are all quick and easy to implement. Keep the others, likely more valuable use cases in the backlog for when your first solution is up and running. This will lower both friction and push-back for the future.

If you followed our guide, we hope that you have gained a thorough understanding of the daily processes that happen in your company and their linkage. Moreover, we hope you have built an analytical eye to spot opportunities for AI optimization and automation wherever they occur.

After you chose the first use case, make sure you stay with it and implement it correctly. Click here for a guide on how to successfully implement AI-powered automation.

Share on social media: 

Now that you're here

colabel is a tool that allows you to train AI models on images, documents, and text data. You can rebuild manual workflows and connect everything to your existing systems without writing a single line of code.

If you liked this blog post, you'll probably love colabel.

Thank you! We'll be in touch soon.
We are sorry - something went wrong. Please try again!

More from our Blog

Human-in-the-loop in machine learning: What is it and how does it work?

Human in the loop (HITL) is a well-known and powerful concept for reaching outstanding collaboration and performance in artificial intelligence.

Read story

5 easy ways to save money using AI in your business

Almost any company can leverage the benefits of AI – here are five ways that can be implemented rather easily.

Read story

3 mistakes people make when thinking about AI

There are some conventional pearls of wisdom about AI that are simply wrong or no longer true in an absolute sense.

Read story

Stay inspired.

Sign up and get thoughtfully curated content delivered to your inbox.
We will never share your email address with third parties. And we don't like spam either.