ByAhsan Nadeem@ashtech52Lead Full-Stack Engineer
Custom software vs SaaS comes down to one sum: what you would really pay the vendor over five years, once unused seats are gone, against what it costs to build and then own the replacement. For most tools, the answer stays "keep the SaaS".
Building wins when the bill is large, rising and still large after a seat clean-up. In our worked example, a $120,000 tool with 10% yearly rises breaks even in about 3.5 years. Remove the unused seats first and that becomes about 8 years.
A declaration of interest: we sell custom business software, so you should check our sums. Every assumption below is visible so you can replace it with your own. The figures are estimates, not quotes, and they don't come from a client project.
Our view: AI coding tools may lower the cost of building. They don't lower the cost of owning. So the test needs to weigh compounding renewals against the upkeep you take on, and in the example, seat count moves the answer far more than build price does, while higher upkeep undoes about what a 30% cheaper build saves.
Custom software vs SaaS: your default, by situation
Your situation | Default |
|---|---|
Tool does a job every company does, such as payroll or email | Keep the SaaS |
Bill is small, or mostly unused seats | Keep the SaaS and cut seats |
Tool is fine, but your workflow spans several tools | Keep them and build a thin layer on top |
Large bill, heavy use, and you've bent your process to fit | Run the break-even test |
Your process is what customers pay you for, and nothing fits | Build, and budget for ownership |
What is pushing the question now: renewal rises and AI bundling
Buyers are asking because renewals got harder to absorb. Nearly every number here comes from a company selling SaaS management or buying services, so read them as vendor data.
- Named list prices. Microsoft raised its commercial prices on 1 July 2026, with existing customers moving at renewal. Its table lists Business Basic up 16%, Business Standard up 12%, Microsoft 365 E3 up 8% and frontline F1 up 33%. Zylo reports that Salesforce raised Enterprise and Unlimited editions by 6% in August 2025, after 9% in 2023.
- Vertice's inflation index (vendor). Readings in 2026 were 16.4% in June, 12.9% in July, 12.7% in August and an early 14.3% in September, drawn from over $75 billion of spend it processed. Its August 2025 report said "over 12%". The number moves every month, so quote the month.
- Zylo's index (vendor). It says 79% of IT leaders met a price rise at renewal in the past 12 months, without saying how large those rises were. Its pages don't give a sample size. Spend rose 8% while application counts barely moved.
- Tropic (vendor). It puts the uplift from AI bundled into renewals at 20 to 37%, from its own contract analysis, with no sample size given.
- Vertice again (vendor). It has noticed prices 49 to 63% higher where SaaS includes AI features, again with no sample size. That compares products, whereas Tropic's figure measures renewals, so don't combine them.
Microsoft's table is hard evidence for one vendor. None of this tells you about your own contract, so use your last three renewal letters.
What Retool's survey does and doesn't show
The most quoted figure in this debate is that 35% have already replaced SaaS with custom software. It comes from Retool's 2026 Build vs. Buy report, and Retool sells a platform for building internal tools.
What it shows:
- 817 Retool customers and builders answered in late 2025, and 36% were software engineers or developers.
- 35% have replaced at least one SaaS tool, and 78% expect to build more in 2026.
- Pressure is highest on workflow automations (35%), internal admin tools (33%) and BI tools (29%), and lower on customer support (21%).
- Only 8% use AI-generated code unchanged, and maintenance burden is named as a blocker by 26%.
What it doesn't show:
- That a third of enterprises left SaaS. The headline says "enterprises", the body says "teams", and the sample is the vendor's own users. One replaced tool out of an average portfolio of 305 (Zylo's figure, from a different dataset) isn't a migration.
- That anyone saved money. The pages we read give no savings figure.
- That the replacements lasted. Nothing tracks them a year or two on.
- What was done and what was planned. The category chart doesn't separate the two.
The more useful number may be that 60% built something outside IT oversight in the past year. Retool's own line is apt here: "replacement without measurement is just churn".
The break-even test, worked through with stated assumptions
Published answers don't help much. A seller of custom builds claims 18 to 36 months. A vendor's build-or-buy calculator shows a sample where building never pays back within ten years. Both can be true, because the inputs decide it. So run your own.
Step 1: right-size the bill. Remove unused seats first, because that's money you can save without building anything. Zylo's customers use 54% of their licences on average. Yours may differ, so pull your own login data.
Step 2: compound the rises. Use your actual renewal history, then a higher and a lower case.
Step 3: cost the ownership. Include upkeep, hosting, security tooling and someone who owns the product.
Step 4: add one-off costs. Include migration and training, and keep paying for the SaaS while you build.
Our quick version:
Years to payback = upfront cost / (right-sized annual bill - annual running cost) + build time in years
The quick version ignores renewal rises, so treat it as a cautious first pass. On the example below it gives about 4.5 years, close to the 4.7 in the flat-price row, while 10% rises bring the full model down to about 3.5.
If the answer is well over four years, keep the SaaS. If it's near four, run the full table with your real rises. Four is our rule of thumb, not a law. Beyond that, you're betting on a forecast of vendors, staff and your own process.
