How To Validate Startup Ideas Using AI

Save months validating startup ideas

Idea validation is all about building something people want. What if you could cut down startup idea validation from 3 months to 3 weeks?

In this post, I explore how AI can help you validate startup ideas, leaving no stone unturned, and be fully aware of your most critical assumptions and what to focus on next.

To demonstrate the role of AI in startup idea validation, I specifically go deeper into the task of “talking to customers” by proposing AI tools that don’t exist yet, demonstrating useful prompts using ChatGPT, and testing available tools where possible. I’m using an example idea, Inbox Zero, for demonstration purposes.

I skip other steps like market research, competitor analysis, etc., and assume AI could play a similar role as unpacked in this post for the process of “talking to customers”.

What is idea validation and why is it important?

There is a lot of stuff written on this topic and it can feel a bit noisy. Ultimately, the goal is to make sure you build something people want and you know where to find your initial users. Lenny’s Newsletter is a pretty good resource to see how idea validation fits into the process of starting a startup.

More often than not, founders don’t really follow an A > B > C approach. They poke and poke until they feel convinced enough to go after a problem.

That being said, building something no one wants is an easy mistake to make, especially if you skip a few critical steps, like:

• Talking to customers

• Researching the market landscape

• Starting with a problem, instead of a solution

• Testing if people will pay

• etc.

Even if you follow these steps, you may still end up getting stuck with common pitfalls:

• Overcomplicating the initial solution

• Sunk fallacy cost and fear of pivoting

• Misreading the market

• Neglecting distribution and marketing

• Wrong timing

• etc.

Even if you manage to take care of the critical steps and avoid the common pitfalls, you may still end up falling short or quitting too soon. There is a high level of uncertainty running through a founder’s head:

• Did I talk to enough customers?

• Did I ask the right questions?

• Did I go after the right market segment?

• What needs to change for the timing to be right?

• etc.

What you want is customers knocking down the door even though your product may still be half-baked. More often than not, this is not actually the case. For example, Airbnb launched 3 times and it took months before they saw any kind of growth effect.

How do founders approach idea validation?

I have talked to a couple of my founder friends who have a product in the market and generating revenue. I asked them a series of questions to get a better idea of how they approached startup idea validation and provided a few responses below. The goal is to identify things that are unusual or unique and would be a challenge for an AI to do.

1. How did you decide what to work on in the beginning?

Everyone had a different entry point for identifying a problem to work on.

Philip from Offerzen started with deep introspection:

The intersection of things I’m passionate about and what will make a good business. Lots of soul searching, research, talking to other founders etc. - Philip Joubert from Offerzen

Jos van der Westhuizen from Remio went through 3 phases:

Brainstorming and building - Eventually we realized we were not talking to customers enough.

Talk to customers - We switched to talking to customers within our expertise but struggled to find clear problem overlaps. Later we decided to look at problems outside of our domain but realized these problems did not make us feel motivated enough when asking ourselves if we would work on it for 5 or 10 years.

Pre-hype cycle - Currently we are focused on areas that are pre-hype cycle and try to get really good at it. - Jos van der Westhuizen from Remio

Gabi identified a striking problem that inspired her:

26 million software engineers in the world and we need dramatically more tech skills. Software engineers also don’t make the best teachers. 26 million teacher shortfall globally today. How do you get more teachers? How do you get software engineers in the classroom? - Gabi Immelman from Mindjoy

2. Were you solving your own problem? If not, what did you do to better understand the problem/domain?

Yes, to some degree. I signed up for 100+ products in the space, spoke to dozens of potential clients and to founders in the space. - Philip Joubert from Offerzen

Not for the bigger problems. Basically, reach out to people the Mom Test way. - Jos van der Westhuizen from Remio

If you are not familiar with The Mom Test, it emphasizes that when seeking feedback on a new idea or product, you should ask objective, non-leading questions that even your mom couldn't lie to you about in her response.

