Make Better Build or Buy Calls in Software Teams

The best build versus buy calls come from identifying where latency is most expensive. Some capabilities are not valuable because they are hard; they are valuable because waiting on someone else creates drag across product, marketing, finance, and leadership. In those cases, the hidden cost is not the invoice; it is the delay between signal and response.

One decision that improved speed dramatically was bringing a core analytics function inside after realizing every outside handoff diluted urgency. I had been paying for competent output, yet the business kept reacting a week late. Once the feedback loop lived with the team closest to revenue, decisions became sharper, faster, and far less political.

My test is whether the capability is something my customers would notice if it were slightly worse. If they'd notice, I build it. If they'd never know who did it, I buy it.

Running a group consulting program with thousands of entrepreneurs in it, the coaching, the curriculum, and the judgment calls people pay for all stay with my team. But scheduling, transcription, editing, and the plumbing around delivery, I've bought every time. When I tried to build that kind of infrastructure internally, I ended up with a half-finished tool and a team distracted from the work students enrolled for.

Buying breaks down when a vendor sits between me and my customer's experience. When a service touches how someone feels about the program, like support responses, onboarding, and anything with my voice on it, outsourcing it costs me more in quality than it saves in hours.

So I ask whether the capability is a differentiator or a dependency. I build differentiators, slowly and expensively, because they compound. I buy dependencies this week so I can get back to the differentiator.

We can almost always build it. Somebody here always can. What decides it is the date.

A hiring push last year needed applicant tracking before interviews started three weeks out, so we bought the dullest tool on the shortlist and moved on. The same month someone here built a small dashboard for our outreach numbers in four days, because nothing was waiting on it. My rule is that anything with a date attached gets bought. Anything we would still want in three years gets built in-house first, badly, and improved later. We are about 60 people with a tech team you could fit in one call, so the cost that matters is whose week it lands on afterwards. That dashboard took four days to build.

When my team needs a new capability, I decide build versus buy by first asking whether the work can be delivered as a small, well-defined task with clear context and decision points; if so, we prototype in-house; otherwise, we buy. The single rule that most improved speed and reduced regret is to only build when you can set tight boundaries and provide the context needed for reliable execution. I used this approach with Claude Code in the development loop by giving it bounded tasks, supplying the relevant context, and keeping human review at key decision points. That practice let us iterate quickly and avoid wasting time on large, uncertain builds.

My rule of thumb is frequency, not cost. If we need the capability weekly, we build it or hire for it. If we need it in bursts, we buy it. Most build versus buy regret I have seen came from teams building something they used twice a quarter.

Our own clearest example is choosing to stay on Webflow rather than building custom front ends. We could build custom. But the thing clients pay us for is turning a request around in hours, and every custom system we own is time we spend maintaining instead of shipping. Buying the platform is what makes the speed possible.

We see the mirror image on the client side. Founders often ask whether to hire an in-house designer or keep an agency on retainer. Same test: if there is a full week of design work every week, hire. If it comes in waves around launches and fundraises, buying that capacity is cheaper and faster than carrying it.

My rule of thumb is to lock down non-negotiables first, then evaluate the trade-offs across total cost of ownership, time-to-value, operational overhead, and vendor lock-in. We focus strictly on the variables that could flip the decision and use a targeted POC to validate our highest-risk assumptions.

When our event-streaming setup hit a scalability wall, we weighed managed services against running stateful Kafka brokers on Kubernetes. A managed service offered faster deployment, but brought platform dependencies and a significant infrastructure premium without the control we required. Rather than assuming running stateful Kafka workloads would create unsustainable overhead, we ran a focused POC to test technical feasibility and evaluate how the setup integrated with our existing operational workflows. It demonstrated that our team could manage the operational load while giving us the control and cost profile we needed.

Build vs. buy isn't about finding a universally correct answer. It's about choosing the option that best fits your constraints.

I run TKEG Expat, a corporate-services firm managing 120 companies across 22 jurisdictions. We decide build versus buy on maintenance surface instead of price. We buy anything the whole market already maintains, and we build anything that encodes our own jurisdiction data.

For example, the Council of the EU's PRADO register catalogues 3,253 identity and travel document descriptions across 198 issuing countries, territories and organisations, which is the maintenance surface any firm takes on if it builds identity checking. Therefore we buy it at roughly a euro and a quarter per verification, on Stripe's published EU rate. Moreover, our AML policy names that verification as outsourcing of a technical function instead of section 40 third-party reliance, so the due diligence responsibility stays with us.

Our jurisdiction dataset carries the corporate income tax return, payment and estimate due dates on all 89 records, and our obligation register hangs 266 rows off 55 named managed companies, each row linked to a company record. No vendor sells that shape off the shelf, so we build it.

However, the decision I regret was a migration we ran ourselves. When we folded a separate accounting system into the main platform in April 2026, every record class moved but the attachments did not. Project invoice files came across 0 of 32, and files that lived only on the old file store were deleted in the move and cannot be recovered.

Since then I treat "we can migrate later" as the most expensive assumption.

Buy by default, that is my rule. Build only when the first option is forcing a compromise on something your customers judge you on.

My company's website was built on Webflow by a web agency. It worked for us but it was not super fast and not well optimized. We are a web hosting company and we sell speed. Having a slow website or slower than needed was not an inconvenience, it was a credibility problem. That is what tipped it. If this was for any other type of business I would still be paying the subscription. So I made the decision to move the website to Astro on Vercel. It became static HTML which made it quite faster. The obvious cost was that we lost Webflow's CMS capabilities - exactly the thing that made us stay with the platform even though we had outgrown it. The solution was simpler this time, I built one for us. It is AI based, we describe what we want and it produces it with our design, brand voice and philosophy.

A few years ago this decision would have been hard to make. Building a CMS meant hiring developers and project managing, and the maintenance burden afterwards was the reason to buy instead. AI changed that. I am a CTO, not a career developer and I built it in an afternoon.

That is the part I would flag to anyone who has to make this decision now. A lot of these calls were settled years ago based on what building used to cost in time and money. Those assumptions no longer hold. It is worth re-asking yourself the question on things you decided to subscribe to years ago. In particular, whether the rented version is making you compromise on something core to what you sell.