The assumptions
We found no independent study of upkeep as a share of build cost. The 20% below is our assumption, inside the 10 to 25% range offered by that vendor calculator.
Item | Example value |
|---|---|
SaaS bill, year one | $120,000 |
Build cost, over year one, with SaaS paid in full that year | $180,000 |
Migration and training | $20,000 |
Upkeep from year two, rising 5% a year | 20% of build, so $36,000 |
Hosting, monitoring and security tools | $12,000 a year |
Internal product owner, part time | $15,000 a year |
That's $63,000 of running cost in year two. There's no discounting, tax or financing in these sums, and discounting would push break-even later.
The results
Yearly renewal rise | SaaS bill in year five | Break-even | Five-year net from building |
|---|---|---|---|
0% | $120,000 | About 4.7 years | +$17,000 |
5% | $145,900 | About 4.0 years | +$80,000 |
10% | $175,700 | About 3.5 years | +$149,000 |
15% | $209,900 | About 3.3 years | +$226,000 |
Compounding matters, but look at how little break-even moves between 10% and 15%. Now hold the rise at 10% and change something else:
What changes | Break-even | Five-year net |
|---|---|---|
Base case | About 3.5 years | +$149,000 |
Build 30% cheaper, upkeep unchanged (hypothetical) | About 3.0 years | +$203,000 |
Upkeep is 30% of build, not 20% | About 4.2 years | +$72,000 |
You negotiate 15% off the renewal ($102,000) | About 4.3 years | +$58,000 |
Build overruns by 50%, and upkeep scales with it | About 5.2 years | -$18,000 |
Seats cut to 54% ($64,800) | About 8.1 years | -$132,000 |
Seats cut to 54%, with rises of 5% | Not within 10 years | -$170,000 |
The negotiation case is realistic, since Zylo reports average savings of 16.8% at renewal (vendor figure). Talking to your vendor is the cheapest test you can run.
The seat line is the one to look at. The quick formula, which ignores rises, gives $200,000 divided by $1,800 of yearly saving, or 111 years. The full model with 10% rises gets it to about 8. An inflated bill can make a build look sensible when the honest comparison says otherwise.
Small tools rarely pay back at all. Take a $24,000 tool with a $60,000 build, $5,000 of migration, 20% upkeep, $3,000 of hosting and a $5,000 owner. Running cost is $20,000 a year, so the saving is about $4,000. With 10% rises it breaks even in about 6.5 years, and with 5% rises not within seven.
Costs that get left out on both sides
Left out of the SaaS column:
- Seat growth as you hire, on top of price rises.
- Add-ons, plus surprise usage charges: Zylo says 78% of IT leaders met unexpected consumption or AI-related charges (vendor figure).
- Time your staff spend on workarounds. Measure a real week rather than estimating it.
- The price of leaving, covered below.
Left out of the custom column:
- Upkeep that never ends: dependency updates, security patches, browser and API changes.
- Someone who answers when it breaks at 8pm.
- A named owner, because knowledge leaves with people.
- Audit and compliance work the vendor used to carry.
- The features your SaaS would have shipped for free.
Exit costs are changing in the EU
If you buy in the EU, leaving SaaS may soon cost less, which makes staying less risky and weakens one common argument for building. The EU Data Act covers SaaS as well as infrastructure. According to Germany's Bundesnetzagentur and the European Commission:
- From 12 January 2027, providers can't impose switching charges, including data egress charges. Until then, they may charge only direct switching costs.
- A provider has up to two months to prepare, then a transition of up to 30 calendar days. If that isn't feasible, it must explain why within 14 working days, and the transition can run to seven months.
- Your data must stay retrievable for at least 30 calendar days afterwards.
There are limits. Individually designed services not offered at broad commercial scale are partly exempt. A Digital Omnibus proposal would also widen exemptions for older contracts, and it was still being negotiated when Greenberg Traurig reviewed it on 15 July 2026. IAPP reported on 22 October 2026 that the Council had agreed its negotiating position, with talks to follow once Parliament has its own, so check where those exemptions ended up. This is an EU rule, so UK and US buyers shouldn't assume it applies. Check early termination terms with your lawyer too, because we couldn't settle that point from the summaries we read. This isn't legal advice.
What AI coding tools change, and what they don't
The evidence is unsettled:
- METR, 2025. In a randomised trial of 16 experienced open-source developers, tasks took 19% longer with AI, while the developers believed they'd been faster. METR now says those results are out of date.
- METR, 2026. The follow-up found 18% less task time for returning developers and 4% less for new recruits, and both ranges include no change. METR calls this very weak evidence of size, and suspects the true speed-up is higher because developers refused to work without AI.
- DORA, 2025. Its report treats AI as an amplifier of an organisation's existing strengths and weaknesses. RedMonk's review says the link with throughput turned positive, while the link with delivery instability persists.
So assume building may get cheaper. Our table tests that directly: a build 30% cheaper, with the same upkeep, moves break-even by about seven months. A seat clean-up moves it by more than four years. That 30% is a what-if, not a measurement.
