DigitalTransformationTalk: Using AI to help process automation reach peak efficiency

On 23 April 2024, Digital Transformation host Kevin Crane was joined by Ayman Husain, Director, Customer Success, Data, AI & Advanced Analytics, Azure Intelligent Cloud, Microsoft; Jamie Grabert, President/Co-Founder, The Consultancy Group; and Maxime Vermeir, Senior Director - AI Strategy, ABBYY.

Views on news

Adobe is building its own video-generating AI model like OpenAI’s Sora. The company behind Photoshop and Premiere Pro software is now offering artists roughly $3 per minute of video footage to train its new AI model. Adobe wants "more than 100 short clips" of humans showing various emotions, completing different physical activities or doing things with objects, or clips that show the human anatomy.

Generative AI wasn’t developed but discovered, and now we are at the stage of finding out more about its capabilities, which requires a lot of data. On an organisation level, businesses have already made a lot of efforts to get their data accessible, structured and retrievable. Now it’s time to take it to the next level and ensure their accuracy and the absence of bias too. The wild west period of the early days of cloud computing when data was up for grabs is about to end now with ever more regulations coming into force.

While AI companies are running out of public-facing data on the internet, businesses keep an increasing amount of data to themselves for proprietary or compliance reasons. As a result, companies now gravitate towards creating their own models. As a result, they get much more customised and niche solutions. The example in the article shows that we are entering a more mature phase of data procurement where terms are negotiable between the owners of the data and tech companies. And the better the quality of the images put into these models, the better the outcomes will be for users. 

Opportunity risk versus the risks of AI deployments

AI has been around for quite a long time but the launch of ChatGPT marks the line when it became available for the masses – which has raised public awareness of its possibilities. The automation of tax returns is a good example of how different types of AI (small and large language models, symbolic AI) are combined to automate a complex process end-to-end. However, there are plenty of typical pain points where AI can help address problems. Some labour shortage can be addressed by chatbots, AI tools enabling data entry and systems that can talk to each other by liberating time in supply chains as well. For SMEs, deploying AI can make a lot of impact on their growth trajectory. 

Processes that lend themselves readily to automation include sales, marketing communications (demand generation can be fully automated) and data analysis. Other opportunities on the operational side are procure-to-pay including accounts payable and receivable – which are now regarded as low-hanging fruits for automation. The other major application opportunity is in customer facing processes, which can be more impactful but are also more risk fraught. 

Brexit in the UK has resulted in the sharp increase of overheads – customs clearance and other documentations – but Ai purposefully built for the automation of document-driven processes (there are 159 touchpoints involving documentation when goods are shipped from the UK to Europe or vice versa) could automate 90 per cent of this workload. Some processes, however, shouldn’t be automated with regards to safety to humans. Although robots can remove a cataract now in 20 seconds per eye, a surgeon will always be there to supervise the robot. Autopilot on planes has been around for almost 40 years but we still need human pilots in the last mile of decision making. 

In some areas like HR, the deployment of AI shouldn’t be pushed too far as data quality and the cleanliness of data is still an issue in many cases. Another concern is whether AI is used ethically, and if people are informed as they should be that they are interacting with it. Hallucinations are a unique feature of GenAI – you won’t encounter this problem with other types, such as NLP or ML, which are discriminative. You can combine the two to ensure that GenAI gets the right data and avoids hallucinations. Retrieval augmented generation is a technique that is being extensively used to keep hallucinations to a minimum. 

Another source of risk is that companies don’t always understand their own processes. Ideally, a process is made up of not more than 10-15 steps but, in reality, most of them tends to be much more complicated than that – sorting out these processes should be the first step to automation. There are also cases when older technologies such as OCR are taken to the next level by increasing computational power to enable it to transform, for example, unstructured corporate data into the right format.

The panel’s advice 

• You need a combination of different AI tools to fully harness AI ‘s capabilities for the design of enterprise level tools.  
• The opportunity cost of avoiding AI deployments is losing out to competitors who do implement these tools.
• AI can play a pivotal role in modelling scenarios for supply chains or enabling decision making. 
  To make the most of AI, skill up.
• Use ChatGPT to learn about AI. You’ll need the skill even if it’s not your core comfort. 

For Abbeyy’s podcast on AI, click here.