On Resisting the Urge to Boil the Ocean (with AI)

Completely, 100% written by a human.

In my last post, I broke down the technique that separates ‘sounds good’ AI implementations from high-value, ROI-seeking ones. That technique was this: prioritize the problems you want to solve for, over chasing the best models (or tech) available to get there.

See my post with the cheeky title ‘Chase problems, not models‘.

But there is a downside to chasing problems too vigorously, if you’re not careful. In your quest to feed up all your org’s problems & bottlenecks to AI, you may find yourself overwhelmed for another, contrarian reason.

You want AI to solve for everything, right now.

AI is not going to solve all of your problems overnight, if at all.

Time to say the un-fun thing.

People are correct when they say that AI is going to drastically change the way we do business/have impact. It’s already changing the ecosystems in which our orgs work, as AI exacerbates both the boons and detriments of modern society. (Whether your org enlists AI to fight the good fight is only a secondary question.)

And for those who work in any kind of office setting, you’d be a rarity if the subject of AI hasn’t come up yet. Best case, your employer is causing you to contemplate AI more than you otherwise would. Worst case, it’s upending your work life completely.

But conflating AI’s capabilities to reason or automate – which are great – with its ability to solve every efficiency issue is an error in judgment. And here’s why:

• Not all efficiency challenges can be automated away. Automation can easily breed more inefficiency — like when it lacks the governance to ensure its reliability, or when it puts additional secondary work on us to verify or correct. (AI has screwed me on both.)
• Not all efficiency challenges are tech gaps. Every org, even the most high-performing & impactful, carries some degree of unique cultural dysfunction. Unless you’re the anomaly, it’s unlikely that all of your org’s challenges are due to lacking the right tech for the job. Sometimes the challenge is something deeper with how your staff operates, and tech will never be able to address that. (Consider this the human element of doing work that, for better and worse, AI still can’t replace.)

Once you accept that AI isn’t a silver bullet for every issue, the next step is accepting that impactful adoption takes time. Part of the reason is on you: your org has limited capacity & resources to tackle everything all at once. (And surely, this has been true for every other domain where you’ve tried it.)

But it’s also AI’s fault. Though the tech has been around for years, it’s become smarter and more widely available in ways that weren’t true before. And that is the newness we’re all contending with: orgs are racing to embrace a state of working with AI that simply hasn’t been time tested.

We’re building the bike while it rides us.

Even the big AI providers like Anthropic or OpenAI don’t know how it will all go down. That’s why the approach to models, agents & token usage seems to shift overnight.

And yet here we are, each trying to steer our respective ships with AI, while simultaneously along for the ride.

Prioritize the problems that are low risk and high reward.

When you’re in AI experiment mode – which I partly define as not having the governance nor resources to properly integrate AI into your workflows – the best problems to take a stab at first are those that:

• Provide the greatest potential for time or cost savings – either because they optimize an existing bottleneck OR the “implementation cost” (the time or money it takes to set the thing up) is small. AND ALSO
• Introduce the least amount of risk – security & privacy are top of mind, but this can also include reputational, financial or programmatic risk

Striking the right balance between these two is the experiment sweet spot. For one, you ensure your team doesn’t start by trying to tackle issues that are too big or high-consequence for AI to get wrong: you can afford for these experiments to miss the mark.

And on the opposite end, you’re not wasting time on items that are too insignificant for your team to begin seeing tangible benefits. This is, after all, the only reason why any org should race to adopt new & constantly evolving technology before it’s had time to mature & settle.

How you define ‘low risk’ is obviously distinct to your org. Every org using AI should absolutely have a conversation about the risks they perceive & how they rank in terms of severity.

But for anyone who has zero idea how to start thinking about this, some candidates for low risk AI experiments are generally workflows that:

• don’t require any personally identifiable or sensitive information. (LOVE anonymized, aggregate data here at Employed for Good 💃)
  
• don’t require sharing intellectual property/the inner workings of your organization. (This is especially true when using generative AI tools without an enterprise license, which tend to offer higher data protections than typical consumer accounts)
  
• produce outputs that are internal-facing (like research or memos) vs external-facing (like your board, funders, or beneficiaries)
  
• are easy to verify the accuracy for and/or are low consequence in the event the information is wrong

And similarly, high reward AI experiments are generally workflows that:

• Follow a consistent pattern that AI can replicate
  
• Your team doesn’t enjoy OR takes a good chunk of their time. To reframe, what would solving this challenge free them up to do?
  
• Won’t lack anything substantial if a human doesn’t own the entire production. Because some tasks do need more humanity than just human-in-the-loop.

At the end of the day, stay grounded.

Somehow, this ends up being the bottom line of every AI post on this site.

We’re currently in a phase where we are going to hear seemingly endless AI awe stories: how it’s revolutionizing work, how it’s helping orgs do things faster and better, how it’s enabling technological innovation never seen before.

And we’ll also continue to hear how now is the time to get on board. Because if not now, it might as well be never! You’ll certainly be left behind!

However you choose to internalize these narratives (or hopefully don’t) – choosing to start your AI journey by throwing it at every challenge is the digital equivalent to throwing spaghetti at the wall.

Yes, solving every bottleneck tomorrow would be absolutely game-changing. But no, it’s not quite realistic…even with AI on the horizon.

So pick your battles. Your AI experiments – along with your team’s focus & sanity – will be all the better for it.