Product & Design — Amy Bonsall

Republished with permission from the FrUn SMB AI Guide. Canonical chapter: Superhuman Docs → · All chapters

Amy Bonsall · Founder, Light Actions · LinkedIn

1. Mindset — What an SMB Should Believe Before Spending a Dollar on AI

Don’t start with AI. Start with your customer or your colleagues. What is their world like? What can we do to make their lives easier?

Then add AI: What can we do to answer those questions today that we couldn’t do a year ago — or three weeks ago?

The trap: If you don’t have a good process for product management, product evolution, and new product development, AI will cause you problems. You need human discernment to understand what people need and how they behave. AI comes after that.

For every AI initiative, ask at the outset: Where do people need to lead, and what human touchpoints should we build in?

AI is lifting the floor on a lot of things — coding, design, prototyping. But the floor isn’t the ceiling. Everything AI touches benefits from expert judgment on top of it.

2. The SMB Tool Stack — Start Cheap, Stay Simple

BASIC STACK

A general-purpose AI tool (pick a paid tier, either individual or team). Your thinking partner and builder. Use it to analyze customer feedback, synthesize patterns, draft product briefs, pressure-test assumptions, and generate prototype concepts. Claude, ChatGPT, Gemini — pick one you like or have access to and use it consistently. I prefer Claude right now, but the tools are changing so quickly. The tool matters less than the habit.

Vapi (free tier to start). An AI voice tool for having conversations with your customers at a scale and frequency you couldn’t manage before. Set up a basic discussion guide, offer it as a post-purchase opt-in, and start hearing things surveys never surface.

GitHub (free tier). Smart version management for everything from prototypes to your website. A little geeky, but someone tech-savvy on your team can own it once setup is done.

SKIP FOR NOW

There’s a whole category of dedicated product management platforms built to organize backlogs, track feature adoption, and align teams around roadmaps. They’re well-designed — for product teams that already have a defined process and enough users to generate meaningful data. If you’re a small team with one person wearing the product hat, that’s not your starting point. Get the discipline first, then add the platform.

ZERO BUDGET STARTING POINT

A free AI tool and a simple process for listening to customers and logging their needs. Paste ten customer emails or reviews into Claude or ChatGPT and ask: “What are the top three things these customers are struggling with, and what would help them most?” That’s a product insight session.

3. The First 30 Days — Where to Start

Your job in product and design is to keep serving customers as their lives change. We’re in a period of accelerated change — what’s possible, how we live and work, is shifting fast. Your products and services need to keep pace.

Start with purpose.

Step 1: Establish your why (human discernment)

The best companies have purposes specific enough to guide decisions but broad enough to grow into. Google’s “organize the world’s information” is as relevant in 2026 as in 1998. Write yours. One sentence. It should stand the test of time.

Step 2: Build a process for hearing from customers regularly (automation)

This is the most important continuous habit in product development.

• What do you want to know?
• List your questions about customers’ experience and needs — a mix of what they’re trying to accomplish (independent of your product) and how your product is serving them. What are they trying to do? What alternatives have they tried? What’s working? What’s not?

How do you get continuous input?

B2C: Set up an AI voice tool like Vapi with a basic discussion guide. Offer it as a post-purchase opt-in.

B2B: Same, at different stages of the client journey. But the fastest win: record client conversations with AI and mine them monthly for opportunities.

• What are you learning?
• Synthesize into a running list of customer needs. Once a month, paste your recent conversations into an AI tool and ask: “What patterns do you see about unmet needs?” Track what comes back.

Step 3: Connect that learning to a tracking tool (automation)

GitHub, Asana, a spreadsheet — whatever you already use. The point is that what you hear from customers flows into what you’re building. Close the loop.

Step 4: Hold regular prioritization meetings (human discernment)

Monthly or quarterly. The input is steps 2 and 3. The decision is your team’s.

Step 5: Build toward the need (AI + humans)

Describe what you want to build in an AI tool, along with the customer need it serves. Ask for a looks-like, feels-like prototype, or even a simple functional prototype (if you feel more confident building in AI). Test with real customers. Iterate. Then take it into production.

Operationalizing this

Week 1: Review your purpose. Ask: Does it guide decisions? Does it hold up in the AI era? Is it still inspiring? Adjust if needed. Some favorites for inspiration are here.

Weeks 2–4:

Define what you want to learn from customers and which customers to hear from.

Set up an AI voice tool to start collecting input conversationally.

Test it with a handful of customers, refine, and decide what comes next.

