Procuring Technology

Build or buy in the age of GenAI?

Our first post in AI for SMBs was nearly a year ago. From the start, the goal has been to help you understand how to do things with generative AI. The tools themselves were hard and clumsy to use - counterintuitive and strange to us humans. We focused on helping you overcome those challenges. Figuring out how to do the things you’ve done for a long time, but faster, better, and/or cheaper with generative AI.

Early on in AI for SMBs, we included massive sample prompts. Long-time readers will have noticed there are far fewer lately. Generative AI models no longer need special incantations to unlock great outcomes; just be verbose, specific, and clear.

If we wrote a step-by-step guide today, it would be wildly out of date within a couple of weeks. The interfaces and capabilities of the LLMs are shifting that fast.

We’re trying to be an example of what we think will be true for all businesses moving forward: by the end of 2026, every business will need to remix itself if it wants to thrive in the new paradigm.

So it’s time to remix this blog yet again; just a few months after the last change. Something tells me this won’t be our last remix this year.

As AI for SMBs evolves, we’re going to focus on mindsets, approaches, and attitudes. Especially for leaders and executives in small and medium-sized businesses. You’ll see us publishing a bit less frequently.

This week’s post is an entirely new format for us: a lightly edited transcript of a conversation Jason and I had about how small and medium-sized businesses should think about building vs buying technology right now. It’s in a new-ish format we both love in which two commentators at a large publication share their own internal chats. Enjoy!

It's the start of a new calendar year. Many teams' budgets have been refreshed. Given recent developments in generative AI, we're seeing a lot of organizations increase their technology budgets.

Exciting! But also precarious. If you’re procuring technology today the way you bought it two years ago, you’re setting yourself up for problems later this year+.

We hope this week’s conversation gets you thinking about how to bring technology into your business in a way that meets your needs today without limiting your ability to take advantage of what’s coming this year and beyond.
💬
Justin Massa: You’ve bought a lot of technology in your life. How much, roughly?

Jason Rubinstein: I’d guess I’ve directly or indirectly influenced tens of millions of dollars worth of software and services purchased over my career. Which is both exciting and frightening at the same time.

Justin: For most of our careers, the model was: commit a bunch of money upfront to buy the right thing, implement it correctly, recoup that investment over multiple years. What was the shortest ROI time horizon you’d normally think about?

Jason: It’s not simple. Typically, you joined a company where there was already a foundation: your email, your CRM, your messaging apps. But based on what your business does, you’d have specialty applications you either built yourself, or bought. That build-versus-buy decision becomes critical in this new AI era.

For that base layer of productivity, I didn’t really see CFOs or CIOs heavily measuring ROI (beyond cost) on core applications. What I always saw, both as a buyer and a builder/seller of software, was having to justify why you wanted something new or different above the baseline. What value would it add? That’s what made the business case. And those ROIs were typically measured in years.

Justin: I remember when I was at IDEO and we changed our CRM. We agonized over it. Found the right partner, the right implementer, finally pulled the trigger. We thought we’d be ahead on the investment in less than two years. Too much money upfront, too much training, too much transition complexity. And just figuring out what we needed and who would implement it sometimes took six months. So what’s different now?

Jason: Two pieces. First, how GenAI software is actually being implemented today. Second, the software we can build using GenAI ourselves.

On the first – for those of us in Microsoft or Google shops, it was an easy yes for Copilot or Gemini as an add-on. That was really the beginning of everyone going, “Oh, I got this thing, for free or at a discount, that just showed up in my core office suite. Maybe I got an email about it.” But nobody really knew how to use it. We were reading more headlines about kids cheating in school than AI business transformation case studies.

Let’s peg that moment at 2-3 years ago. This massively powerful capability just sort of turned up one day on the desktop, but we had no idea about it. It’s just there. It’s “free.” It plugs into my company’s tech stack. We became desensitized.

Justin: And now?

Jason: Now we’re at a point where, if we have the guts, we can start to replace the foundational layer software ourselves. The way we want it to look and be. No extra features!

A few weeks ago, you texted me on a Sunday night saying you’d canceled our CRM subscription because you started building a new one. I replied, “We should have talked about that first!” A couple of weeks later, I’m all in. I’ve built out a bunch of additional capabilities on my own. Between the two of us, we’re saving over a hundred hours, and $170, a month.

Justin: What I didn’t think about until this conversation: while the cost and complexity of buying technology is coming down, the training isn’t. I don’t even mean training on what buttons to click. The mindsets, attitudes, and growth orientation to leverage generative AI effectively matter even more than with prior waves of technology. The complexity of how it integrates into day-to-day work is racing ahead.

Jason: Right. And this is where all companies are not created equal. Who’s best positioned to roll up their sleeves and rebuild their friendly neighborhood CRM application? Small businesses. They don’t have a CIO who would say no way. Small businesses say yes to themselves, run experiments in parallel, and see how it works. Fortune 500s won’t be ripping out major SaaS apps like Salesforce anytime soon. But small companies can punch way above their weight class by building what they need instead of being forced to accept the small business package with fewer features.

