When I started my career at one of the world’s largest mobile dev shops, Piotr Gajos taught me the design-led engineering process: gather requirements, create the information architecture, create wireframes, and finally add the visual design — the pretty stuff.

I’ve worked with dozens of designers and design shops, which more or less followed some variation of this process. But over time, I noticed designers started skipping the rest to “save time.” First, they dove into wireframes without producing an information architecture — because most business leaders don’t understand this phase. Then, they began jumping straight to visual design, making clickable prototypes with InVision.

Then, Figma came along and ate InVision’s lunch (sorry, Clark Valberg, InVision was awesome). With Figma, executives could comment and make changes directly in the designs. So, slowly but surely, the design process has become focused on visual design, at the expense of user experience.

I’ve lost count of the number of truly stunning product designs I’ve seen that were utterly unusable.

Now comes the next big shift: AI-led design — and we’re already screwing it up.

What’s AI-Led Design’s current trajectory?

With tools like Base44, Loveable, Replit, and Blink.New, anyone can become a developer — skipping the design process entirely. Based on the best designs I’ve seen from these tools, I don’t have much hope for the future of design unless we define a better process — starting now.

Where does AI-Led Design need to go?

First, we must answer this question about integrating AI into our products:

What can we do now that we couldn’t do before?

In many cases, the answer is surprisingly simple: Keep It Simple, Stupid (KISS). Shazam, an early AI pioneer, had an app with one button. Today, we’re trying to weld together legacy systems and LLMs. Starting with a blank sheet feels daunting — especially for enterprise apps. But, if AI can remove the complexity of the UX, then a paper prototype might be 3-10 sheets of printer paper — not 387.

The ideal process should look like this diagram.

Here’s what a better AI-led design process could look like:

Step 1: Design your product on paper

Grab a stack of printer paper, scissors, and a black pen. Create your prototype in 10–30 minutes. It’s faster than you think — and totally fine if you can’t draw well. Just stick to the process.

If AI is handling most of the heavy lifting behind the scenes, you won’t need to draw dozens of pages. Once you’ve sketched the key screens, capture pictures or a video of your prototype, and move on to step two.

Step 2: Upload a video or images of your paper sketches to a low-fidelity AI prototype builder

This tool may not exist yet — but maybe someone will build it after reading this post. The goal is to turn your sketches into a simple, interactive prototype you can show to customers.

It should feel more like Balsamiq than a polished Figma design. Keep the focus on user experience and the content, not visual finesse.

Step 3: Add the shiny bits — the visual design

With the push of a button, your low-fidelity mockup becomes a high-fidelity product. That button should let you apply a visual style — either one you choose or one based on your brand’s style guide.

Maybe the backend gets built automatically, too. Maybe not. But that’s not the point — the design comes first.

After every step: Test with real customers

Test. Test. Test. Find people who match your ideal customer profile — and keep testing. Keep going until customers start pre-paying for the product.

This isn’t a complete guide to AI-Led Design, but I hope it serves as a blueprint to help move the digital design industry forward.

About a decade ago, a startup would get coverage on TechCrunch for its funding round, and a Chinese company would clone that product in about a week — regardless of how good (or stupid) the idea was. Today, you can use tools like Replit, ClaudeCode, and 8n8 to prototype a clone in 18 minutes and ship an MVP in about a day.

But that’s you — just one person. Not a team of 38 developers sitting in Shanghai.

Right now, building products this way still requires some technical skill — ideally, you’re an experienced software developer. But tomorrow? Tomorrow, a non-technical founder will ship a first-version product in nine minutes.

Think about that for a second. You’ve spent 10× that amount of time in at least one meeting with no outcome. What are the implications for the startup ecosystem as time to market approaches zero?

What are the implications for the startup ecosystem as time to market approaches zero?

You need a defensible AI strategy that builds a moat around your startup idea. Because ideas are no longer a dime a dozen — real, functional products are now too.

In the age of AI, your product’s success depends on four forces: community, network effect, distribution, and critical mass. These aren’t new concepts — but they’re no longer nice-to-haves. They need to be baked into the DNA of your company.

Now, if you’ve got an idea that sparks curiosity and has real revenue potential, by all means — ship it and generate cash flow. Then, focus all your energy on making it defensible.

Community

Building a community is more than just throwing up a Discord server and calling it a day. Discord, forums, and other platforms are just that — tools.

To build a real community, you need:

• A clear pain point — When you solve a problem that many people feel, the community often builds itself. Don’t just be another tool — create something people have been waiting for.

