Custom software with AI vs off-the-shelf SaaS: when each one wins
Count the subscriptions your company pays for this month. The CRM. The invoicing tool. The e-signature one. Email. Projects. Support. Each looked cheap the day you signed up. Together, they're one of your biggest fixed costs, and not one of them does exactly what your business needs.
Here's the question almost no one asks in time: keep stacking off-the-shelf subscriptions, or build the software that actually fits your operation? For years the answer was easy: buy, because building bespoke was expensive and slow. That has changed. This guide explains when each option makes sense, what each one really costs, and how to decide with numbers instead of by inertia.
The problem isn't one tool, it's the sum
SaaS isn't expensive because of one subscription. It's expensive because of the accumulation. A company with fewer than 500 employees runs around 150 different applications on average, and software spend sits near 4,800 dollars per employee a year, up 22% on the year before. The bill grows on its own, subscription by subscription, without anyone making a conscious decision.
And a good chunk of that money is wasted. 51% of the SaaS licences companies buy go unused: the highest waste rate ever recorded. You pay for seats no one fills, for features no one touches and for tools that overlap each other. The spend is there; the value isn't.
In Spain the pattern is the same, with a telling twist: 44% of SMEs use management software and 7 out of 10 already run on a cloud ERP, but only 28% use software specific to their sector. Translated: most buy generic tools and bend their business to fit them, rather than the other way round. That's where the invisible cost lives.
What "buying" SaaS really means
The price on the pricing page is the small part. According to Gartner's 2025 SaaS Economics analysis, integration, training and mandatory customization push the total cost of ownership 150% to 200% above the advertised price. The real cost of a SaaS tool ends up being 2.5 to 4 times the fee you saw when you signed up.
The five-year breakdown confirms it: the licence itself accounts for just 36% of total spend. The remaining 64% goes on implementation, dedicated staff, training and constant adjustments. Buying SaaS isn't "pay and use". It's pay, integrate, train, maintain and, when the tool falls short, rig up a workaround with another tool and another subscription.
On top of that sits what no one measures: the process you bend to fit the software, the data you copy from one app to another because they don't talk to each other, the limit you can't cross because it isn't your product. You rent the tool, but you also rent its restrictions.
What changed: bespoke is no longer expensive or slow
The historical argument against custom development was time and money. AI has moved both. Today, teams developing with AI assistance complete tasks up to 55% faster, according to a GitHub and Accenture study of 4,800 developers. McKinsey puts the cut in time-to-market at 30%. And for bounded initial builds, development cost can fall by up to 80%.
This isn't a promise about the future. AI already generates around 46% of the code written on GitHub, and Gartner expects it to reach 60% of all new code by the end of 2026. The area where it pays off most is exactly what an SME needs: internal apps, back-office tools, portals, automations, integrations between systems. Well-defined problems, with clear rules, where AI genuinely accelerates the work.
The result is that the floor has dropped. What two years ago only made sense for a large company with a six-figure budget is today within reach of a small business that wants software fitting its way of working, not the other way round.
When to buy (SaaS) and when to build
The question isn't ideological. It's a matter of judgement. And it comes down to one thing: whether that software is your edge or a common service.
Buy when it's a standard function
There's ground where reinventing the wheel adds nothing. Email, accounting, payroll, e-signature, video calls. These are identical functions for you and your competition, solved by mature, cheap products. Here buying is the sensible move: you deploy fast, maintain nothing and the per-user cost is justified. Building what already exists and works is burning money.
Build when it's how you operate
The calculation flips when the software touches the core of your business: the process that sets you apart, the flow you repeat a thousand times a month, the integration no off-the-shelf tool covers. There, generic SaaS forces you to choose between adapting or stacking subscriptions that never quite fit. A custom build does exactly what you need, connects your systems and grows with you. And unlike the subscription, it's an asset you own, not a perpetual rental.
The practical rule: buy the common, build what makes you different. Most SMEs do the opposite, and that's why they pay more to do less.
The math: deciding with numbers
The decision fits on a napkin. On one side, the real cost of SaaS: the per-user fee —which can run from tens to several hundred euros a month depending on the tool— multiplied by seats and by months, year after year, plus the hidden implementation and maintenance. Remember the fee is only 36% of what you'll end up paying.
On the other side, the cost of the custom build: a bounded investment up front and far lower maintenance after. Build-vs-buy analyses are consistent: for an operation with stable workflows and around 20 users or more, custom software stops being more expensive at roughly 24 months and is clearly cheaper from the third year on. The subscription rises every year; the custom asset pays itself off and then works for you.
The factor that weighs most shows up on no invoice: the cost of adapting your business to a tool that wasn't built for it. Add that in, and the scale tips sooner. We develop this framework with figures in how much it costs to automate a process with AI.
How to decide without getting it wrong
It's not about picking a side forever, but about ordering the decision piece by piece.
First, take inventory: list every subscription, its real annual cost and what you actually use it for. You'll be surprised how much overlaps and how much goes untouched. Second, separate the common from the proprietary: mark which tools solve standard functions —those stay— and which ones try, and fail, to hold up a process that's yours. Third, for the proprietary, work out the three-year cost of staying on SaaS versus building bespoke, including the cost of adapting. Fourth, start with the highest-return piece, measure it and expand only once the saving is proven.
Each well-chosen piece pays for itself before the next one begins. And every subscription you cut is a fixed cost that doesn't come back.
The shortcut: build just enough, with AI
Deciding well between buying and building doesn't mean choosing custom development for everything. It means buying what's commodity and building, fast and cheap, what makes you win. That's how Obsidy works: we look at your tool stack, identify which process is costing you money in subscriptions and workarounds, tell you what to build, what it would cost and in how many months it pays back, and leave it running with your team at the centre.
Are you paying every month for software that almost fits and forces your business to adapt? Let's talk. Write to us at hola@obsidy.com or visit obsidy.com and in a twenty-minute call we'll tell you what you'd buy, what you'd build and how much you'd save against your current stack of subscriptions.
Sources: Zylo 2025 SaaS Management Index and JumpCloud 2025 (average SaaS spend ≈$4,830 per employee/year, +21.9% YoY; companies of 1-500 employees ≈152 applications; 51% of licences unused, all-time waste high); Gartner SaaS Economics 2025 (integration, training and mandatory customization raise TCO by 150%-200% over the advertised price; true total cost 2.5x-4x the fee); TrustRadius (the direct licence is 36% of total cost over five years); build-vs-buy analyses 2025-2026 (SaaS fee from tens to hundreds of € per user/month; custom-software break-even ≈24 months with stable workflows and 20+ users, clearly cheaper from year three); GitHub + Accenture, 4,800 developers (up to 55% faster with AI); McKinsey (≈30% lower time-to-market; up to 80% lower cost on initial builds); GitHub and Gartner (AI generates ≈46% of code, projected 60% of new code by end of 2026); IDC 2025 and Spain SME digitalization data (44% use management software, 7 of 10 on cloud ERP, only 28% use sector-specific software); own case work and ROI framework (obsidy.com).