Constructing AI-Powered SaaS Companies – O’Reilly

In preparation for our upcoming Constructing SaaS Companies with AI Superstream, I sat down with occasion chair Jason Gilmore to debate the total lifecycle of an AI-powered SaaS product, from preliminary ideation all the best way to a profitable launch.

Jason Gilmore is CTO of Adalo, a preferred no-code cell app builder. A technologist and software program product chief with over 25 years of business expertise, Jason’s spent 13 years constructing SaaS merchandise at corporations together with Gatherit.co and the extremely profitable Nomorobo and because the CEO of the coding schooling platform Treehouse. He’s additionally a veteran of Xenon Companions, the place he leads technical M&A due diligence, advises their portfolio of SaaS corporations on AI adoption, and beforehand served as CTO of DreamFactory.

Right here’s our interview, edited for readability and size.

Ideation

Michelle Smith: As a SaaS developer, what are the primary steps you are taking when starting the ideation course of for a brand new product?

Jason Gilmore: I all the time begin by discovering a reputation that I really like, shopping for the area, after which making a brand. As soon as I’ve carried out this, I really feel like the concept is turning into actual. This was a torturous course of, however due to AI, my course of is now fairly easy. I generate product names by asking ChatGPT for 10 candidates, refining them till I’ve three most well-liked choices, after which checking availability through Lean Area Search. I normally use ChatGPT to assist with logos, however curiously, whereas I used to be utilizing Cursor, the favored AI-powered coding editor, it robotically created a brand for ContributorIQ because it arrange the touchdown web page. I hadn’t even requested for one, but it surely appeared nice, so I went with it!

As soon as I nail down a reputation and brand, I’ll return to ChatGPT but once more and use it like a rubber duck. After all, I’m not doing any coding or debugging at this level; as an alternative, I’m simply utilizing ChatGPT as a sounding board, asking it to increase upon my thought, poke holes in it, and so forth.

Subsequent, I’ll create a GitHub repository and begin including points (mainly characteristic requests). I’ve used the GitHub kanban board up to now and have additionally been a heavy Trello person at numerous instances. Nonetheless, nowadays I preserve it easy and create GitHub points till I really feel I’ve sufficient to represent an MVP. Then I’ll use the GitHub MCP server along with Claude Code or Cursor to tug and implement these points.

Earlier than committing sources to growth, how do you strategy preliminary validation to make sure the market alternative exists for a brand new SaaS product?

The reply to this query is straightforward. I don’t. If the issue is sufficiently annoying that I ultimately can’t resist constructing one thing to unravel it, then that’s sufficient for me. That stated, as soon as I’ve an MVP, I’ll begin telling everyone I learn about it and actually attempt to decrease the barrier related to getting began.

As an example, if somebody expresses curiosity in utilizing SecurityBot, I’ll proactively volunteer to assist them validate their web site through DNS. If somebody desires to offer ContributorIQ a strive, I’ll ask to fulfill with the individual working due diligence to make sure they’ll efficiently connect with their GitHub group. It’s in these early phases of buyer acquisition you could decide what customers actually need somewhat than merely making an attempt to duplicate what rivals are doing.

Execution, Instruments, and Code

When deciding to construct a brand new SaaS product, what’s probably the most important strategic query you search to reply earlier than writing any code?

Personally, the query I ask myself is whether or not I significantly imagine I’ll use the product on daily basis. If the reply is an adamant sure, then I proceed. If it’s something however a “heck sure,” then I’ve discovered that it’s greatest to sit down on the concept for just a few extra weeks earlier than investing any further time.

Which instruments do you suggest, and why?

I usually use plenty of completely different instruments for constructing software program, together with Cursor and Claude Code for AI-assisted coding and growth, Laravel Forge for deployment, Cloudflare and SecurityBot for safety, and Google Analytics and Search Console for analytics. Take a look at my complete checklist on the finish of this text for extra particulars.

How do you precisely measure the success and adoption of your product? What key metrics (KPIs) do you prioritize monitoring instantly after launch?

One thing I’ve discovered the exhausting approach is that being in such a rush to launch a product signifies that you neglect so as to add an acceptable degree of monitoring. I’m not essentially referring to monitoring within the sense of Sentry or Datadog; somewhat I’m referring to easily understanding when any individual begins a trial.

At a minimal, it is best to add a restricted admin dashboard to your SaaS which shows numerous KPIs resembling who began a trial and when. You must also be capable to rapidly decide when trialers attain a key milestone. As an example, at SecurityBot, that key milestone is connecting their Slack, as a result of as soon as that occurs, trialers will periodically obtain helpful notifications proper within the very place the place they spend a big a part of their day.

On construct versus purchase: What’s your important determination framework for selecting to make use of prebuilt frameworks and third-party platforms?