• A captivating vision — Don’t try to boil the ocean on day one. Go all in on one neglected pain point, but show your customers — your community in the making — where you’re heading.

• Engaged customers — If people say “that’s cool” but don’t buy your $49 Starter Plan, you’re not on the path to building a community. You need customers who are excited to pay — and ask for features they’ll pay a premium for.

• Your presence — Waiting a week to answer a question that 83 others also have is a great way to kill momentum. Show up daily. Multiple times a day. Over time, your presence will lead to peer-to-peer support — but only if you’re a consistent voice within your community.

Network Effect

As more people use your AI, the experience should get better. The benefits of network effects are well-established, but few AI products today have them baked into their model.

Over the next five years, 1,000+ companies will offer AI agents for call centers and other inbound conversations. But because these calls happen in silos — one customer, one AI Agent — customer Jack gains nothing from customer Jane’s interaction.

To create a network effect in your AI product, think about crowd-sourcing, collaboration, or marketplace features. A product like Gong benefits from a network effect: as an organization’s conversational intelligence dataset grows, the value of the product increases — and switching costs rise.

Distribution

Buying clicks shouldn’t be your long-term growth strategy. While plenty of people hawk step-by-step guides for “repeatable” growth strategies, proceed with caution when following their advice. Instead, here are four ideas you should consider:

• Become the Distributor — Platforms like Roblox and Salesforce enable others to build and distribute on top of them. How can you help AI companies elegantly distribute their products?

• Become a Thought Leader — Most “thought leadership” is outsourced fluff. Real thought leaders provoke new conversations. What perspective can you offer that cuts through the noise?

• Become a Storyteller — A great story generates buzz, referrals, press — and momentum. It’s how you become Product of the Week on Product Hunt or go viral on YouTube. What’s the story that will spark an emotional response from your target market?

• Become a Hustler — There are dozens of AI SDR tools on the market — use one. What irresistible offer can you put in front of your target market today?

Yes, these are high-level ideas and not a detailed playbook. An entire book could be written on distribution for AI products — and we’ll go deeper in future posts.

Critical Mass

You can have every other piece of your defensible position in place — but if you don’t reach critical mass fast enough, a competitor can still swoop in and steal your community.

A niche AI product for ski slope operators in North America may only need 50–100 customers to reach critical mass. But a broadly applicable product that any company could use like an AI SDR tool may need 500K–1M customers to gain the kind of mindshare that’s hard to replicate.

Critical mass is about planting doubt in a prospect’s mind. You want them to disregard anything that isn’t your product without hesitation. You want them thinking:

“Everyone in my network uses this. They rave about it. I already know it’s packed with value — why would I even consider anything else?”

When you reach critical mass — and layer in the other three defenses — you don’t just fend off competitors. You unlock compounding growth.

What part of your product experience gets better as more people use it?

If you find yourself telling your engineering team to “just ship something with AI already,” you don’t have an AI strategy — and you’ve already lost.

This is like a shepherd moving her flock to a field and telling them, “Be good little sheep. I’ll be back in a week to move you to your next pasture.” No shepherd in her right mind would leave her flock unattended for a day, let alone a week.

Your AI team needs a shepherd too — someone who stays close, steers progress, wards off threats, and guides the herd to greener pastures.

We’ve seen the carnage of shepherdless herds before. During the last tech wave, companies spent millions digitizing paper processes — but many failed to realize meaningful productivity gains. Why? Because they didn’t reimagine workflows to take advantage of new cloud and mobile tech advancements, settling instead for glorified web forms.

Today, most companies are duct-taping LLMs and AI features onto legacy systems and calling it a strategy. It’s not. That’s like building a field services app that makes you upload photos from a separate digital camera, type detailed notes on a tiny keyboard, and manually enter the job location — about as efficient as herding sheep with a spreadsheet.

You need an intentional AI strategy — one that reimagines your experience from the ground up. Don’t ask, “Where do we bolt on AI?” Ask, “What’s now possible that wasn’t before?”

Your engineers may resist. They’ll ask, “Why build a new experience?” or “Why did we hire an AI strategist? I can add AI to our existing product in ten minutes.” Or, “Can’t we just display OpenAI’s output here?” A good shepherd knows when — and how — to fight off these wolves.

A great one ensures the team is guided by customer feedback — often multiple times a day.

To keep pace with upstarts and incumbents, forward-looking AI teams use tools like Replit and Claude Code to ship new products in weeks — not quarters. Your shepherd must lead a team that moves just as fast.

