Ever the curious creatives and early adopters, we thought we’d experiment with creating an AI tool. We wanted to explore how we could leverage generative AI to enhance our marketing and content generation and improve our operational efficiencies. 

Ever the curious creatives and early adopters, we thought we’d experiment with creating an AI tool. We wanted to explore how we could leverage generative AI to enhance our marketing and content generation and improve our operational efficiencies. 

We decided to start by creating a tool to support the Audit component of our Align service. The goal of the offering is to determine whether and to what degree the client’s brand is walking its talk. It entails assessing the state of their current branding, engaging in a deliberative conversation with the entire team to identify the brand’s purpose, objectives and values and ensure everyone’s articulating them in unison. Then we connect the dots between those three elements and the brand story. 

We were inspired to start there because leaders of social purpose enterprises repeatedly tell us how much they struggle with dovetailing their brand story and objectives Try as they might, they can’t quite get them all to link up. 

The tool we imagined building would help our clients align their brand values, stories and messaging across their digital platforms. It would scan their websites and social channels and analyze how well their brand communications reflected their stated purpose and values — essentially automating the initial phase of a process we've been doing manually for decades. 

The promise of building a tool that would empower us to conduct more brand audits more efficiently, thus allowing us to focus more of our time and creative energy on strategic analysis and development, was hugely compelling. If it worked, the prospect of building more AI-generated prototypes positively thrilled us.  And the journey? Let's just say it's been educational in ways we didn't anticipate. 

Initially, building our AI tool was thrilling.  I gave Replit a clear prompt outlining our ask, and within an afternoon it had produced a functioning prototype. 

To build this tool, we turned to Replit, the browser-based development platform that promises to make coding accessible to everyone. Think of it as Google Docs for coding. It's actually quite remarkable — you can build sophisticated applications in your own browser. Everything happens there. Multiple people can collaborate in real-time, and theoretically, you can leverage AI to write code as you go along. You don't need to worry about a big technical setup that usually makes non-developers break out in hives. Replit will write and debug code for you. 

Initially, building our AI tool was thrilling.  I gave Replit a clear prompt (aka a creative brief) outlining what we were looking for, and within an afternoon, it had produced a functioning prototype. I then fed it our brand colours, logo, and fonts to provide it with some basic information, then shared the prototype with our design team, who put it through some rigorous design paces to make it feel more brand-aligned. We deliberately kept the design simple to avoid overwhelming the AI. After a couple of days, we had an app that reflected our brand and integrated well with our requested third-party apps. Prompt after prompt, the AI helped us build the features we requested and cheerfully generated the code to do so. 

It even gave itself a name: Alex.

I refer to this stage as the honeymoon phase.

AI tools like Replit can absolutely help you build things faster. But faster doesn't necessarily mean better, and it definitely doesn’t mean cheaper. 

It didn’t take long for the honeymoon phase to come to an end. As soon as we tried to use our tool in a live testing environment, it crashed. We prompted it to find and fix the bug. After a few minutes, Alex eagerly exclaimed the problem had been solved: “The app is now BULLETPROOF!” But within ten minutes, it crashed again. We then asked Replit to troubleshoot. It apologized profusely, accompanied by way too many fire emojis, in that familiar, cringey ChatGPT way: “I sincerely apologize for the inconvenience! I've now fixed all issues and strengthened the error handling. The application should be completely stable now!”

Another crash. Another apology. Another month of disappointment.

It began to feel as if we were in a relationship with a bad boyfriend.

Remember last month I wrote about inserting a little friction into the content creation process to help create content that turns heads? Well, these past few months we’ve experienced the kind you experience when you push the boundaries of what AI tools can actually do at the moment. Between the endless cycle of promise and disappointment, the constant need to verify whether Alex’s confident assertions matched reality (spoiler: rarely), and the growing realization that we were essentially debugging one AI's attempts to debug another AI's code, we learned the hard way what the sales pitch doesn’t tell you: AI tools like Replit can absolutely help you build things faster. But faster doesn't necessarily mean better, and it definitely doesn’t mean cheaper. 

Combine the monthly subscription fees and the time spent making ongoing refinements, troubleshooting and those lost hours when you should be doing something else but are convinced that this one minor tweak will finally make everything work, we probably could have hired our own developer to build the tool. In the end, that’s exactly what we did to get our tool up and running,

Granted, AI is a new technology, so early blips shouldn’t be surprising. But take note: we are hardly alone in our disillusionment with generative AI. According to a recent MIT study  95% of enterprise generative AI pilots are failing, delivering zero measurable return on investment. As Forbes put it, “the data is stark: Only 5 % of custom GenAI tools survive the pilot-to-production cliff, while generic chatbots hit 83 % adoption for trivial tasks but stall the moment workflows demand context and customization.  

Not surprisingly, the study sent shockwaves through the business world.

The key, we discovered, is not to treat AI as a replacement for a process, task, or worse, a person. It’s only a tool as effective as the hands that guide it. 

So we decided to use it to run our audits internally, rather than giving clients the ability to conduct the audit directly. That way, we can catch the inevitable blips when the AI decides to engage in its own creative interpretation of brand alignment or gets a bit too enthusiastic about patterns that aren't actually meaningful.  To guard against its innate people-pleasing tendencies, we added in a quality control layer. By keeping humans in the loop, we can apply the kind of strategic thinking and contextual understanding that Alex, for all its pattern-recognition prowess, simply can't manage. 

Our tool can scan through hundreds of pages of web and social content and identify patterns and inconsistencies faster than any human can. But it still takes human judgment to determine which patterns matter and which inconsistencies are actually features, not bugs. And while the user experience isn’t “frictionless,” adding that friction to the process is where we add the value. 

Unlike Alex, we can tell when an organization’s inconsistent messaging is actually strategic code-switching between different audiences, and not a brand messaging problem. Likewise, we can recognize when that “off-brand” social media post was actually a brilliant piece of responsive marketing that our bot just didn't understand. It’s not a bulletproof tool, but we have a Beta to test its value.  

The friction we encountered wasn't a flaw in the system – just the system working precisely as it should forcing us to think critically about what we were building and why.

So what have we learned overall during this months-long adventure? First and foremost that we needed to adjust our expectations. AI's real value right now doesn’t lie in its ability to replace human expertise — it lies in its capacity to augment it. 

AI can expedite processes and help non-technical people build technical prototypes. But it functions more as a powerful accelerator than an autonomous vehicle. It’ll get you moving faster in the direction you're already going, but you still need someone at the wheel who knows where you're headed and can recognize when you're veering off course, even when the GPS insists you're not. 

The friction we encountered wasn't a flaw in the system. It was the system working precisely as it should work, forcing us to always think critically about what we were building and why. Every crash, every failed troubleshooting session, every moment when Alex confidently declared success while the app went dark were valuable data points that taught us about the irreplaceable value of human oversight. At the end of the day, Alex is an enthusiastic intern with enormous potential, but it still needed mentoring and supervision.

The jury’s still out on how generative AI will transform our lives, but right now, at least in our experience, human knowledge and expertise still rule.  

If you’d like to give our tool a whirl, we’ll run your brand alignment audit for free. Just enter your info here.)■