Decision framework9 min read
Custom CRM or off-the-shelf: which is worth it?
The answer comes down to six commercial questions: process fit, three-year cost, integrations, ownership, time to useful adoption and expected change.
- Author
- Xavier Barlow
- Reviewed by
- Félix Tarnóy
- Published
- Last reviewed
A Monday management meeting, one question on the table: how many open quotations do we have right now, and what are they worth? The sales lead says he will have it by Wednesday — from three spreadsheets, two inboxes and the notes of a colleague on leave. The number arrives, and it is roughly right. It is the word “roughly” that produces two proposals a fortnight later.
One is an established CRM product: a monthly fee per user, configurable in a few weeks. The other is a bespoke system built around how the business actually works, delivered over months. The two answer different questions, and the decision rarely turns on which is the better piece of software.
This article does not argue for building. For a great many companies the off-the-shelf product is the right answer, and that is worth saying before anyone starts looking for a developer. The six dimensions below make the decision follow from your own operation rather than from whoever presented last.
The six dimensions that settle it
This is not a technology choice; it is six commercial questions that arrive dressed as one. Answer them honestly and the decision usually makes itself.
- Process fit. How far does the way you work depart from what mainstream products assume by default?
- Three-year total cost. Not the entry price: licences, hosting, maintenance, development and internal hours added together.
- Integration. How many other systems must talk to this one reliably, and do ready-made connectors exist for them?
- Data ownership and exit. Who holds the data, in what form can it leave, and what does switching cost?
- Time to first useful use. From what point does the system help the daily work rather than slow it down?
- Expected change. How much will the process move in the next three years, and who decides when it moves?
The decision typically arrives at a particular size: past spreadsheets, short of an in-house IT function. Hungary had 33,189 active enterprises in the 10–49 band in 2024 — a band the statistical office builds on persons occupied, which counts working proprietors and unpaid family workers alongside employees.
33,189
Looked at more broadly, digital maturity reads modestly against the EU: in 2024, 57.4% of Hungarian SMEs reached at least a basic level of digital intensity, against an EU average of 72.9%. That is one specific index covering small and medium-sized enterprises as a whole, not a verdict on Hungarian firms’ technical competence — on data analytics, Hungary sits above the EU average.
Process fit: how unusual is what you actually do?
Start with a tedious, cheap exercise. Write out the process from enquiry to invoice, step by step, as it genuinely runs rather than as the policy describes it. Then mark the steps a comparable company in a different sector would not recognise. If one or two steps out of ten are marked, configuring a proven product will almost certainly be faster and cheaper. If five are, read on.
Where off-the-shelf is usually enough
- The sale runs in a line: enquiry, quotation, negotiation, contract, invoice.
- A quotation is a handful of product lines rather than a fresh calculation every time.
- Pricing is list price plus discount, not a project-level cost estimate.
- One customer means one agreement, not several running in parallel on different clocks.
- Everyone sees broadly the same thing — no data that must be sharply filtered by role.
Where off-the-shelf runs out
- The quotation is itself a calculation from dimensions, materials, installation time and site conditions — and that calculation is the firm’s real expertise.
- The process runs on several tracks: one order is simultaneously a sale, a production slot and a site visit.
- External partners see into the system, but only the part that concerns them.
- The operation requires a record that statute or a professional standard prescribes and the product has never heard of.
- The licence grows with headcount while half the users enter one field a day.
The most common misconception is that a custom build produces a better process. It does not. It reproduces the existing process in software, faults included, at greater expense. An off-the-shelf product can be an advantage precisely here: it imposes a proven way of working that the company would never have adopted on its own.
The three-year total cost
You will find no prices in this article, because every figure follows scope. The shape of the cost, though, is predictable on both sides — and worth filling in with your own numbers before either proposal is signed.
- Subscription on the real user count. On the tier you actually need, with three years of likely price movement included.
- Implementation on the off-the-shelf side. Configuration, migration, training, and the add-on modules without which the product does not cover the process.
- Missing integrations on the off-the-shelf side. Anything without a ready connector has to be written and then maintained.
- Building the smallest useful custom version. Design, development, data migration, rollout.
- Running the custom system. Hosting, monitoring, security updates, support, defect fixes — a monthly line, not a one-off.
- Internal hours, on both sides. Who curates the reference data, who grants access, who decides what changes.
An illustrative scenario with stated assumptions — not client work, and not a promise. Assume twelve people would use the system today, twenty in three years, and the process stays substantially as it is. On those assumptions the subscription line rises roughly with headcount, while a one-off build costs the same at twelve users as at twenty. The operating line does rise, but with the number of integrations and the rate of change requests rather than headcount. Change any assumption and the answer can flip.
It is also worth separating what you buy from what you will use. CRM products commonly place artificial intelligence features on higher tiers, and adopting them is a project in itself. In 2025, 10.4% of Hungarian enterprises with 10 or more persons employed used at least one artificial intelligence technology, against an EU average of 20.0%. Paying for a tier whose headline features nobody ever switches on is the most common hidden cost in this decision.
