Since ChatGPT launched in November 2022, the ubiquity of words like “inference,” “reasoning,” and “training data” is a sign of how much AI has taken over our consciousness. These words, once heard only within the hallways of computer science labs or within the conference rooms of major tech firms, can now be heard in bars and on the subway.
Much has been written (and far can be written) about the right way to make AI agents and copilots higher decision makers. Yet sometimes we forget that, at the least within the near future, AI will complement fairly than replace human decision making. A pleasant example is the enterprise data corner of the AI world, with players (on the time of this text's publication) starting from ChatGPT to Glean to Perplexity. It's not hard to assume a scenario where a product marketing manager asks her text-to-SQL AI tool, “Which customer segments gave us the bottom NPS rating?”, gets the reply she wants, perhaps asks a couple of follow-up questions, “…and what if you happen to segmented it by geography?”, after which uses those insights to regulate her promoting strategy planning.
This is AI that augments humans.
Looking even further into the long run, there’ll likely be a world where a CEO can say, “Design me an promoting strategy based on existing data, industry best practices on the subject, and what we learned from the last product launch,” and the AI will provide you with a method comparable to that of human product marketing manager. There might even be a world where the AI is self-directed and decides that an promoting strategy could be idea and starts working on it autonomously to share with the CEO—that’s, it acts as an autonomous CMO.
Overall, it's protected to say that until artificial general intelligence (AGI) is introduced, humans will likely be involved in making essential decisions. While everyone has opinions on what AI will change about our skilled lives, I need to come back back to what it won't change (anytime soon): good human decision-making. Imagine your small business intelligence team and its bevy of AI agents are putting together an evaluation for you on a brand new promoting strategy. How do you employ that data to make the very best possible decision? Here are some tried-and-true (and lab-tested) ideas that I live by:
Before you view the information:
- Set the go/no-go criteria before you see the information: People are known to maneuver the goalposts within the moment. It can sound something like, “We’re so close, I believe one other 12 months of investing on this project will get us the outcomes we would like.” That’s the sort of thing that makes leaders keep pursuing projects long after they’re viable. An easy tip from behavioral science will help: set your decision criteria before you see the information, after which keep on with them once you have a look at the information. That will likely result in a much smarter decision. For example, resolve, “We should pursue the product line if >80% of survey respondents say they’d pay $100 for it tomorrow.” At that time, you’re unbiased and may make decisions like an independent expert. When the information is available in, you’ll know what you’re in search of and stick to the factors you set, fairly than coming up with recent ones within the moment based on various other aspects like how the information looks or the mood within the room. For more information, see the endowment effect.
When the information:
- Have all decision makers document their opinions before sharing them with one another. We've all been in rooms where you or one other senior person has announced, “That looks so great—I can't wait to do it!” and one other person excitedly nods in agreement. If another person on the team who’s accustomed to the information has serious reservations about what the information says, how can they voice those concerns without fearing backlash? Behavioral science tells us that no discussion needs to be allowed after the information is presented, aside from asking questions for clarification. After the information is presented, have all decision makers/experts within the room silently and independently document their thoughts (you possibly can be as structured or unstructured here as you want). Then share all and sundry's written thoughts with the group and discuss areas of disagreement. This will be sure that you’re really making use of the group's broad expertise and never suppressing it because someone (normally in authority) has influenced the group and (unwittingly) prevented disagreement from occurring in the primary place. For more information, see Asch's conformity studies.
When making decisions:
- Discuss the “mediating judgments”: Cognitive scientists Daniel Kahneman has taught us that each big yes/no decision is definitely a series of smaller decisions that, taken together, determine the large decision. For example, replacing your L1 customer support with an AI chatbot is an enormous yes/no decision made up of many smaller decisions, equivalent to “How does the fee of the AI chatbot compare to humans today and as we scale?”, “Will the AI chatbot be the identical or more accurate than humans?” When we answer the one big query, we implicitly take into consideration all of the smaller questions. Behavioral science tells us that making these implicit questions explicit can improve decision quality. So make sure you discuss all of the smaller decisions explicitly before talking concerning the big decision, fairly than jumping straight to “So should we proceed here?”
- Document the rationale behind the choice: We've all seen bad decisions that inadvertently result in good results and vice versa. Documenting the rationale behind your decision, equivalent to “We expect our costs to diminish by at the least 20% and customer satisfaction to stay the identical inside 9 months of implementation,” will help you truthfully revisit the choice the subsequent time you do a business review and determine what you probably did right and what you probably did fallacious. Building this data-driven feedback loop can provide help to hold all decision makers in your organization to a better standard and separate skill from luck.
- Set your “kill criteria”: Refer to the choice criteria documentation before you see the information. Set criteria that, if not met quarters after launch, will indicate that the project isn’t working and needs to be terminated. It is likely to be something like this: “>50% of shoppers interacting with our chatbot ask to be transferred to a human after interacting with the bot for at the least 1 minute.” It’s the identical concept that moves the goalposts: you get “equipped” with the project when you get the green light to do it, and begin to develop selective blindness to signs of performance deficiencies. By setting the kill criteria up front, you’re sure by the mental honesty of your former unbiased self, and make the fitting decision to proceed or terminate the project once the outcomes start coming in.
If you're considering, “That appears like quite a lot of extra work,” you'll find that this approach will quickly develop into second nature to your leadership team, and each overtime invested will bring a high ROI: It ensures that your entire company's expertise is utilized and puts safeguards in place to limit the downside of the choice and to learn from it, whether the choice goes well or badly.
As long as humans are involved, working with data and analytics generated by human and AI agents will remain a particularly priceless skill – especially navigating the minefield of cognitive biases when working with data.