3. Looking back what was the high-level process/steps to validate your startup idea? How long did each phase take?

The whole thing took 2 months. It wasn’t a super clear A => B => C process. - Philip Joubert from Offerzen

First idea took 2 months to invalidate until we realized users wouldn’t pay for it. The second idea took longer to invalidate. For smaller ideas, we took the usual approach: landing pages. Set a target for something in x amount of time. If things go way over, it is a clear sign. - Jos van der Westhuizen from Remio

Jos and his co-founders did something next level to help them invalidate their ideas faster:

We offered customers something else. For example, a free year on the subscription or a free pizza. - Jos van der Westhuizen from Remio

Obviously, if customers take the free pizza instead of the free subscription, whatever you are doing is not valuable enough for them.

4. Did the thing you originally set out to work on change? In what way?

Clearly, it can go either way:

Not really - it’s remarkably similar. - Philip Joubert from Offerzen

Yes, it changed a lot. Learned from customers what they actually want. VR products changed from collaboration tools, team building tools, and consumer games. - Jos van der Westhuizen from Remio

5. What approach did you use for defining your target customer?

We are building a marketplace, so used marketplace theory to pick a hyper-targeted audience (tech startups in CT) - Philip Joubert from Offerzen

What is interesting about Remio is how their approach changed as they pivoted from enterprise to consumer.

On the enterprise side:

Spent time getting to know the person and how they think inside their role. Figuring out who needs it, why do they need it, why don’t they need it. Not a specific process other than getting to know the customer instead of just doing a sales call. You try to read between the lines of what do they need. - Jos van der Westhuizen from Remio

On the consumer side:

Spending time with the user and seeing what they are doing. - Jos van der Westhuizen from Remio

6. How did you find and engage with your target customer?

Offerzen required a lot of face time, ideally in person.

LinkedIn, networking, conferences. - Philip Joubert from Offerzen

Again in the case of Remio, they had two different approaches.

On the enterprise side:

Full spectrum: cold emails, paid ads, stopping in the app, and chatting with them. - Jos van der Westhuizen from Remio

On the consumer side:

Engagements are hard. Not everyone wants to speak to you. - Jos van der Westhuizen from Remio

7. What did you measure to help you validate the value proposition?

We didn’t measure quantitative things - customer validation is too subjective for that. You live in the qualitative realm. - Philip Joubert from Offerzen

It was a bit tricky to do this. We looked at how team engagement increased before and after sessions. There were positive results, but did not move the needle. We did testing ourselves too. For consumer games, measure how much time they spend playing. - Jos van der Westhuizen from Remio

8. How much of your validation was a gut feeling vs. actual data?

80% qualitative. - Philip Joubert from Offerzen

I don’t think any of our validation came from gut feeling, but I also don’t think we ever got something that is fully validated. For the consumer thing we are doing now, the data is speaking for itself. Gut feeling plays a role, but it always comes back to the data. - Jos van der Westhuizen from Remio

9. Did you create what customers asked, or did you land on a need that customers didn’t even know they wanted?

We never asked customers what they wanted. We asked about their problems and alternatives to solve them. - Philip Joubert from Offerzen

We did not land on a need customers did not know they wanted. Customers ask for something specific, but it is actually not the feedback session that matters, but it is more when they ask you like “How do I do X” - Jos van der Westhuizen from Remio

10. How did you approach validating your growth thesis?

Setting some targets based on what other similar apps have done. - Jos van der Westhuizen from Remio

11. What tools did you use, how long did it take, what worked?

We experimented with a lot of different things until we found channels that scaled. - Philip Joubert from Offerzen

Landing pages, email campaigns, customer conversations. Try not to bias their answers and see what they say. Typically try to do it in a week or two, but that is perhaps a bit short. I do know of startup stories of iterating on 1 idea every week. - Jos van der Westhuizen from Remio

12. How long did it take to validate your startup idea altogether?

3 years, still in it. Try not to take longer than 6 or 9 months to invalidate. Some were short, like a week or two. - Jos van der Westhuizen from Remio