Don’t just ship something — anything — with AI. Find your shepherd, and lead them somewhere worth going.

When I started blogging here a few years ago, the generative AI movement was still in its infancy. Since then, the pace of investment has exceeded $200 billion annually, and AI innovations continue to surprise me daily. I’ve also spent over 500 hours writing prompts for personal projects — and well over 1,000 hours generating value with ChatGPT.

Moving forward, this blog will focus on how you can change the world with your AI. I firmly believe you can create a movement — through your agent, startup, or product — that makes humanity better off. While some AI disruptions will displace jobs, new paths can be forged and movements sparked that bring the power and economic benefits of AI back to the people, where it belongs.

The posts that follow offer practical advice on how to create AI that makes the world a better place.

Referral programs supercharged by software enable you to grow your existing customer base by a consistent multiple month-over-month. For most SaaS and marketplace startups, a referral program is a must-implement customer acquisition strategy that turns your current customers into your most prominent evangelists. Referral programs don't make your app "go viral." Instead, they incentivize word-of-mouth marketing.

Any referral program's goal is to lower your average customer acquisition cost significantly. If you're spending $100 for every new paying customer, your referral program should generate new customers for pennies on the dollar up to $15. It's always easier to increase the value you give referring customers than to pull back the rewards you offer. If you need to provide more value to generate referrals, continue improving your offer until you've found something that excites your customers.

What you need to get started

A great referral program can transform your business, but your company needs to get a few things in place before you get started.

1. Customers

Don't waste your time or resources building a referral program into your product on day one. Without an existing customer base, your referral program will fail to recoup its implementation costs quickly. You're better off focusing on other marketing channels until you have a few thousand monthly active users. Once you've achieved product-market fit and have a few channels driving customers to your product, you can start architecting your referral system. Remember, it's OK if you employ a freemium model and only a percentage of your customers have upgraded to paid plans.

2. A product worth talking about

Your customers need to love, love, love, loooove your product. That's right; your product fulfills an unmet need in the market, or you've solved a problem 10x better than any competitor. Paid and unpaid customers want to sing your praises from the podium of your industry's largest conference. You can still see results from a referral system if your Net Promoter Score isn't close to 100 — but the more your customers love what you do, the better your program will perform.

3. A self-serve product

While creating a referral program for your enterprise software product with a 6-12 month sales cycle is possible, the results will always lag when compared to self-serve SaaS and marketplace products. When customers can enter their credit card on your site or app, they complete the entire referral loop without involving anyone from your team. Automated software processes reduce steps and friction, increasing the conversion rate of referrals. If you want consistent month-over-month growth from your referral program, you must remove your employees from the process and allow your automated money-printing system to work independently.

4. Something to give away

What can you give away to incentivize people to refer their friends? Dropbox gives away storage space. Airbnb gives away cash. Your VOIP app can give away calling minutes. You want to give away something that costs you much less than your current cost of customer acquisition.

Think about features or other value you can let your customers unlock that cost you pennies per month. Dropbox's model is so brilliant because they give customers something that costs them almost nothing for making a referral. Dropbox makes customers happy by giving them more storage space while gaining new users who may upgrade to a paid plan. There's likely something your customers value that's inexpensive for you to deliver — that's what you need to give away.

Your first inclination may be to give away a free month of your SaaS service, but that's a trap you want to avoid or consider carefully. When you give away a free month, you snowball the complexity of your messaging and implementation instead of keeping it simple.

What if a customer with 30 seats and yearly billing refers five new customers in a month? Would they even care about the free credits they've earned? How would you credit customers of your mobile app that uses in-app purchases when you earn $0.70 for every dollar you charge? If a customer refers 12 friends, do they get to use your app for free for a year? What if they upgrade to a plan with more features later in the year? How do you apply the credits they earned and keep them happy?

There are dozens of other edge cases, so do yourself a favor and think of something more inventive than a month of your service that you can give away — strive for simplicity.

5. A trigger event for the reward

You don't want your customers to feel like they must climb a mountain to earn your reward. Your trigger event is the customer action that triggers the referral bonus. Depending on the cost of the reward you decide to give away, the trigger event could be as simple as a new user signing up for a free account. Some businesses with freemium models aim to fill the top of their funnel and then focus on converting a large percentage of each cohort to a paid plan.

You may want to tie your trigger event to a revenue event trigger so each new customer referred to your product pays for themself — Airbnb uses this strategy to grow its business consistently. Requesting someone to enter a credit card is a high-friction event, but if you offer that referral something to motivate them, you should see enough conversions to make your referral program successful.

