Addressing Workflows with AI

How do AI developers build effective solutions that customers actually want to use? We’ve spent plenty of time unpacking how technologists and end users can implement and utilize AI tools, but we’ve spent less time talking to the folks behind the tech.

?

So when the creator of an AI tool I use in my own workflow reached out, I had to learn more about his approach and process. Draft Horse AI and Swell AI CEO Cody Schneider joined me for a recent bonus episode of The Business of Tech , and I think there’s a lot to learn from his overall AI philosophy.

?

From selecting training data to keeping applications practical to convincing customers to give AI a shot, here’s a rundown of our conversation on workflow automation.

?

Defining Success in Workflow Automation

?

Schneider focuses on automating and enhancing workflows with AI. When I asked him about his overall approach to this application, he said that AI is best at taking unstructured data and structuring it.

?

In practice, his goal is what we’ve heard from similar AI experts – to have AI take over the repetitive, boring backend work so that workers can focus on the more human aspects of business, eventually cutting costs while improving efficiency.

?

The Data Management Process

?

Now, when we talk about unstructured data, it’s often a chaotic mess that feels too disorganized to throw into an AI. How do his companies and clients handle data management to extract success from AI?

?

In short, Schneider’s clients work with a lot of public information or use internal documents that don’t need too much cleaning up. His approach is simply to not overthink it.

?

For example, in the content space, his products can use data from podcast episodes, google results, YouTube videos, internal documentation, branding documents, and white papers written by clients to generate results:

?

“I can take these ten different white papers based on this current client situation, and I can now create a custom report or a custom output for them by providing this unstructured version of the data. It doesn't have to be perfect. It can extract that knowledge and information and then transform it or translate it into the format that you're trying to get it,” he said.

?

Differentiating Between Hype and Practicality

?

Some of the scenarios Schneider mentioned – sales, workflows, content transformation – were reminiscent of the overhyped AI applications that gained and lost traction in the last couple of years. His companies are quite successful, so how does he differentiate between that hype and practice applications? What are the areas he finds most useful?

?

He agreed that AI tools based solely on general internet knowledge produce average results. So, one way to glean true value is to use data unique to your organization’s expertise:

?

“Expert knowledge has actually become more valuable than ever before. And so what we see as the real value is when you have your own data sets, you have your own libraries of content, you have your own proprietary information that you're then using AI to work off of,” he said.

?

Another area of reliable value is focusing on automating the types of tasks people truly can’t stand doing – the types of tasks people might quit a job over. He believes that 9 times out of 10, there’s some way to augment that piece of the workflow, leading to more retention and better client success.

?

Avoiding Average Outputs

?

Let’s zero in on the idea of average results – how does Schneider help users avoid that fate?

?

At an expert level, Schneider pointed to big companies like Stripe and Arrow Electronics, who have had success modifying ChatGPT-style models with their own branding and style.

?

Narrowing that approach down to a small business level, he aims to help customers create AI style guides with a defined persona, style, tone, voice, and prompt approach. This helps create consistency and avoid ‘garbage in, garbage out,’ results:

?

“Once you have that library, that back catalog information or knowledge, AI is very capable of understanding the language that a company speaks,” he said. “You're just going to get a way higher quality piece generated.”

?

Minimum Iteration

?

One of the main reasons I wanted to chat with Schneider is because he has a theme of ‘minimum interaction’ across his products. What does that mean in the AI arena?

?

Schneider explained that, like any other early-stage company, they wanted to get their products in the hands of the public as fast as possible. That was his North Star goal while developing AI, but he knew that he had to take small steps to get there and sometimes even veer a bit off course to reach the final destination. Development-wise, this philosophy enables him to tap into the mindset of customers on the ground quickly:

?

“You're gonna end up in a different place, but you don't know that until you take a step forward, right? Relating that to the product development side, that's how we try to approach all of this… It’s the only way that we found to actually make things that people want to buy. We have theories about what people want to purchase, but in reality, until it's in the public, and they can interact with it, you have no idea,” he said.

?

Schneider’s Advice: Get Comfortable With Chaos

?

So, what’s the best practice Schneider recommends for service folks wanting to implement AI?

?

He says to focus on getting it good enough because you can always improve it. If you can get an AI system 80% of the way there in 24 hours, it will still generate impact for customers, especially when you get it to 100% a month down the line. He also believes this approach works even better for small businesses because they’re way more flexible, and that it gives you a better chance of getting clients excited about a tool:

?

“Especially with this newer technology, there's no way to make it perfect… every once in a while, it's just going to totally mess up, and we just have to kind of be OK with that. But again, if it's 80 percent of the way there, they're going to be ecstatic. And that creates a high-impact activity that you can do for these people,” he said.

?

While the ‘progress over perfection’ approach is all fine and good for developers, MSPs aren’t used to that kind of fluidity. How does Schneider wrestle with those opposing realities?

?

When clients start using an AI tool, he suggests positioning them as curators rather than producers. Aim to get them to accept that AI isn’t perfect, but that it can be good enough for handing off work and focusing on human-driven magic.

?

“The framing that I'm trying to brainwash as many founders to think about is, how can I make my team that's already exceptional perform like they're almost superhuman by augmenting them with these tools?” he said.

?

?

Do you see value in using the developer mindset when implementing AI solutions for customers? As always, my inbox is open for reactions, insights, stories, or whatever else is on your mind.