13. In hindsight, what do you think could have helped you to save time?

I think the answer might be pushing towards an app or tool, but I’m not sure if that would have helped. I feel the problem might be more around how you approach the whole startup thing. The people who have done cool things might end up spending a year, two, or three getting into the problem themselves. Partnering with someone who is deep into something could help. Speaking more to customers from the early days, instead of brainstorming by ourselves. Taking it slower. You end up rushing to find something, but spending time with customers to understand what they want. - Jos van der Westhuizen from Remio

14. What tools did you use in the process of validating your startup idea?

Mailchimp, Hubspot, Mixpanel, Google Analytics, Landing Page Suites. You can also just keep score on a spreadsheet: I have spoken to this many customers, etc. - Jos van der Westhuizen from Remio

15. Did you build anything to help you validate your startup idea by getting early user feedback? How much time did you spend on your early MVPs before getting it in the hands of potential customers?

We made rough mockups for a few things but didn’t spend too much time validating the solution. We were confident that if we could get developers on the platform then companies would use us. - Philip Joubert from Offerzen

Definitely on the wrong side of building too much for most of our apps. There is definitely a lot of things you can validate without building anything. Now that we have something in customers’ hands we can more easily validate new concepts. Our iterations are on a weekly cycle. A week to 3 months is our range. - Jos van der Westhuizen from Remio

16. What was your biggest hurdle in validating your startup idea?

Getting the initial users. - Philip Joubert from Offerzen

Reaching the right customers quickly enough. Depending on the problem, sometimes you need 5 to 10 people’s input on it. But actually finding 5 or 10 people to speak to you about a problem can be hard. I guess that if you can’t find the right people that is already a sign that you should not work on the problem. Reaching the right people to test this with. - Jos van der Westhuizen from Remio

17. What aspects of the startup validation phase did you automate? If you did it now, what would you consider to automate? Are you using AI already?

Not too much. I think you could build a prototype/mockup with AI much faster now. - Philip Joubert from Offerzen

We actually use AI in idea validation already. When describing our personas we take LinkedIn profiles, we put them into a spreadsheet with an app script and GPT API. You can say: help me create a persona. Here is my value prop or sales pitch. Does the person match my ICP (Ideal Customer Profile), why, why not? We discovered our buyer is the school leadership and the teacher is the enabler. The actual end user is the student. - Gabi Immelman from Mindjoy

18. What did you do in the startup validation phase that you think is outside of the typical advice?

Lean startup style “ask customers” has significant limitations. Assuming you’re competent and have decent business knowledge, it’s valuable to have a vision and stick to it (to a degree) - Philip Joubert from Offerzen

I don’t think we did anything extraordinary to validate ideas. Each idea is just slightly different. The typical advice applies, but how you apply it is the tricky translation you have to do. - Jos van der Westhuizen from Remio

19. What did you do in the process to validate your startup idea that you believe an AI would never have been able to do?

Speak to people and get useful nuggets. You need them to trust you. - Philip Joubert from Offerzen

I think AI would be able to do all the steps. - Jos van der Westhuizen from Remio

Three things stood out from the responses above:

1. Make sure your target customer is very very specific and test your value proposition accordingly.

2. Talking to customers is not just about learning what customers want, but also about reading between the lines and building trust.

3. Push towards validating or invalidating your idea quickly.

Next, we’ll take these three observations to see how AI can help with idea validation.

How can AI help with idea validation?

Idea validation takes a lot of time (especially to do right). I am going to dig into the topic of “talking to customers” for the Zero Inbox idea. This step alone consists of a lot of sub-tasks:

• Defining a target customer upfront

• Searching target customers online

• Joining a target community

• Researching your existing network

• Launch a landing page

• Run an advert

• Request for intro

• Cold outreach

Breaking down the steps, as shown above, you realize that AI tools could be a major aid to save time and resources. A lot of these services I highlight below do not yet exist and are perhaps startup ideas themselves.

Defining a target customer using AI

It is not always clear how to define the ideal initial customer. For example, let’s say you are building Slack or Zoom. It makes sense that any company would find it useful eventually, but where should you start? As Gabi highlighted, AI could play the role of a sparring partner to help you develop an argument for why you need to go after a specific target customer.

In this example, I am using ChatGPT to explore who to target as my initial customer for my Inbox Zero idea.