Dropbox makes users download their desktop app to unlock referral rewards. The most engaged users of your product may take specific actions within your application after they sign up. You could require that all users complete a few onboarding steps as your trigger event, but be careful. As you add additional steps, your overall conversion rate will decrease because some customers will drop out at every step you add to the process.

Marketing channels for referral programs

You can give more than one option for your customers to refer their friends, but you want to promote referral options based on the strength of interpersonal relationships. Channels should be organized and included or excluded from your list as follows — SMS > direct mail > Whatsapp, Facebook Messenger, or Telegram > email > Facebook, Linkedin, or Twitter shares > link shares.

For example, there's a high probability that the person you send an SMS message to has you saved in their contacts. Most people read every text message, so the chances of getting someone to read your message and take action are much higher with SMS. Similarly, only a tiny percentage of your Facebook friends will see the link you share, and your generic post is impersonal. The chances of someone taking action on a social share are much less than the chances of them converting from a one-on-one message.

Direct mail is a channel, and that's no mistake — it can convert as well or better than SMS messages for the right product. You get a lot of junk mail at home, but you rarely get mail sent to your workplace. NextDoor uses direct mail and social proof messaging to get neighbors to sign up for their service. Twilio sends T-Shirts and other schwag to people who sign up for a developer account. There are now services that send snail mail via an API call, which makes sending postcards and company-branded items to customers as easy as sending an email.

Experiment with different channels when you design your referral program, but always promote the most interpersonal channels first.

Employing powerful physiological tactics

So far, we've discussed what you need to set up your referral program, but that's only half the battle. You must utilize messaging to motivate your customer and the referred person to complete the referral loop. Often you'll need to A/B test the messaging to see what physiological principles work best to inspire everyone to act.

Influence: The Psychology of Persuasion is a must-read book to learn about the physiology behind people's buying decisions. While there are ways to incorporate Commitment/Consistency and Authority principles in your messaging, we will focus on reciprocity, social proof, liking, and scarcity, which referral programs often leverage.

Reciprocity

When you go to your local farmer's market, and a beekeeper named Jasmine is giving out free samples of wildflower honey, you think to yourself, "Why not? It's free, right?" So, you try some, and it's delicious. Jasmine asks if you want to buy a jar, and you think, "I don't need more honey right now, but Jasmine gave me a free sample, which was tasty." So, an hour later, you're driving home in your Toyota Tacoma with some heirloom tomatoes and a jar of wildflower honey. How did this happen? You've fallen prey to the principle of reciprocity — you feel you have to repay Jasmine for her kind gesture.

How do you channel this power within your product? You implement your referral system and messaging, so both parties get something from the relationship. You can replace honey with storage space if you're DropBox and cash if you're Airbnb. The best-converting referral systems give the referer and the person they refer something of value.

Imagine your customer asks their friend to create an account with their referral link because they will get $50. The friend will feel your customer is selfish. Now imagine you flip the script and tell your customer they can give their friend $25 to get $25. In this second case, both the customer and friend get something. It's a win-win referral system. Replace money with storage space, calling minutes, or a month of that killer feature everyone wants, and you have a referral model that scales indefinitely.

Social proof

On your way home, you see a line out the door at that new brunch spot. You make a mental note to save it to Yelp. Later that day, you plan with friends to go there next weekend. Social proof is the herd mentality that you'll try a new product or service because others like it too. Why do all your competitors have the logos of their customers plastered all over their websites? Social proof!

Social proof for your referral system comes in two forms. First, social proof is built into the interaction from the start because your customer refers their friend. Then, you can reinforce social proof by displaying logos and testimonials from your customers at just the right time. You get bonus points if you use a service like Clearbit that lets you personalize a landing page with customer testimonials and logos relevant to the referred person.

Liking

It's next weekend. You're standing in line with your friends for the restaurant because you failed to beat the crowd. Your friends dress like you, talk like you, and work in the pharmaceutical industry just like you. When you asked them if they wanted to join you for brunch last weekend, they agreed because they like you. As you reach the front of the line, the hostess says, "I like your jacket." You blush and say, "Thank you!"

By hanging out with like-minded people and feeling good about the hostess' compliment, you're a walking demonstration of the liking principle — you're more likely to comply with requests from people you know and like. Liking has more facets than these examples, but you can read about them in Influence.

