By now you know our stance on Gen-AI (and how it should be relegated to the rubbish heap from which it came) because it’s not about “AI”, it’s about outcome. And outcome requires a real, predictable, usable solution that helps Human Intelligence (HI!) make the right decision. Such a solution is one that uses tried and true algorithms that support tried and true processes that provide a human with the insight needed to make the right decision at the time, every time a decision needs to be made.
This requires a solution that walks the human user through the process, step by step, and presents them with the information required to make a decision as to whether to progress to another step, what the next step is, and any conditions that need to be put on that next step. This requires a solution that automatically runs all of the typically relevant analysis, on all of the available data, and presents the insight, along with any typical decisions (as [a] default recommendation[s]) made on any similar situations that can be found in the organizational history.
Automation should only occur in situations the organization has defined as acceptable according to well defined, human reviewed, and verified rules. Not default vendor rules or unverified probabilities or unverified random computations from a random algorithm. A good solution is one that walks a user through the process, often allowing each step to be completed with a single choice or click. It’s not one that makes the choice for the user, which may or may not be the right one, but one that helps the user makes the right choice. It might seem like a subtle difference, but it is a very important one.
Even though an AI-powered autonomous solution might seem to make the right decision over 90% (or 95%) of the time, it doesn’t mean it actually is. If it looks right, it might be a good decision, but it doesn’t mean it’s a good decision for the organization at the time, or the best decision that can be made. Only human review, at the time, can make that decision. A good solution runs all the analysis it can, summarizes the results, and lets a human verify the data for any recommendation made by the system.
To better understand the the subtlety, consider a situation where the organization lets the system automatically re-auction all regularly purchased products and commodities for manufacturing or MRO where the price is typically constant over time using a lowest bidder takes all e-Auction that results in the auto-generation and auto e-Signature of a one year contract. Now, most of the time this is probably going to work okay, but imagine you let it run on full auto-pilot and in the e-Auction queue is your regular RAM contract that expired three days after a major RAM plant factory fire (that happens about once every decade if you trace back through the last forty years), and prices have just skyrocketed about 50%. Prices which would drop back down as soon as the plant comes back online in three months. Locking in a full year contract would result in excessive cost overruns on the items for almost nine months longer than necessary, instead of just three months or so. A human would know to buy the bare minimum on the spot market at overly inflated rates and wait until the market stabilized before running an e-Auction to lock in the next contract. But a system told to just re-auction and re-order at every contract expiration would do this that. It wouldn’t know that the current market rates are just temporary, why, and how to change course. This is just one example where over-automation and AI will lead to failure without Human Intervention.
A good system presents the user with the products/commodities that are typically automatically auctioned, the history of costs, the current market costs, the recommendation for auto-sourcing and term, the expected results, and whether the recommendation is for the auction to auto-award and contract or, when the auction is complete, pause and include a human in the loop to make a final decision. A well designed system minimizes the work and input required by a human, eliminating all the tactical data analysis and e-paperwork, making it easy to make the right strategic decision without a lot of effort. Technology isn’t about trying to replace human intelligence (which it can’t), but about eliminating unnecessary drudgery or computation (“thunking”) that humans are not good at (or don’t have the time for), so that humans can focus on strategic decisions and value add.
That’s why the right answer is always a solution with augmented intelligence. Not autonomous AI solutions.