You can use AI to quickly help you narrow down a potential target customer. It can help you test your assumptions and challenge your responses as a brainstorming exercise.

I went a step further asking ChatGPT, what alternative target customer I should consider.

Now imagine taking this a step further where the AI scores your answers against a weighted matrix to return a ranked list of potential target customers. I asked ChatGPT to create such a matrix.

A win? Using AI you can run through multiple target customer segments quickly. I have seen teams spend hours mapping their customer segments. This whole process could be accelerated with an AI tool specializing in this domain to help surface the best starting point.

Searching prospects online using AI

Imagine you could instruct an AI Agent to create a list of prospects with their name, LinkedIn, Twitter, personal messages, and contact details.

There are plenty of non-AI tools that can help with prospecting leads, for example, LinkedIn Sales Navigator, Apollo, Hubspot, Zoom Info, etc. However, these tools tend to require quite a bit of manual effort to build up a prospect list, pre-draft a compelling message, craft follow-up messages, and send it off. It can easily end up taking 3 to 5 minutes per message. If you assume 1 response per 100 cold emails, it translates to at least 5 hours of effort to reach 1 customer.

Hyperbound adapts automated emails by researching prospects using domain-specific knowledge in your CRM and public internet data sources to generate high-quality, reliable personalized emails at scale.

A win? It would be amazing if you could get a list of prospects without having to do any manual reviews, while conversions still perform comparably with more human-heavy outbound campaigns.

Joining a target online community using AI

Imagine you could ask an AI Agent to search and join communities, provide summaries, identify the most active members, and help you craft a strategy to build up reputation and trust in the ecosystem.

A win? Not spending multiple days to get up to speed with who’s who and what’s what in a new community. It would still be important to participate in real life. As Philip pointed out they focused on building trust with their early customers.

I could not find a service that specializes in joining and analyzing online communities. Please let me know if you are working on this.

Simulating customer responses using AI

Instead of reaching out to all the prospects right away, what if you could simulate their responses to a survey to get an initial idea if you have identified a real problem?

Roundtable does exactly this. You can define your target customer and create a survey. The AI then populates the answers based on the target demographic. In the example below the question is asked whether AI will replace certain types of tasks in order to increase efficiency.

Roundtable is still very early. The demo I tried only provided a bar chart on multiple-choice questions. Imagine getting actual simulated responses from possible customers to help you unlock small insights into your potential target customer. For example, for the Zero Inbox idea, I asked ChatGPT to answer questions as a customer support manager and a founder respectively:

A win? As you can see, without talking to actual customers I can already get a high-level idea of how the different needs would translate to different feature requests, potential distribution partners, pricing models, etc.

Researching existing networks using AI

Imagine you could drop in your LinkedIn profile and let an AI tool dissect your LinkedIn network to identify the most ideal candidates, both for direct outreach or for asking for introductions.

A win? You never know, maybe a friend from school or university is working in the exact industry you are exploring. If AI could point you to that person in a matter of minutes, it could be a huge win.

I could not find a service that specializes in analyzing your existing LinkedIn or Twitter network. Please let me know if you are working on this.

Launch a landing page using AI

Creating a landing page is typically a bit further down the funnel than simply talking to customers. However, now that it is becoming increasingly easier to spin up a landing page with tools like LampBuilder, it could be beneficial in a couple of ways:

• Quick explainer - Instead of writing a very long cold email, you could link your landing page that explains your solution in a bit more detail.

• Call to action - Depending on where you are in your problem discovery you could play with the call to action to either being a calendar link, complete a survey, join the waiting list, early buy button, etc.

• Ads-conversion - In order to drive conversion for your ads experiment you’ll typically need a landing page to explain and convert potential customers.

Jos pointed out that it would be a time saver if conversions on landing pages could automatically be tracked.

Here is an example site for the Inbox Zero idea.

A win? It is really quick to spin up a landing page. It is now even possible to spin up multiple landing pages, brands, layouts, pricing, etc. to test different target audiences at the same time.

Run an advert using AI

When it comes to "Talking to customers," running an ad campaign could be a useful way to gauge if you are able to reach/attract your target audience.

