Costs and decisionsUpdated 10 min read
Custom vs. off-the-shelf software: build or buy?

In this article
Before you ask anyone for a quote, settle one question: do you buy off-the-shelf software or have something built for you? Most build vs. buy guides answer "it depends" and stop there. This one gives you a score: ten questions worth 0 to 3 points each and four possible outcomes. And a warning up front: if the math says buy, buy.
What each one is
Off-the-shelf software is a product built for thousands of companies at once. You configure it (users, taxes, warehouses), but the vendor designed the workflow, and changes ship when the vendor decides. In exchange, you can start fast, the upfront cost is low and someone else keeps it updated.
Custom software is built around how your company works: your steps, your pricing rules, your approvals. It costs more upfront and takes longer to go live, but it doesn't force you to change how you sell to fit the system.
SaaS isn't the opposite of custom. SaaS describes how software is delivered and paid for (in the cloud, by subscription), not who designed it. A custom system can also be billed per user per month, with hosting and support included.
The rule: buy the plumbing, build the edge
Martin Fowler tells the story of a prospective client who said software was like the plumbing: it has to work, and nobody should have to think about it (Fowler, 2010). Fowler splits software into utility and strategic. Utility software you buy, and you bend your process to fit the package; strategic software, the kind that sets you apart, you build. Joel Spolsky boiled it down to one line: if it's a core business function, do it yourself (Spolsky, 2001).
Scaled down to a small business: a distributor doesn't win customers with its payroll or with how it sends invoices. It wins them with how its reps take orders in the field, the prices it negotiates with each account and how fast it ships. The first group you buy; the second is a candidate to build. That's the rule: buy the plumbing, build the edge.
The uncomfortable part is that almost everything you use should be off-the-shelf. The useful question is which part of your operation makes you different, and whether you run it in a system today or in spreadsheets.
Four paths, not two
There are two paths between buying and building, and plenty of companies end up on one of them.
| Path | What it means | When it fits |
|---|---|---|
| Buy | You use the product as it comes and adapt your process. | The process works like it does anywhere: accounting, payroll, email. |
| Configure | You tailor a broad product with its own settings: fields, workflows, permissions and reports. | The package covers almost everything, and its settings handle the rest. |
| Buy and integrate | You connect several off-the-shelf products so data moves between them on its own. | Every piece exists, but today someone copies data between them by hand. |
| Build | You build the part that sets you apart and connect it to the off-the-shelf tools that already work. | How you operate is the advantage, and no package covers it without spreadsheets on the side. |
Configuring isn't instant either. In Panorama's report on ERP rollouts (170 projects, mostly at larger companies), the median project took 9 months and 22.3% finished late (Panorama, 2026). Compare the real timelines of both options, not the demo date against the delivery date.
The scorecard: 10 questions to decide
Score one specific process (field sales, inventory, collections), not the whole company. Give each question 0 to 3 points: 0 pushes you toward buying, 3 toward building. If you're torn between two numbers, take the lower one: anyone already picturing their own system tends to give themselves extra points.
| Question | 0 points | 3 points |
|---|---|---|
| 1. Does this process help you win or keep customers? | You do it like everyone else | It's a reason customers buy from you |
| 2. How much of what you need does the best off-the-shelf option you tried cover? | 90% or more | Less than 60% |
| 3. How many spreadsheets run alongside the system? | None | Four or more, used weekly |
| 4. How many tools need to share data? | One | Four or more, and someone copies between them |
| 5. What does off-the-shelf cost over 3 years compared with custom? | Clearly less | The same or more |
| 6. When do you need it running? | In under 4 weeks | You can wait 2 to 3 months |
| 7. Who will maintain it? | Nobody, and you don't want to pay for it | A vendor with support and improvements in the monthly fee |
| 8. How much does it hurt to depend on another vendor's roadmap? | Not at all | A feature that never shipped or a price hike has held you back |
| 9. Do you need data from several teams in one report? | The package's reports are enough | Yes, and you build it by hand every month |
| 10. How often do your rules change (pricing, commissions, approvals)? | Almost never | Every month, or they differ by customer |
Question 5 takes some math. Off-the-shelf side (you can add up your current licenses with the software cost calculator): license per user × users × 36 months, plus the add-ons and connectors billed separately, plus the hours your team would keep spending in spreadsheets. Custom side: development plus 36 monthly fees.
What your score tells you
| Score | Path | What to do |
|---|---|---|
| 0 to 7 | Buy | Pick the most widely used product in your industry and adapt your process. Building here means overpaying. |
| 8 to 14 | Configure | Look for a package that can be tailored and ask for a trial with your real data. |
| 15 to 21 | Buy and integrate | The pieces exist, but they don't talk to each other. Connect what you have before replacing it. |
| 22 to 30 | Build | Build the process you scored and keep the plumbing (accounting, payroll) off-the-shelf, connected to it. |
One cutoff rule: if question 1 scored 0, don't build that process even if the total is over 22. If it doesn't set you apart, configuring or integrating will fix the pain.