I feel it’s an amazing mistake to attempt to reinvent the wheel. Frameworks and libraries resembling Ruby on Rails, Laravel, Django, and others are what’s often called “batteries included,” which means they supply the whole lot 99% of what builders require to construct a tremendously helpful, scalable, and maintainable software program product. In case your intention is to construct a profitable SaaS product, then it is best to focus completely on constructing a high quality product and buying clients, interval. The rest is simply enjoying with computer systems. And there’s nothing improper with enjoying with computer systems! It’s my favourite factor to do on this planet. But it surely’s not the identical factor as constructing a software program enterprise.

High quality and Safety

What distinctive safety and high quality assurance (QA) protocols does an clever SaaS product require that a typical, non-AI utility doesn’t?

The 2 most necessary are immediate administration and output monitoring. To attenuate response drift (the LLM’s tendency for inventive, inconsistent interpretation), it is best to rigorously check and tightly outline the LLM immediate. This should be repeatedly examined in opposition to numerous datasets to make sure constant and desired habits.

Builders ought to look past common OpenAI APIs and think about specialised customized fashions (like the two.2 million out there on Hugging Face) which are higher fitted to particular duties.

To make sure high quality and stop hurt, you’ll additionally have to proactively monitor and overview the LLM’s output (notably when it’s low-confidence or probably delicate) and repeatedly refine and tune the immediate. Retaining a human within the loop (HITL) is crucial: At Nomorobo, as an example, we manually reviewed low-confidence robocall categorizations to enhance the mannequin. At Adalo, we’ve reviewed hundreds of app-building immediate responses to make sure desired outcomes.

Critically, companies should transparently talk to customers precisely how their information and mental property are getting used, notably earlier than passing it to a third-party LLM service.

It’s additionally necessary to distinguish when AI is actually vital. Generally, AI can be utilized most successfully to improve non-AI instruments—as an example, utilizing an LLM to generate advanced, difficult-to-write scripts or reviewing schemas for database optimization—somewhat than making an attempt to unravel the core drawback with a big, common mannequin.

Advertising, Launch, and Enterprise Success

What are your prime two methods for launching a product?

For early-stage development, founders ought to focus intently on two core methods: prioritizing Search engine optimisation and proactively selling the product.

I like to recommend prioritizing Search engine optimisation early and aggressively. At present, nearly all of natural visitors nonetheless comes from conventional search outcomes, not AI-generated solutions (GEO). We’re nevertheless actually seeing GEO being attributed to a bigger share of holiday makers. So when you ought to give attention to Google natural visitors, I additionally counsel spending time tuning your advertising and marketing pages for AI crawlers.

Implement a feature-to-landing web page workflow: For SecurityBot, practically all visitors was pushed by making a devoted Search engine optimisation-friendly touchdown web page for each new characteristic. AI instruments like Cursor can automate the creation of those pages, together with producing vital belongings like screenshots and promotional tweets. Touchdown pages for options like Damaged Hyperlink Checker and PageSpeed Insights have been 100% created by Cursor and Sonnet 4.5.

Many technical founders hesitate to advertise their work, however visibility is essential. Overcome founder shyness: Be vocal about your product and get it on the market. Share your product instantly with mates, colleagues, and former clients to begin gaining early traction and suggestions.

Mastering these two methods is greater than sufficient to maintain your workforce busy and successfully drive preliminary development.

On scaling: What’s the only greatest operational hurdle when making an attempt to scale what you are promoting from a handful of customers to a big, paying person base?

I’ve had the chance to see enterprise scaling hurdles firsthand, not solely at Xenon but in addition through the M&A course of, in addition to inside my very own tasks. The most important operational hurdle, by far, is sustaining give attention to buyer acquisition. It’s so tempting to construct “only one extra characteristic” as an alternative of making one other video or writing a weblog publish.

Conversely, for these corporations that do attain a measure of product-market match, my statement is they have a tendency to focus far an excessive amount of on buyer acquisition at the price of buyer retention. There’s an idea in subscription-based companies often called “max MRR,” which identifies the purpose at which what you are promoting will merely cease rising as soon as income misplaced as a consequence of buyer churn reaches an absolute greenback level that erases any income positive factors made by way of buyer acquisition. In brief, at a sure level, it’s essential to give attention to each, and that’s troublesome to do.

We’ll finish with monetization. What’s probably the most profitable and dependable monetization technique you’ve seen for a brand new AI-powered SaaS characteristic? Is it usage-based, feature-gated, or a premium tier?

We’re actually seeing usage-based monetization fashions take off nowadays, and I feel for sure forms of companies, that makes a whole lot of sense. Nonetheless, my recommendation to these making an attempt to construct a brand new SaaS enterprise is to maintain your subscription mannequin as easy and comprehensible as attainable in an effort to maximize buyer acquisition alternatives.

Thanks, Jason.

Appendix: Really helpful Instruments