In the relatively brief period of history since OpenAI released ChatGPT and captured the world’s attention, the Artificial Intelligence (AI) surge has already made its way into the daily processes of countless businesses. Is it at the level to justify the breathless amount of hype that we are collectively being bombarded with every day? Maybe not, but there’s more happening across all aspects of business, than many might think.
There’s a lot of AI enabled ghostwriting taking place as part of daily, routine business communications. It’s actually happening all over the place, yet employees are often using it outside of the formal realm of any kind of named, official “IT Initiative”.
Sure, it’s a focus group of two, but consider these examples:
I recently discovered my 24 year old daughter uses ChatGPT to clean up her weekly performance updates to her management. She spent a little time tuning the prompt and now simply loads up the facts and receives a well crafted executive summary in less than five seconds. She’s not an IT person and she’s not yet a confident business writer, but she’s saving hours of time and anxiety each week while simultaneously optimizing vital executive communications.
In another example, The top IT executive at a software company disclosed to me that he used ChatGPT to write every piece of internal correspondence. For him, the choice was all about efficiency and a desire to satisfy a curiosity about whether AI could become a viable part of the office tool suite. He never admitted it, but he had become dependent on AI to help him with business communications. He will likely never write his own emails ever again. Think about that.
It’s been well over a decade since Siri and Alexa complemented smartphone keyboards with a language interface. AI’s ability to communicate is mature. There’s an entire generation of workers who grew up with it. But bringing AI into the business operations fold is just getting started.
When the Internet Isn’t an Option for Teaching LLMs
In many aspects of business, AI is being applied to more than just writing copy. It’s being asked to perform specific business functions that are often specialized, yet highly redundant and administrative in nature.
Unlike turning a discombobulated set of notes or ideas into polished essays and blog posts about a common topic, the Internet is not as worthy of a knowledge source when content (data) and workflows pertain to a specific process, in a specific industry, at a specific company.
In a private value chain setting, the data required to achieve acceptable levels of AI-powered operational precision is not always readily available. ChatGPT can scour the Internet and find billions, or possibly trillions of data points that are readily available sources of data for the AI to work its magic. It’s worth noting that most of that Web content was created by humans and could very well become increasingly restricted as copyright holders fight back.
Private settings present a different AI dynamic.
For example, if the AI is trying to understand and process complex logistics documents that are non-standard, in different languages, with different units of measure and then apply the right data elements into specialized workflows, the Internet is not really very helpful. Converting kilograms to pounds as a standard unit of measure is one thing, determining how to match a commodity code with line item cost in a commercial invoice and then reprocessing it for a 25 year old AP system so accruals are in policy is a different animal. Then, scale it up to 10,000 times a day across 200 different supplier invoice formats and even the mighty ChatGPT will struggle.
Modern solutions that are built around AI come with the ability to address org-specific LLM training. This capability is built into the workflow orchestration tools so the human operators are able to drive the AI knowledge based to benefit their company’s nomenclature and processes.
The path forward is not a magical AI black box where garbage goes in and perfection comes out. The industry-centric solutions of tomorrow will have a mix of broad LLMs (time zones, currency, locations, document types) that all companies might use combined with partitioned zones, where the models are being tuned to support the processes, systems and nomenclature of a single organization.
“We Need to Train our Models”
Several recent Raft customer calls have included some really profound AI moments. It’s not because they are talking about AI, but rather we routinely see a group of business users talking about how they can train and optimize their LLMs so the AI-powered process automation keeps getting better.
“Training our Models” is becoming part of business vocabulary. It’s not just in IT circles. The customer calls described above were operations people - customs brokers, logistics managers, finance managers - who are responsible for keeping inventory moving around the world. But they are also the same people who have been suffering for decades dealing with a massive flow of unstructured communications and documents that arrive via email, PDF, Word, Excel, PNG, EDI, XML, and API.
AI is replacing redundant, slow administrative processes with machine learning that consumes and processes unstructured information and surrounds it with advanced workflow orchestration capabilities. For every document that goes through without the need for human intervention, it’s a “zero touch” moment. If 90% of documents can be handled with AI, that’s a solid new class of KPI because fewer costly, important human resources were required to get the same job done.
While the fear of AI taking all the jobs is common, it’s important to remember that it’s taking on aspects of jobs that nobody really wants to do. In logistics, it also means the vital human resources can be focused more on moving freight and servicing customers. Most people would prefer to spend their days helping customers versus rekeying.
Rob Ardesi, COO at Navia Fright recently said, “AI now takes care of more than 3,000 minutes a week of document processing, which previously were tasks that our staff did not enjoy”
We’re in the early stages of a long AI journey, but if the early results are any indication, what’s beginning to happen today will be with us forever, getting better every day.