Shaw Centre
55 Colonel By Dr, Ottawa ON K1N 9J2

604.449.2284
SAAS NORTH NOW Blog

Five Things Founders Need To Consider Before Diving Into AI

Reid Robinson, Lead AI Product Manager, Zapier

SAAS NORTH NOW #58

Hello to Canada’s SaaS Community,

The right way to use AI is any way that delivers value for your business. And regardless of social media posts declaring that you’re behind for not already being an AI whiz, looking for the right use case will pay dividends. Speaking with SAAS NORTH after his conference talk, Zapier’s Lead Product Manager for AI, Reid Robinson, shared the things founders need to consider before diving into AI.

Key takeaways:

  • Don’t let fear-based messaging get in the way of AI exploration.
  • As you think about implementing AI in your product or operations, take some time away from your work to think holistically about business needs.
  • Be cognizant of both AI’s limitations and your own organization’s limitations such as policies and design decisions.

Dave Tyldesley

Co-Founder/Producer, SAAS NORTH Conference Editor, SAAS NORTH NOW

Advertisement

Create metrics and dashboards everyone will love with Klipfolio PowerMetrics. Start free now.

Advertisement

Learn more about Sage Intacct for subscription and SaaS companies!

Perhaps the note came from an advisor. Or you saw a viral social media post. But the message was similar: get started with AI now cause you’re already behind. So many growing organizations already leverage AI… so you better get something going immediately.

Right?

Yet this fear-grounded messaging hurts more than it helps according to Reid Robinson, a founder who sold his startup and is now the Lead Product Manager for AI at Zapier.

Speaking with SAAS NORTH after his conference talk, Reid shared more about the realities of AI and how to bring it into your startup the right way.

1. A demo is not reality

One of the best things about generative AI is you can show the result in real time. If you develop a novel prompt, the result is instant and can easily wow audiences across the internet.

But that doesn’t mean the demo actually works.

“There’s a lot of things that people can demo and that’s creating a lot of hype,” said Reid. “Like, ‘Oh my God, I made this demo of something that can basically be your entire sales process.’ But that doesn’t work in reality. At least not today.”

Instead, Reid likes to talk about how AI can help solve real problems that often get overlooked. For example, he’s building an internal product at Zapier that assesses overall sentiment and pulls out key issues from support tickets. Without the support of AI, a support rep would need to spend a lot more time manually reviewing. The result? An estimated savings of 11 hours per day of rep time—and a better outcome for customers since reps can focus on solutions rather than distilling data about the problem.

Advertisement

SOFII is a self-sustaining program funded through FedDev Ontario which supports scale-up of high-growth, innovative Small and Medium-sized Enterprises (SMEs) in rural and urban communities across southern Ontario by offering interest-bearing financing solutions.

2. You need room to pause and explore

Rather than thinking about AI as something you have to do, think about how it might help you solve problems or deliver more value, either in your product or in business operations.

For example, Reid said Zapier recently hosted an AI hack week where employees could take time away from their regular jobs to think creatively about what AI could possibly do for the organization.

That idea–taking time away from your desk to brainstorm–is something Reid recommends for all businesses.

“The reason I say that is because otherwise, you get into a cycle where you’re thinking, ‘Can we squeeze AI into an existing feature? Can we squeeze AI into an existing thing?’” said Reid. “But there may very well be this wild moonshot opportunity to massively simplify or expand possibilities by taking a more AI-first approach or by putting AI under the hood.”

3. Building with AI creates design questions

Once you’ve identified where AI can provide value, a new problem comes up: prompt or embed?

For example, let’s say you want to add AI-based personalization into your onboarding. This is a great potential feature, but you need to figure out how best to present it. Do you provide a natural language bot that asks a user questions, personalizing based on their answers? Or do you collect the data in a different way, let the AI work in the background, and the user is then presented with the finished outcome?

There is no clearcut answer—it depends on user needs, cognitive load, and overall experience.

For example, Zapier initially tried an AI rollout where a chatbot would help you develop a zap with AI. The team soon realized, though, that the classic drag-and-drop or dropdown features in Zapier provide a cleaner user experience.

“It felt like a game of 20 questions to do that with the chatbot,” said Reid. “We learned quickly that there were times when a clickable UI, like a dropdown list or a checkbox is just better than chat.”

Advertisement

Ottawa is the destination where meeting delegates often choose to arrive early or stay on after their conference to enjoy all the capital has to offer – mixing business with culture, adventure, nature, or all three! Learn more online and connect with Ottawa Tourism’s Business Events team.

4. You need clear guidelines for privacy and security

One of the murkiest elements of AI is security and privacy. There are already concerns about how OpenAI and similar organizations are using ChatGPT logs. Some companies may not even be able to leverage open tools like those due to privacy concerns.

The issue, though, isn’t privacy concerns—it’s that organizations don’t yet have cohesive policies.

Before you start building with AI, figure out your own limitations, perimeters, and guidelines. Then make them clear so employees can creatively work within your needs rather than accidentally breaking a rule they didn’t know existed.

“It’s easy to follow the demos, but those don’t often talk about the underbelly of all this,” said Reid. “There’s a whole different security world that starts to be in focus.”

