AI is not possible without Automation

As companies build more intelligent systems to utilize data, AI is on the agenda of almost every executive. The problem is that to get the full value from AI, you must automate an entire process. If you think of AI as the intelligent brains of processing data, then automation is the hands. It is critical to get the data and create actions based on the analytics and decisions. That’s why it is not possible to really utilize AI without automation.

Most Systems are not Built Around AI

Let’s look at a simple solution where AI can be used. The solution must collect and process data, make conclusions and decisions, and put the results to operative use. If the whole system is built on AI, the ability to put data-backed decisions into action is severely limited. Think of a self-driving car without an engine: it’s all brains. Most AI solutions are not built to work in a vacuum. Instead, they must work across many systems, including legacy systems.

Let’s take another example to utilize AI, how to automate insurance claim processing. This is a real case I have been involved to implement. There are several phases to complete the whole process:

1. Filing the claim. A policyholder files a claim, which includes other documents, such as receipts, a report of an offence or a medical report. To get it all to a digital format, OCR (Optical Character Recognition) and NLP (Natural Language Processing), might be needed.
2. Collecting additional data. The insurance company collects data from the claim and other sources. The other sources can be, for example, a person’s insurance history from a national database, typical costs in similar cases, credit rating data, criminal records, data from other similar incidents, i.e. all available data that is relevant to see that information in the claim makes sense, is in line with other data sources, within a statistical margin of expected behavior and is not fraudulent.
3. Making a decision. The solution analyzes the data and makes a decision. The decision can be to pay as requested, something less or more, not pay or send the case onward for manual investigation. AI can be used to identify cases for exception handling and to reduce the number of claims that require manual investigation to a fraction.
4. Notifying the claimant. When the decision has been made the solution must create and send a decision letter or email to the policy holder, save the decision and all documents to the systems, start the payment process, inform third parties (for example, national insurance database, health care provider, other parties in the incident).
5. Potential appeal. After this the policy holder might not be happy with the decision and can trigger a new process for that.

AI Brains is 20 Percent of Whole System

As the above example shows, the actual data analytics and decision-making is a small part of the whole process. The other steps, which are handled by automation include: collecting data from different sources, re-formatting data, sending decision data to other systems, and starting actions in other systems. The data typically comes in many different formats, and part of the data could be missing or inaccurate. For example, the claim data from the policyholder might have empty fields, data in many different formats and the attachment can include anything, but relevant attachment still missing.

The rule of thumb in the data business, which I can personally attest to, is that 60-80% of the work is to pre-process the data. That is, when you try to bring AI into an enterprise that uses platforms such as SAP and Netsuite, you’ll need to build infrastructure for the AI to work—the AI will end up being 20% of the entire process. Many people get caught up in the AI side and disparage the other 80%.

Get AI Hands

An excellent term to describe the required automation components is AI Hands. Picture hands collecting data from a variety of systems, formatting it, and then getting the results of the AI processing to have a real use at the end. It is easy to forget or ignore the development of hands, when it is fancier to focus on the latest innovations for the brains.

Automation tools such as RPA that work across systems and devices are a perfect example of AI hands. RPA can also bring together software components (e.g. OCR, NLP, data cleaning, APIs) to get the data and then trigger actions (e.g. trigger emails, payments and deliveries). This makes it possible to work with large sums of data in different systems and formats. To get real value, it is mandatory to get all components to work.

The answer is not, however, to put a little bit of AI and a little bit of automation on top of your old processes and systems. This is not making them any more digital or intelligent. Instead, you are only adding one more layer of complexity and the potential for technical problems.

Open Source and Real Developer Tools are Needed

Open source is often the best way to connect multiple platforms, from small and rare systems to major and common ones. To give you an example, there are so many data formats out there. Absent a single format, no company can implement support for all those in their proprietary system; open source is the only option. Vendor lock is also a real risk with these implementations. If the solution is totally tied to one proprietary system and vendor, you can only hope that your future needs are supported.

In an ideal world, the AI hands and AI brains should be based on commonly used and widely available programming languages—Python for example—that help to get brains and hands work together and utilize open source components. Developers must have tools to integrate these solutions to other systems. Sometimes it can mean data scraping, using well-documented APIs, and or even integration with other software components. It is an illusion that you make all this with a drag-and-drop or low-code solution. To build a process with data from several sources and integrate it with machine learning, proper developer tools really are required.

Conclusions

To truly put AI to use, we need to provide it with automation—with AI hands. Leadership must also see that they must invest in automation to implement and utilize AI. If AI is on your agenda, then it is mandatory to look for automation solutions that complement and enable AI. Otherwise, results from AI projects will be limited and likely frustrating. Most AI initiatives fall short at the stage when they should be put to production or automated as a part of a process.

While some new companies can build digital solutions from scratch that don’t require cross-platform integration, this is not possible for most companies. Most of us are left with systems we already have in place, and we must get them to work together. To remain competitive, you need automation and AI solutions that can work with all existing systems and can get AI based processes to work end-to-end.

At Robocorp, we have built automation solutions that can be integrated with other systems and software components. We empower AI and automation developers to work together and get all required components to work together and to create synergy and sustainable value in AI and automation projects. We don’t seek to simply automate some small isolated routines; we offer scalable solutions to automate large-scale processes and to fully utilize the power and potential of AI.

Jouko Ahvenainen is Co-Founder and COO of Robocorp, a serial entrepreneur, and a tech and business pioneer.