Because of the liking principle, you don't design your referral program like Linkedin's early invite system — spamming everyone in your contacts list. Instead, you focus on getting referrals from people who know and like your customers. In your referral program messaging, you utilize a friendly tone that feels like someone the referred person knows. Maybe you throw in a compliment too!

Scarcity

As you and your friends settle at your table, the waitress says, "We only have three orders left of the Millionaire's Bacon sample platter, which is great for groups. If you want to try some for the table, you should place your order with me now because it's going fast." Looking at the menu, you think, "Millionaire's Bacon? What is that? $24 for bacon? It must be good because they're almost out, and it's only 11:37 AM, right? Or covered in gold leaf?"

You've just experienced the scarcity principle in full effect, or FOMO as you may know it. You fear missing out on "experiencing" millionaire's bacon more than you desire to gain something from buying overpriced bacon.

Some of the most successful software product launches have used scarcity to generate interest in their products. Gmail gave every user a few invites and two additional invites for each referral who accepted. Mailbox app created a virtual waitlist for their Gmail competitor, and you could move closer to the front of the line by inviting your friends. You can take ideas from these examples or figure out how to make some aspect of your referral system scarce.

Summary

The system you design can take queues from some examples here, or feel free to go out and create your path. Whether you're defining the basics of your program or picking the physiological principle you plan to employ, you now have the tools you need to create a successful program that converts referrals into new customers.

You have intuition about the messaging that will work, but prospective customers will tell you what “they think they want.” It’s your job to recognize patterns from this qualitative feedback and develop micro-campaigns that test your hypothesis. The fastest (and cheapest) way to quickly validate each hypothesis is via cold outreach, which your sales team does daily.

I say cold outreach because well-run sales teams don’t just make phone calls. Modern sales organizations use sequencers like Outreach, Apollo, or Salesloft to create multi-touch campaigns to target prospect lists. Some sequences have 10 to 20+ steps per prospect, including multiple channels like email, LinkedIn InMails, connection requests, phone calls, and SMS.

When your sales team gets a “no” from a prospect, they often get a “hell no.” But that’s OK. They also get some “maybe” and “yes” responses as you help them test different messaging. So, why should you, the product marketer, work closely with sales? That’s simple — it’s the quickest way to run marketing experiments.

In a matter of minutes, you can create a list of 100 United States-based CMOs who work at companies with 100-500 employees in the retail industry. Why only 100? That’s all you should need to test your idea. Your sales team should get some meetings with the right pitch, and with the wrong messaging, they'll get crickets.

Good sequences get around a 5% response rate, and great ones achieve a response rate of 20% or more.

Your job is to listen to recorded calls and demos and look at email chains for insights. When customers ask questions, what are they saying? Do customers say they want some feature your product team says will be released in six months? How big would the deal be if you had that feature today? How much pipeline? What can you show customers to close the deal?

Revenue and signed contracts change product roadmap priorities. Help your sales team land a deal, and that integration with Figma will get moved up to the next sprint.

Sell the vision, not today’s functionality

There are a thousand things your product “could” do. What 1-3 features have customers repeatedly requested? If your product did those things today, would your prospective customers sign a contract?

It’s your job to push your messaging ahead of your product. Don’t promise a HIPAA-compliant video chat feature when you’re an agile project management tool. But do promise that feature that lets people add new tasks directly from their Google Workspace account (if it’s a $20K+ deal).

As a product marketer, you should know the complexity of each customer’s feature request. Entertain features that take weeks — not months, to implement.

In an ideal world, every customer loves what your product does today. In the real world, you're always chasing that sweet spot where your marketing message and product offering perfectly align with your customers’ wants, needs, and desires.

So, when searching for those first customers and that repeatable customer acquisition process, you need to give your sales team the tools to “sell the dream.”

How to sell the vision

In the product marketing department, you have resources. Can you grab a few hours from a designer to mock up a few screens? Can you publish a landing page describing that requested feature? Or perhaps you create a video showing a "user" clicking through a Figma prototype?

You can use many sleight-of-hand tricks to show that killer feature everyone wants. Some tech-savvy customers will see what you’re doing and walk, but others may not care. They may sign a contract if they’ve seen your product demo and like your feature prototype. If you’ve established trust, that may be all you need.

Selling the vision is about showing just enough to get your customer to sign on the dotted line.

How this all accelerates your marketing campaigns

Working closely with sales, you understand the pain points that resonate with buyers. You could start a campaign and buy ads on Google, LinkedIn, and other channels, trying to achieve the same results. But you'd be lighting a few thousand dollars on fire before seeing results. Your boss will appreciate your ingenuity when you integrate with the sales team to learn what works quickly.