Justin: And here’s what’s wild: the very process of building software for your business teaches you how to use the technology. Double payoff. We spent quite a bit of time and money on Claude in January. If the only benefit were replacing some systems, it would have been worth it. But we also got deeply comfortable using these tools. That’s the double bottom line.

Even if we toss out what we built a month from now and go back to some SaaS application, none of this would be wasted. We’d make a far better buying decision because we’d deeply understand how hard and how valuable every single feature is. We built them ourselves.

Jason: Picking up on that: the big enterprise apps are fighting back (“supporting end user needs”) by putting in prompts. I was messing around with Figma’s then-beta AI features last fall, looking for buttons or menus to execute some basic operations…then I realized… there are no buttons or menus. Just a prompt - like the beginning of Zork. I told it what I wanted and sat amazed as a coding window immediately opened and started writing software in real-time.

That’s how the incumbents are responding. Hundreds of millions of people a day are getting used to a prompt - let’s make that the UI and shortcut building out new features and extend our SaaS application. And pray, because we’re stuck between two or three business models!

We’ve seen the same thing with a recent client. In their industry, there’s a major AI-native software application that’s in headlines. We asked them how they handle custom feature requests. They spun up a demo and said, “We put a custom workflow builder in. As the end user, you can build your own workflow in real time.” The workflow builder is a prompt. And the software company doesn’t charge extra for it.

Justin: Which brings up something everyone procuring technology needs to investigate: that unbelievably cool feature you see in a new enterprise SaaS product, how hard is it actually to pull off yourself? Some of these magic tricks are truly magic because they’re powered by generative AI. With an hour of work, you could do the exact same thing in your own environment.

I think a lot of buyers get excited about magical features without realizing that unless the vendor is making the model itself, they’re just leveraging off-the-shelf capabilities and baking them into their product. Try replicating it in ChatGPT or Claude before you pay someone else. Sometimes you’ll realize it’s genuinely hard. Sometimes you’ll be dumbfounded that you did it yourself in fifteen minutes.

Jason: On the flip side, it’s important to recognize where we are. Building software this way is still a solitary experience and generally for the benefit of one user. None of the major LLMs have talked about or rolled out what a “vibe-coded enterprise software deployment” looks like for teams. But that’s where small companies have an advantage. They can mess around, configure things in different ways, and figure it out. That’s the next adventure. And it’s what the incumbent SaaS providers need to be very wary of.

Justin: Okay, some rules of the road for people procuring technology right now. I’ll start. One: you don’t want a long-term contract. If it’s two dollars more per user for month-to-month, do month-to-month. You will want to switch.

Jason: Two: if you’re willing to try recreating something, don’t turn off your existing system first! Run the new thing as a parallel experiment until you’re sure it works and can be sustained. Check all of your key boxes before you do something dramatic like shut down working software (Justin!). ;-)

Justin: Three: data portability matters. These enterprise apps will get jealous about guarding your data. Ask direct questions (prompts!) about export, what you can save, what you can move. The capability jumps we’re seeing mean making a change could be an exponential improvement. You don’t want to be locked out of that.

Jason: Four: understand the real roadmap of what you want your new or replacement app to do versus what the LLM or coding tool you’re using is capable of. For example, you couldn’t have an agent log into paywalled apps as “you” a few months ago. Now Claude has a beta version with Chrome that is enabling that. Knowing what’s actually available today versus hoping for what will be possible soon is critical so you don’t have higher expectations than reality.

Justin: Five: if you rely on proprietary data feeds, start having conversations with those vendors about their AI strategy. You don’t want to buy something that locks you into their bolt-on AI tool. You want the option to pull that data into your own environment and use it across multiple AI tools.

Jason: Six: if you’re building a one-for-one swap of something you currently pay for, your ROI window should be three months or less. If you can only build part of what you need, six months. But even then, you’re getting a head start. You’ll be ready before anyone else to make the leap when full replacement becomes possible.

Justin: Last thought. When you buy big enterprise software and everything lives in the cloud, the only mechanic you get to call is the dealer. If something goes wrong in your OpenAI web account, you deal with OpenAI. No choice.

But if you’re just buying tokens of intelligence, which are really a commodity, you’re not calling Shell Oil when you need your car fixed. You can take it to any local mechanic. If you’re building your own apps, you’re in control. Your friendly neighborhood software engineer becomes your mechanic. You don’t need big implementers. That radically expands, and lowers the cost of, who will help you.

Jason: And your implementer (we call them AI Business Architects) doesn’t have to be an engineer. It can be someone who’s a non-engineer, but technical enough to understand how these things work together. Like you or me.

Justin: For the first time in a long time, technology has become an advantage for SMBs over big enterprises. I’m trying to think of other examples in my career and I struggle. Large technology expenditures have almost always started in government, then gone to corporate, then eventually made their way to SMBs and non-profits. Most of the small businesses we work with are now using generative AI in significantly more sophisticated ways than big enterprises.

Jason: It’s the unfair advantage for small, growing businesses. Revenge of the (SMB) Nerds. Return of the Jedi. You get it.

Questions? Email us at info@remixpartners.ai - we read every message.