AI doesn't remove the owner, the on-call rota, the security review or the audit. Retool's data points the same way: even among its builders, only 8% use AI-generated code unchanged. If a prototype took a weekend, that's a good way to learn what you need. It isn't evidence about year three. For a sense of what a first version costs when done properly, see our guide to SaaS MVP development cost.
Keep, replace or wrap: a decision table by type of tool
"Wrap" means you keep the SaaS as the system of record, and build a small layer on top through its API. The defaults below are our judgement. The Retool column is the share of its respondents naming that category as at risk (vendor data).
Type of tool | Retool | Default | Why |
|---|---|---|---|
Workflow automation | 35% | Wrap or replace | Logic is yours, and the scope is usually small |
Internal admin tools | 33% | Replace if heavily used | Few users and simple permissions, but give it an owner |
BI and dashboards | 29% | Wrap | The data model is the hard part, not the charts |
CRM and forms | 25% | Keep, or wrap the CRM | Integrations and history are costly to move; simple forms are fine |
Project management | 23% | Keep | Commodity, with little to gain |
Customer support | 21% | Keep or wrap | Channels and reporting are costly to reproduce |
Accounting, payroll, tax | Not listed | Keep | The rules keep changing |
Identity, email, security | Not listed | Keep | A failure is expensive and the vendors specialise |
Your core process | Not listed | Build if nothing fits | This is where owning it can be an advantage |
What not to build
- Anything regulated that changes often. Payroll and tax are the clearest cases. UAE e-invoicing alone had four ministerial decisions and a new penalties decision in 2026, as our UAE e-invoicing guide shows. Let a vendor carry that.
- Security-critical plumbing. Don't write your own login, identity or email delivery.
- Tools under the payback line. If the full test, with your real rises, says over four years, stop.
- Replacements with no owner. If nobody will be named in the contract, keep the SaaS.
- A copy of the whole product. You probably need 20 screens, not all of them, and a well-scoped project forces that discussion early. Whether you then use a dedicated developer or an agency is a separate decision.
Frequently asked questions
Is custom software cheaper than SaaS in the long run?
Sometimes, and only for large, heavily used tools. In our worked estimate, a $120,000 tool with 10% rises came out about $149,000 ahead over five years, but a smaller $24,000 tool was still behind after five years. Unused seats change the answer most. Run the sum on your own numbers, including upkeep.
When should you replace SaaS with custom software?
When your right-sized bill stays well above the yearly cost of owning a replacement, payback comes in under about four years, and someone will own the result. Also when your process is what customers pay for and no product fits. Try negotiating, removing seats and building a thin layer on top first.
How long does break-even take?
There's no standard answer. One seller of custom builds claims 18 to 36 months, while a vendor calculator's sample never pays back within ten years. Our example gave about 3.3 to 4.7 years depending on renewal rises, and about 8 years after a seat clean-up. Your inputs decide it.
Can AI tools build our replacement for us?
They can help, and may make the build cheaper, though the research on speed is unsettled. METR's latest study calls its own result very weak evidence. AI doesn't remove upkeep, security review or the need for an owner. In Retool's survey, a vendor source, only 8% of respondents used AI-generated code unchanged.
Who maintains custom software afterwards?
You do, or someone you pay. Budget for dependency updates, security patches, hosting, monitoring and a named owner. We assumed 20% of the build cost each year for upkeep, plus hosting and an internal owner. We found no independent study of that percentage, so treat it as an assumption and ask any supplier for theirs in writing.
Can we use both?
Yes, and it's often the best result. Keep commodity tools such as payroll and email, and build only where your process is different. A thin custom layer over a SaaS system of record, connected through its API, can fix a workflow without taking on the whole product. Check the vendor's API limits and pricing first.
Before asking anyone for a quote, export your seat logins, get your last three renewal letters and run the test on your largest tool. If it comes out around four years or under and you'd like the assumptions challenged, book a free consultation. If the sum says keep the SaaS, we'll say so.
Sources
- Microsoft, 2026 Microsoft 365 packaging and pricing updates
- Retool, 2026 Build vs. Buy report announcement (vendor, 17 February 2026)
- Retool, the build vs. buy shift (vendor)
- Vertice, SaaS inflation rate (vendor, updated October 2026)
- Vertice, the blueprint for the perfect SaaS deal in 2026 (vendor, August 2025)
- Zylo, SaaS pricing trends (vendor, 9 June 2026)
- Zylo, SaaS statistics (vendor, updated 8 February 2026)
- Tropic, AI tax in software pricing (vendor, updated 26 September 2026)
- Bitvea founder, custom software vs SaaS on dev.to (seller of custom builds)
- TestMu, build vs buy calculator (vendor)
- METR, early-2025 AI and experienced open-source developers (10 July 2025)
- METR, uplift update (24 February 2026)
- DORA, 2025 report
- RedMonk, review of DORA 2025 (18 December 2025)
- Bundesnetzagentur, Data Act fact sheet on switching (regulator, June 2026)
- European Commission, Data Act explained
- Greenberg Traurig, Digital Omnibus amendments to the Data Act (15 July 2026)
- IAPP, EU member states finalise Digital Omnibus stance (22 October 2026)
- UAE Ministry of Finance, e-invoicing legislation list