Now that you’ve identified messaging and target personas that work. You can invest in ad campaigns on various marketing channels. Instead of burning piles of cash, you turn your marketing budget into annual recurring revenue (ARR).

This method lets you learn quickly and build a repeatable customer acquisition strategy.

If you’re like most startups, you’ve spent most of your seed capital building your digital product, leaving little left for sales and marketing. You can tap into your network for early adopters, but your warm introductions and leads will soon dry up.

Why your customers won’t come

Remember that scene from your favorite movie. After spending a third of the film building up the founder’s story and “great idea,” the movie cuts to 37 seconds of B-roll, showing how hard the founder worked to grow their business. Then, magically “success.” Don’t fall for this lie.

What you see on TV and read about in the news isn’t reality. Many startups grind it out for months or years, looking everywhere for their customers. Slack “ate their own dog food” and worked out the kinks with early adopters before their big press push that led to a hockey-stick growth trajectory. Countless other startups you’ve never heard of have vanished — CrunchBase their only trace — as their founders failed to find customers.

Finding customers is hard because no one wants to be your first customer. Sure, your Uncle Leo may have bought a subscription to your SaaS product for scheduling meetings, but he runs a pet store. How many meetings could he have? What you need are referenceable customers. Specifically, you want people with budget authority who trust you and are willing to take a leap of faith.

How to find your customers

If your first test is shipping, you passed. But you may fail your second test if you start talking to customers when your product is live. You must speak with potential customers during the entire product development process. Creating your product in a vacuum will never replicate the success of products like Slack.

You must create a repeatable marketing process to reach your next company milestone. Customer feedback helps you find your repeatable process faster.

Repeatable doesn’t mean sending an email blast to 10,000 sales VPs for your revenue growth product that integrates with Salesforce. And it doesn’t mean convincing TechCrunch to write about your startup, either. A repeatable process targets a niche big enough to drive consistent customer revenue growth for the next 6-12 months — ideally much longer.

Think about your niche-like search filters on your Linkedin Sales Navigator account. Yes, you need to pay for Linkedin if you’re serious about growing your customer base. As you add search filters, you narrow in on a niche and your Ideal Customer Profile (ICP) — the people who will buy your Software-as-a-Service or marketplace product.

You may think you know your ICP and the messaging that will drive them into your funnel, and you might be right. But, chances are, only some of your assumptions hold.

Test. Learn. Repeat. The process is: no, no, no, no, no, no, no, no, maybe, no, no, maybe, yes, maybe, yes, yes, no, maybe, yes. Sometimes that chain of noes is one or two hundred long before you get that first “maybe.”

To get to that first “maybe,” you must start small and focused. You’ve thought of 14 marketing and sales channels you can use to reach your ICP. Great! Don’t try everything all at once. Try two channels at most, but it’s better to start with just one. Stack rank the list and think like your customer.

Most decision-makers only respond to three messages: reducing cost, increasing revenue, and eliminating risk. Get as creative as you like with your messaging, but ensure your prospective customers can easily understand how your product solves one of these three problems.

Over this iterative process, you run dozens of tiny experiments until you find yourself saying, “hmmm… that’s interesting.” Hopefully, not too long after, you’ll reach an AHA moment — then continue iterating until your quantity and quality of customer conversations are predictable.

Frequent conversations with people interested in your product lead you to your customers. Congratulations! You’ve found them.

As you started your career in Silicon Valley, you played with every AI tool published on TechCrunch. You put SIRI through its paces before and after the Apple acquisition. You scheduled meetings with x.ai when that was the cool thing to do. You even placed a hockey puck on your desk when Alexa was all the rage. But nothing even remotely lived up to your expectations. At some point, you thought, “all AI is vaporware, and I’ll never live to see the day.”

Then, a few months ago, you tried ChatGPT for the first time. And everything changed after you entered your first query.

You: Which cryptocurrencies are not scams?

ChatGPT: It is difficult to definitively say which cryptocurrencies are scams, as the crypto space is still largely unregulated, and the definition of a "scam" can vary. However, many well-established and widely-used cryptocurrencies, such as Bitcoin, Ethereum, and Litecoin, have a strong track record and are generally considered to be legitimate.

You fall out of your chair. This response must be a fluke. You get up, adjusting your standing desk so as not to repeat yourself, and continue to enter twenty additional questions. With each question progressively harder, ChatGPT nails them all — or appeared to, at least. As you continue testing ChatGPT’s knowledgebase, you realize generative AI isn’t a better Google. No, it’s much more revolutionary than that — it’s Jarvis. It’s what you’ve been waiting for all these years.

