Based on the speed of AI development, it is clear that artificial intelligence can monitor more data streams, make faster decisions, and in some cases, make better choices than human intercessors. Meanwhile, purely human workflows—with their gaps in information sharing, endless bottlenecks, and convoluted chains of command—often introduce more inefficiencies.
Research from IDC shows that companies can lose up to 30% of their revenue due to inefficient processes. The emphasis on business communication means that the average US worker spends over five hours a day checking email. Due to the pressure of admin work, salespeople spend only 30% of their time selling products. One of the promises of AI is that it can shave down and automate the tasks that can keep us from doing our actual jobs.
For most essential workflows, AI will always need a human in the loop, because AI will probably never be perfect. The question is this—how do you balance AI input with human governance to take the best advantage of both?
While it is not a good idea for businesses to run AI workflows without humans in the loop, adding too much human interaction with AI is a recipe for failure. In an ideal world, you’d want every decision-maker in your company to be able to access AI, but most of those decision-makers won’t be technical users. Frankly, if you subject these users to high-touch interactions with AI products, they’re likely to put a lot of demand on your IT support desk.
This is why Gartner research projects that up to 85% of AI projects will fail. Companies start too ambitious projects too fast; they give their workforce keys to products that they don’t have a use case for yet, and their personnel rejects the experiment.
Rather than start with AI decision support, companies need to place that concept at the end of their tech tree. Instead, they need to start with less flashy concepts that arguably provide more immediate benefit for employees.
Technologies like robotic process automation (RPA) can replace a vast amount of manual effort, for example. You can imagine a salesperson who uses RPA to automatically append call recordings to the CRM, mark opportunities as won or lost, and then dial the phone to call the next prospect. As data accumulates over time, the business can use AI to make small decisions—such as creating a cadence to schedule follow-up calls and emails. Finally, as the business implements more technology over time, it can use AI to help with decisions that require human approval.
Once the workforce is fully automated, there are many different ways to introduce AI to the decision-making process.
One way is to add AI into the strategic decision-making process. By studying how other industries behave in the market and accumulating data on top-performing companies, your AI product may be able to offer input on how to allocate budget to your various departments to maximize growth. This is a decision that you’re free to overrule—growth may not be the be-all/end-all for your organization—but it’s an area where a well-constructed AI is unlikely to be wrong.
Outside of strategic decision-making, AI can excel at decision-support when it offers human participants a range of options instead of binary choices. As opposed to saying, “choose this inventory mix to maximize profit,” AI could offer one mix for profit maximization, one that maximizes customer experience, another one that hedges against supply chain shocks, etc. Offering this range of choices gives human participants more agency while still decreasing their workload.
As the last example, decision support may take the form of a workflow generated via AI as the result of human input. A marketing leader might ask their AI tool to assemble a content campaign with a cadence of emails and social ads designed to maximize conversions. Thus instructed, the AI assumes control of a large number of integrated automation tools and begins configuring them to produce this result. All the executive has to do is sit back and watch their KPIs go up.
In short, the way you add AI decision support has a lot to do with the comfort level of the people in charge and a lot less to do with the particular abilities of AI. Once you successfully add AI to your workflows, you’ll begin to see improvement as a given—you just need to choose the right product and implement it in the right way.
Want to learn more about Bitvore? Download our latest white paper: Using Sentiment Analysis on Unstructured Data to Identify Emerging Risk.