Worked example: field sales at a distributor
This example is made up, with round numbers. A distributor has 18 users, 8 of them sales reps in the field. Accounting runs fine on an off-the-shelf product. Orders come in over WhatsApp and someone retypes them, customer pricing lives in one spreadsheet and commissions get calculated at month-end in another. Two people spend about 6 hours a week each copying data.
| Question | Points | Why |
|---|---|---|
| 1. Differentiation | 3 | Reps show up with today's price and stock |
| 2. Coverage | 2 | The package covers about 70%; customer pricing and routes are missing |
| 3. Side spreadsheets | 3 | Five: pricing, routes, commissions, orders and receivables |
| 4. Tools | 2 | Accounting, WhatsApp and spreadsheets |
| 5. 3-year cost | 2 | The math is below |
| 6. Urgency | 1 | They want something in about 6 weeks |
| 7. Maintenance | 2 | They'd rather pay a monthly fee with support |
| 8. Vendor dependence | 2 | The offline app the package announced still hasn't shipped |
| 9. Cross-team data | 3 | Sales by rep vs. receivables, by hand every month |
| 10. Rules | 3 | Prices change monthly, commissions vary by product line |
| Total | 23 | Build |
The math for question 5, off-the-shelf side: a US$40 per user per month license for 18 users comes to US$25,920 over 3 years. The field sales add-on, at US$25 per rep per month, adds US$7,200, and a US$150-a-month connector adds another US$5,400. Total: US$38,520, and customer pricing still lives in the spreadsheet. That's the number to set next to the custom quote: if the quote comes in clearly below it, the question scores 3; if it's close, 1 or 2.
Result: 23 points, build, but not everything. Accounting stays where it is. What gets built is field ordering, customer pricing, commissions and the cross-team report, connected to accounting so invoices go out without anyone retyping. Urgency tells you where to start: orders first, commissions in the second release.
When custom isn't worth it
This is the section a lot of custom software shops skip. Don't build if:
- It's accounting, payroll or tax compliance. The rules change with regulation, and an off-the-shelf vendor updates them for thousands of customers at once. Connect to it; don't replace it.
- You have fewer than 5 users and ordinary processes. A well-configured package will last you for years.
- You need it next week. Serious development, even a small project, takes weeks.
- Nobody on your team can spend a few hours a week explaining how they work and testing releases. Without that person, the system gets built on guesses.
- You want to copy an existing package screen by screen. Buy the original.
And if you do build, build a little at a time (how much custom software costs breaks down the price). IT projects over US$15 million run 45% over budget on average and deliver 56% less value than predicted (McKinsey and Oxford, 2012). In another study of 1,471 projects, 1 in 6 had an average cost overrun of 200% (Flyvbjerg and Budzier, 2011). One module on a fixed timeline is a much smaller bet than "the whole system."
What changed in 2026: building got cheaper, maintaining didn't
Retool surveyed 817 people in late 2025. 35% had already replaced at least one SaaS tool with something they built, and 78% planned to build more in 2026; the most replaced categories were automations (35%) and internal admin tools (33%) (Retool, 2026). Keep the sample in mind: these are Retool customers, people who already use a platform for building tools.
On the other side is what gets paid for and never used: the average company has 305 SaaS apps and uses only 54% of the licenses it buys (Zylo, 2026). Those are large companies, but it's worth checking how many seats you pay for and how many people actually log in, with a subscription audit.
AI made the first version cheaper, which is the core argument of Retool's report. It didn't make maintaining it cheaper. Robert Glass found that maintenance takes 40% to 80% of a software system's total cost, and that 60% of that maintenance is enhancements, not bug fixes (Glass, 2001). Before you build, ask who will make them and what they'll cost: that's question 7 on the scorecard.
How to combine both without duplicate data
If your scorecard said build, you'll end up with two systems: off-the-shelf for the plumbing and yours for operations. Stripe makes the same case for billing: the pricing strategy is yours, and the infrastructure is bought (Stripe, 2026). It works as long as every piece of data has exactly one owner.
| Data | System of record | What gets sent to the other |
|---|---|---|
| Orders, routes and customer pricing | Custom | The approved order, so it can be invoiced |
| Invoices, receivables and payments | Off-the-shelf (accounting) | Each invoice's status and the balance per customer |
| Customers and products | Whichever one you pick | Every new record or change, always in the same direction |
The test is simple: nothing gets typed twice, and if both systems can create customers, you'll have duplicates within a few months. Set up this way, switching off-the-shelf products later doesn't force you to rebuild your own system: you replace a connection, not your operation.
Sources
- Martin Fowler, "Utility Vs Strategic Dichotomy," 2010 (updated 2016) (opens in a new tab)
- Joel Spolsky, "In Defense of Not-Invented-Here Syndrome," 2001 (opens in a new tab)
- Panorama Consulting Group, "2026 ERP Report," 2026 (opens in a new tab)
- Bloch, Blumberg and Laartz, McKinsey and University of Oxford, "Delivering large-scale IT projects on time, on budget, and on value," 2012 (opens in a new tab)
- Bent Flyvbjerg and Alexander Budzier, "Why Your IT Project May Be Riskier Than You Think," Harvard Business Review, 2011 (opens in a new tab)
- Retool, "2026 Build vs. Buy Report," 2026 (opens in a new tab)
- Zylo, "2026 SaaS Management Index," 2026 (opens in a new tab)
- Robert L. Glass, "Frequently Forgotten Fundamental Facts about Software Engineering," IEEE Software, 2001 (opens in a new tab)
- Stripe, "AI billing infrastructure: Rethinking the build-versus-buy decision," 2026 (opens in a new tab)