There are a lot of existing tools that can help with running ad campaigns. Ad platforms already help with target precision, A/B testing, and optimizing costs. I have not been able to try it yet, but Google is advertising AI-powered marketing solutions.

I can imagine specialized AI tools could take this to the next level to create dynamic ad campaigns and do predictive/sentiment analysis. For example:

• Dynamic Ad Creation - AI tools can create multiple ad versions tailored to different audience segments, optimizing visuals, and messaging for maximum appeal. This video by Paul Covert is a great example of how visual tools like Canva can be used using AI.

• Sentiment Analysis - Beyond just clicks and conversions, AI can analyze user comments and interactions to gauge sentiment, giving deeper insights into potential reservations or praises. For example, if a user joins the waiting list, it could research that user to fine-tune the content further for similar users in the future or follow up with a tailored welcome email to better engage with the user.

• Predictive Analysis - AI can forecast which advertising channels and strategies will yield the best results to get a rough idea of what to expect if pursuing a particular direction.

Combining the ability to spin up AI landing pages and running dynamic ads seems a useful way to really make sure you explore all the corners, before giving up on your idea.

A win? Personally, I have not had any success running ad campaigns for idea validation. However, I can see how an AI tool could in fact abstract a lot of the nuances that I might have missed in setting up an effective campaign in the past.

Cold outreach using AI

Doing cold outreach can be brutal, especially if you do not yet have a strong following or a well-rounded product. Generally, it is advised to try and create as personalized a as possible message. Earlier I mentioned Hyperbound which automates prospecting.

Digging a bit deeper, there are a couple of complementary tools for doing cold outreach. For example, for writing cold emails you can use copy.ai, SmartWriter, Lyne, Instantly, etc. There is even a Udemy course on using AI for cold outreach.

A win? It would be pretty amazing if an AI tool could help you craft a suitable message based on various data points about the target customer and perform on the same level as a more human-involved cold email.

Founder-market fit and AI

In many ways, a big part of idea validation is having access to the target customer. As Philip highlighted the hardest part is to get your initial customers” and to get them to “trust you”.

What will be useful is a tool that gives the founder a realistic view of how well-suited they are for a particular industry. The AI needs more context of your strengths, weaknesses, and experience. For example, if I take the Inbox Zero example a bit further and apply some context to the prompt it may be more accurate in the advice it gives for identifying a suitable initial target customer.

In a scenario where the best-fit options are less obvious, this technique could help founders decide on their next move in some form of qualitative way.

Gut feeling and AI

In the end, you should follow your own gut feeling, vision, or conviction when going after a particular startup idea. An AI can help you make sure you don’t skip critical steps in validating your thinking and help you mold a mental model of the problem space.

What I find interesting is how often founders actually proceed to build startups from qualitative data points instead of quantitive data points. As Philip highlighted, 80% of their conviction to pursue Offerzen was qualitative. This does not mean, you don’t have to talk to customers. In fact, it means you need to talk to customers to such an extent that you are certain about the problem without having to quantify it - you just know it is a problem.

Conclusion

In my opinion, it is clear enough that founders can use AI as a way to develop a deeper insight into a problem space and validate their ideas. It is great hearing how Gabi is already using AI at Mindjoy, not only as part of their core product but also helping them sell better.

It is not a silver bullet, but ultimately AI could help cut out a lot of the manual steps involved in validating the idea. This potentially means being able to validate your startup idea faster and make sure you focus on the biggest assumptions first.

As Philip pointed out “Lean startup style ‘ask customers’ has significant limitations." “Talking to customers” is only one of the steps in validating startup ideas. You will need to consider other factors too, like competitor landscape, MVP development, market size, legal considerations, etc.

A lot of the AI tools do not yet exist for end-to-end start-up idea validation. However, I have seen some capable examples in customer support and sales to assume that we’ll see tools emerge in this space too. I do think AI is opening a whole new product category making it easy to discover problems and help develop products.

Perhaps, what would be cool is an AI tool that takes you through the startup idea validation journey and creates a business canvas as an output.