• What good looks like at 30 days:
• You’ve got a purpose that still holds and a live process for hearing from customers. That’s the foundation everything else builds on. Anything you do from here will be focused — because you know why you’re doing it and who you’re serving.

4. A Real SMB Use Case

A trade publishing company — established, respected — had spent decades building deep expertise in a specialized professional field. They had archives of authoritative content, a loyal subscriber base of working practitioners, and a sense that there was something new to build with it.

But the onus was on the customer to find the information they needed at the right time. Readers had to search, browse, and synthesize on their own. The team knew their audience had specific, practical questions that their content could answer — but there was no interactive way to surface it. The knowledge was there. The access wasn’t.

Wanting to make things easier for the customers, we prototyped an AI-powered tool that draws on the company’s published content and answers practitioner questions in a tone calibrated for their audience. We considered sourcing (published editorial only), response format (conversational answers the way an expert would give them), and interaction design (starter questions written in the actual voice of their readers).

It’s still in development, but the key elements for the company (with my support) are deciding what types of data they want to make available, how they want to ensure it’s accurate and current, and what business model makes the most sense for this opportunity.

For the prototype:

Tools used: Claude to create the prototype with an API to do database search and craft the responses, Netlify for deployment.

Cost: A few hours of build time. Hosting costs are minimal.

Result:

The prototype gave the team something to react to rather than imagine. It moved the conversation from “could we do something like this?” to “here’s what it feels like — now what do we want to do with it?” That shift — from abstract to tangible — is often the most valuable thing product design can provide. It’s also one place where AI has changed the game: a prototype that once would have taken a team weeks now takes a person hours.

5. Guardrails — What an SMB Should NOT Do

Don’t ship AI-generated product decisions without a human review.

AI is exceptional at generating options and surfacing patterns. It is not good at knowing which option is right for your specific customer, your specific market, your specific moment. Nor can it create production-ready code at this stage.

Don’t skip the brief.

AI can generate a lot, quickly. That speed becomes a liability if you haven’t defined what you’re building and why before you start. A two-paragraph brief — who this is for, what problem it solves, what success looks like — is the difference between AI as a productive collaborator and AI as a noise machine. Write the brief first. And importantly, decide what the people own and what AI owns.

Don’t assume AI understands your customers as well as you do.

You have years of context. And you can spot patterns that AI can’t. Ask it to synthesize client needs, but look at the raw input and draw your own conclusions.

Don’t put sensitive data into AI.

Be careful about what customer data you put into AI tools, especially identifying information. Understand whether the tool you’re using trains on your inputs. For sensitive client conversations, use a tool with a clear data policy, and always anonymize data before uploading it.

6. Lessons Learned + One Thing You’d Tell Every SMB Owner

Forget about the tool, focus on what you’re trying to accomplish.

I’ve been in so many conversations about what are the best tools. Honestly, the answer tomorrow could be different than today’s. Instead, and this is good practice anyway, always get clear on what you want to achieve. Then start with a general purpose GPT (I prefer Claude, but you could use ChatGPT, Gemini, or even CoPilot) and ask it: “I want to accomplish X, help me decide how to do this.”

Learn to use a trusted data repository.

To the point about tools changing, you may love Claude today and something else tomorrow. Now is the time to establish a habit of storing all the things you build with AI in a trackable repository. I like GitHub for this. How I use it: Say I make a prototype of a product, and share it with a few customers. They give feedback, and I iterate it. If I stored that original prototype code in a GitHub repository (repo), then I can update it there. I can always go back to the original version later, though, if I realize we lost something important. All this is critical. But importantly, it sits outside of any one AI tool, so I can swap and carry on iterating in any AI tool I want, no lost data, no friction.

Just because you can, doesn’t mean you should.

It’s so easy to create prototypes these days… you could create dozens in a day. That’s where it’s more important than ever to have a process for collecting, reflecting on, and deciding what to do about customer experiences and needs. AI is not a replacement for human discernment. Don’t equate volume or speed with quality.

Remember: you are the expert your customers need

At the end of the day, AI doesn’t know what your customers need. You do. Use AI to hear from them more often, synthesize what they’re telling you more quickly, and build your response faster than you could before. The human judgment at the center of good product design has never been more valuable. Because AI makes it so easy to skip it and miss the mark.

The one thing:

Product and design has always been about one question: How might we better support our clients/customers? Keep this in mind at every step of the process, and you’ll be on the right track.

Part of the FrUn SMB AI Guide. Read on canonical guide →