By this time in your life, your dreams of becoming the world's first trillionaire are long gone, but a tool like this could be invaluable to your company. As a VP, you have a budget for special projects, and your marching orders are to make your company more efficient. So, why not make a ChatGPT-powered app for your company?

OpenAI — the company behind ChatGPT — has APIs. So, how hard could it be? Right?

It's clear now that this is the only path forward for your company. If you don't adopt ChatGPT now, your competitors will. You're determined to keep your organization from becoming a relic of the past like those companies that used to make typewriters. International typewriter? No. That's not it, but it's not important. You've made up your mind.

You carve out the budget and hire a team to get to work, integrating all your legacy software systems, databases, and SaaS tools — the easy part :) — with your shiny new AI system. You call it "Jarvis,” of course. After reading every email, attending every recorded call, and memorizing every CRM data point, Jarvis is ready for your first query. Finally, you can discover insights that will change the trajectory of your company. You’ve created the first dimension of Jarvis (1DJ).

1DJ — Query and Response

To test your newly commissioned system, you decide to find out why some of your retail stores perform worse than others.

You: Which region has the worst-performing stores by revenue?

Jarvis: The Southeast.

You: What are the three worst-performing stores there?

Jarvis: Birmingham, Chattanooga, and Tallahassee.

You: What common theme leads to the poor performance of these stores?

Jarvis: They’re all located in the Southeast.

You: Jarvis, your answer doesn't help!

Though the initial results are inconsistent, you spend the afternoon drinking chai lattes at your desk, set to 46 inches, so you don't repeat your first encounter with ChatGPT. But, you soon realize that even if you were in a chair, you wouldn't fall out of it. No. Your first version of Jarvis is cool, but it doesn't deliver the revolutionary insights you thought it would from day one. You must know what questions to ask and how and when to ask them.

What about the questions you don’t know you should be asking? What if you ask your question the wrong way? What if you ask the right question a week after your retail sales team misses their target for the quarter?

What you’ve created is helpful but reactionary. To deliver on your vision for Jarvis, you need an active system that’s always asking the right questions in the right way at the right time, sharing the answers with the right people. It’s time to start work on the next version. Enter the second dimension of Jarvis (2DJ).

2DJ — Active Learning, Listening, and Questioning

You design this Jarvis version to tell you what you need to know the instant you need to know it. Not just you — your entire company, from retail employees to your CEO.

Despite its shortcomings, your employees heavily use Jarvis, which gives thousands of questions to which your employees want answers. You enrich the question set by asking ChatGPT to craft 20 questions employees in each department and role would find helpful. Once done, you've identified thousands of questions loosely mapped to each employee's job function and seniority level. These are the questions every employee needs the answers to — Jarvis delivers them.

You roll out your new and improved Jarvis as a service that sends three questions and their answers to every employee once per weekday. Jarvis personalizes questions and answers for each employee, so most colleagues see different information. Each employee upvotes or downvotes questions and answers based on their usefulness. As Jarvis learns, she starts to surface pertinent information to the right people at the right time. And since Jarvis is always listening to the new data within your organization, the answers Jarvis shares change daily.

You dust off your old hockey puck, place it on your desk, and fire it up because your team surprised you by extending this new version of Jarvis to Alexa.

You: What do I need to know?

Jarvis: The retail sales team is on pace to hit 84% of their quota. In the Southeast region, 92% of stores are understaffed 63% of the time. Career fairs and other hiring initiatives need to onboard new retail employees faster. The store-within-a-store partnership deal with Nike is in the final stages of negotiations, but there are disagreements on several contract terms.

You: Don’t bring me problems. Bring me solutions.

Jarvis: You’ve only trained me to share summaries of the information I know.

Better than any insights engine you've used, you're impressed, but you still expect more. You want Jarvis to be an extension of yourself, thinking through problems and delivering possible paths forward. There's no doubt in your mind that this version of Jarvis will create tremendous organizational efficiency, but Jarvis still needs a spark. You yearn for Jarvis to do more.

So, you set to work on a version of Jarvis that delivers solutions to problems before they become critical — the Jarvis you always wanted. You're creating the third dimension of Jarvis (3DJ).