5. No GPUs before PMF

You might be tempted to build your own AI model and train it from scratch. The logic here is that you can train something custom for your use case—and it’s good logic. However, it’s often a waste of resources if you don’t have product-market fit.

Or, as Reid quoted from a viral tweet: No GPUs before PMF.

Sometimes, training your own AI model can make sense. But it’s a huge investment that shouldn’t be taken lightly.

“The models that are available at people’s fingertips today are incredibly powerful,” said Reid. “And if you don’t have product-market fit in what’s available today, then you really shouldn’t go out and spend an absurd amount of time and money to train and do your own models.”

Advertisement

Subscribe to hear the latest news about our 2024 conference

SUBSCRIBE NOW

top

Hello to Canada’s SaaS Community,

The right way to use AI is any way that delivers value for your business. And regardless of social media posts declaring that you’re behind for not already being an AI whiz, looking for the right use case will pay dividends. Speaking with SAAS NORTH after his conference talk, Zapier’s Lead Product Manager for AI, Reid Robinson, shared the things founders need to consider before diving into AI.

Key takeaways:

  • Don’t let fear-based messaging get in the way of AI exploration.
  • As you think about implementing AI in your product or operations, take some time away from your work to think holistically about business needs.
  • Be cognizant of both AI’s limitations and your own organization’s limitations such as policies and design decisions.

Perhaps the note came from an advisor. Or you saw a viral social media post. But the message was similar: get started with AI now cause you’re already behind. So many growing organizations already leverage AI… so you better get something going immediately.

Right?

Yet this fear-grounded messaging hurts more than it helps according to Reid Robinson, a founder who sold his startup and is now the Lead Product Manager for AI at Zapier.

Speaking with SAAS NORTH after his conference talk, Reid shared more about the realities of AI and how to bring it into your startup the right way.

1. A demo is not reality

One of the best things about generative AI is you can show the result in real time. If you develop a novel prompt, the result is instant and can easily wow audiences across the internet.

But that doesn’t mean the demo actually works.

“There's a lot of things that people can demo and that's creating a lot of hype,” said Reid. “Like, ‘Oh my God, I made this demo of something that can basically be your entire sales process.’ But that doesn't work in reality. At least not today.”

Instead, Reid likes to talk about how AI can help solve real problems that often get overlooked. For example, he’s building an internal product at Zapier that assesses overall sentiment and pulls out key issues from support tickets. Without the support of AI, a support rep would need to spend a lot more time manually reviewing. The result? An estimated savings of 11 hours per day of rep time—and a better outcome for customers since reps can focus on solutions rather than distilling data about the problem.

2. You need room to pause and explore

Rather than thinking about AI as something you have to do, think about how it might help you solve problems or deliver more value, either in your product or in business operations.

For example, Reid said Zapier recently hosted an AI hack week where employees could take time away from their regular jobs to think creatively about what AI could possibly do for the organization.

That idea–taking time away from your desk to brainstorm–is something Reid recommends for all businesses.

“The reason I say that is because otherwise, you get into a cycle where you're thinking, ‘Can we squeeze AI into an existing feature? Can we squeeze AI into an existing thing?’” said Reid. “But there may very well be this wild moonshot opportunity to massively simplify or expand possibilities by taking a more AI-first approach or by putting AI under the hood.”

3. Building with AI creates design questions

Once you’ve identified where AI can provide value, a new problem comes up: prompt or embed?

For example, let’s say you want to add AI-based personalization into your onboarding. This is a great potential feature, but you need to figure out how best to present it. Do you provide a natural language bot that asks a user questions, personalizing based on their answers? Or do you collect the data in a different way, let the AI work in the background, and the user is then presented with the finished outcome?

There is no clearcut answer—it depends on user needs, cognitive load, and overall experience.

For example, Zapier initially tried an AI rollout where a chatbot would help you develop a zap with AI. The team soon realized, though, that the classic drag-and-drop or dropdown features in Zapier provide a cleaner user experience.

“It felt like a game of 20 questions to do that with the chatbot,” said Reid. “We learned quickly that there were times when a clickable UI, like a dropdown list or a checkbox is just better than chat.”

4. You need clear guidelines for privacy and security

One of the murkiest elements of AI is security and privacy. There are already concerns about how OpenAI and similar organizations are using ChatGPT logs. Some companies may not even be able to leverage open tools like those due to privacy concerns.

The issue, though, isn’t privacy concerns—it’s that organizations don’t yet have cohesive policies.

Before you start building with AI, figure out your own limitations, perimeters, and guidelines. Then make them clear so employees can creatively work within your needs rather than accidentally breaking a rule they didn’t know existed.

“It’s easy to follow the demos, but those don’t often talk about the underbelly of all this,” said Reid. “There’s a whole different security world that starts to be in focus.”

5. No GPUs before PMF

You might be tempted to build your own AI model and train it from scratch. The logic here is that you can train something custom for your use case—and it’s good logic. However, it’s often a waste of resources if you don’t have product-market fit.

Or, as Reid quoted from a viral tweet: No GPUs before PMF.

Sometimes, training your own AI model can make sense. But it’s a huge investment that shouldn’t be taken lightly.

“The models that are available at people's fingertips today are incredibly powerful,” said Reid. “And if you don't have product-market fit in what's available today, then you really shouldn't go out and spend an absurd amount of time and money to train and do your own models.”