Why dimensions and what’s next

In physics, dimensions describe the characteristics of physical objects or systems that allow us to quantify their properties and predict their behavior. While the dimensions of Jarvis don’t relate to physics, they help us describe how generative AI systems in the workplace will evolve. Each dimension progressively adds complexity to the system while bringing more utility to businesses, employees, and customers who use Jarvis.

In Part 2, you’ll continue your journey as we walk you through dimensions three through six.

Sonos has implemented software that supports this exact scenario. In fact, I recently had an experience similar to the one above. But, Sonos uses human agents only available during regular business hours. This process works for Sonos because of their high average cart size and low number of SKUs.

Now, let's go through a similar journey at Home Depot. You just finished renovating your kitchen, complete with new zellige backsplash tile. The portrait you commissioned of your dog, Hudson, fits perfectly in your new coffee bar nook, which has six feet of vertical space. But you need to figure out how to hang the painting of your best buddy without damaging your beautiful new backsplash. So, you visit Home Depot's website to do some research.

Like Sonos' site, you notice a floating image to contact support. You click the button to connect with a live agent and join an instant video call with Linda. You show Linda your space and explain the problem. Linda recommends purchasing some Command Hooks, which are picture hangers with a strong adhesive that can stick to your new backsplash without damaging it. You end the call and order some Command Hooks.

Because Home Depot carries thousands of items that range in price point from screws that cost a few cents to appliances that cost thousands of dollars, they can't provide the same experience as Sonos. Today, you can get this sort of advice in-store, but it's cost prohibitive for Home Depot to provide tailored online support. But customers would love customer service like this scenario, so how can Home Depot deliver?

Home Depot needs to leverage a generative AI software product capable of providing tailored customer support.

ChatGPT is one of several generative AI building blocks that will transform customer and employee experiences across industries, job functions, and tasks. ChatGPT alone can't deliver the Home Depot customer experience described above, but it's one of the building blocks for an automated human-like experience.

In this blog series, we dive deeper into different generative AI technologies and how they will be applied to solve business problems over the next decade. Yes, we'll see hundreds of innovative tools released this year using APIs from OpenAI and other tools. But, it'll take years to develop an automated experience like the Home Depot example above. Together, we'll explore the problems generative AI will tackle in the coming years.

This blog is for founders and marketers looking to find their first customers

More specifically, this blog is for founders and marketers like you who work at SaaS and marketplace companies — and together, we’ll find customers for your new product or product offering. When you’re finding customers for something new, the rules are different. You don’t have historical sales scripts and marketing copy. You don’t have data on successful campaigns and failed ones. All you have is your intuition and usually a small budget to find and close your first few customers.

This blog is about economically finding those first customers with scientific precision. Since you don't work at Amazon (or another large tech company), you can't afford to run a seven-figure launch campaign on seven channels. Instead, you think like an astrophysicist running experiments in your head before you spend a single dollar.

Companies with SaaS and Marketplace business models generally have a clear path to profitability, generating revenue with each new customer. Salesforce, Slack, and Zoom invented new ways to find customers and grow their businesses. In future posts, we’ll dissect strategies used by these well-known companies and other startups to find and grow their early adopters.

Who shouldn’t read this blog

This blog is not for eCommerce and consumer software startups. Plenty of other blogs give step-by-step advice on optimizing your marketing funnel to boost your user growth metrics or sell physical goods. Plus, when you create a consumer product, you’re in the “hits business” — the odds of making the next Facebook, TikTok, or Instagram are never in your favor.

When you’re building TikTok for cats, Pinterest for fashion, or “the next popular consumer software product” for X, your chance of failure is in the high 90th percentile. Most consumer software products either fail to gain much traction or don’t grow beyond a few thousand, hundred thousand, or million early adopters.

Even if your WhatsApp for cooking app grows to three million active users, you must continue on a hockey-stick growth trajectory. If you have a few bad quarters and your user growth stalls, your ability to raise the next funding round to keep your engineers caffeinated and your cloud servers powered on dramatically diminishes. Path and WANELO are two examples of the thousands of funded consumer startups that failed to gain widespread adoption, so they fold or run on fumes.

Whether you’re selling a new consumer device or trying to build the next Warby Parker for pillowcases, what you’re doing is cool, and you deserve great advice to grow your startup. But you should seek advice from someone else who is an eCommerce marketing expert.

Move fast and be frugal

Your competitors bid up the keywords you want to target, making your customer acquisition cost unprofitable. You need to find new channels and run smarter campaigns than your competitors. Together, we'll work through ways you can run quick scientific experiments that help you find your customers quickly. It's going to be a fun journey, finding your customers together!

Subscribe now