Free assessment
Does this sound familiar? Let us discuss the best place to begin.
Your assessment is carried out personally by one of our experts.
Integration, ownership and exit
How many systems have to talk to each other?
List what must connect: invoicing, accounting, the website forms, the advertising accounts, email and the calendar. Then check which of those the candidate product already connects to. A widely used system that plugs straight into your invoicing software is often a stronger proposition than a bespoke build in which the same connection has to be written from scratch. The reverse holds too: if the system that matters most is a Hungarian sector-specific application no major CRM connects to, development is required on either path.
One thing applies in both cases: every integration is a permanent maintenance obligation. Interfaces change, credentials expire, and somebody has to notice when an overnight sync quietly stops running.
Who owns the data, and what does leaving cost?
Ownership is really two questions: who can reach the data, and can you take it with you? Ask the vendor four things, preferably in writing: in what format the complete dataset can be exported; whether attachments, notes and history come with it or only the core records; what notice period applies; and what leaving costs.
At EU level this ground is moving. The Data Act has applied since 12 September 2025, and the European Commission states that switching charges between data processing services — cloud and edge services — including data egress charges, are removed entirely from 12 January 2027. It does not reach every software vendor and is no substitute for reading the contract, but it improves the negotiating position.
One thing does not depend on the choice at all: your company remains the controller. Personal data may be processed only on a valid and documented legal basis under the EU rules, whether it sits on your own infrastructure or inside a vendor’s. What an off-the-shelf product typically covers is the processor agreement and the operational security of the infrastructure. It does not cover the legal basis, the retention period or the access rules.
The two options side by side
| Dimension | Off-the-shelf | Custom build |
|---|---|---|
| Process fit | The process bends to the product; tailoring stops at configuration. | The system bends to the process — including a poor one. |
| Time to first useful use | Weeks: account, fields, imports, training. | Months: even the smallest useful version needs design and migration. |
| Cost shape | Low entry cost, then a subscription that rises with users and tier. | High entry cost, then operation; headcount does not drive it. |
| Integrations | Ready connectors to major systems; anything missing is separate development. | Connects to anything with an interface; every connection is maintenance. |
| Data and exit | Data exports, but the structure belongs to the vendor; switching means redesign. | The database and the source code are yours — and so is the responsibility. |
| Security and operations | Usually the vendor’s job; what it answers for is set by the contract. Granting access stays with you. | Yours: policies, key management, backups, restore drills, updates. |
| Change | Arrives with the product roadmap; you do not set the timing. | What you decide, when you decide it — subject to capacity and cost. |
What building actually commits you to
Choosing to build is an operations decision rather than a development one, so it is worth knowing what lands on the desk after the final invoice is paid. None of the following is a deterrent: on a modern platform these are well-documented facilities you get out of the box. They simply all have to be configured, and then kept in order. The examples below come from the documentation of one such platform, Supabase; the names differ from vendor to vendor, but the work does not.
Access control is the clearest example. Row level security decides which rows a given role may reach inside the database engine — not in the interface and not in an API layer, but where the query finally runs. That is precisely why it still holds when a query arrives by a route the developer never wrote. The documentation records a detail worth understanding: two checks run in sequence. A grant decides whether a role may touch the table at all, a policy decides which rows — and turning policies on does not take existing grants back.
Keys work the same way. The publishable key intended for the browser and the secret key that stays on the server are not interchangeable: a secret key bypasses row level security, so it must never reach a browser, a shipped application or version control. The same page states the least-privilege rule plainly: grant every role only what the application needs to function. Replacing a leaked key is a procedure rather than a click — create the replacement, roll it out, confirm, and only then delete the old one, which cannot be undone.
Changing the schema is disciplined work too: migrations are timestamped files applied in order, and the documentation’s golden rule is never to change the live database directly, because doing so breaks the migration history. Backups are similarly nuanced. Daily backups run automatically, but the retention window is finite and varies by plan, and finer-grained recovery to a chosen moment does not come as standard. The point that matters most is that database backups do not include files uploaded to the storage service, since the database holds only the metadata pointing at them. Signed contracts and site photographs therefore need separate provision.
Access policies and the tests behind them
Every new table and role means a new policy, and a policy is only worth having if a test proves it does what was intended. That test suite then has to be maintained like any other.
Key management
Who knows the secret keys, where they are held, and what happens when someone leaves. Rehearse the replacement rather than learning it during an incident.
Change control
Every schema and function change goes through review and version control, and one named person decides when it reaches production.
Backups and a restore drill
A backup existing is not protection. At least once a year, restore a copy and time how long it takes to have a working system again.
Dependency and access review
Frameworks move, and people join and leave. Reviewing who can reach what is a quarterly task, not a line in the closure report.
None of this is an argument against building. It sets out the size of the job a subscription usually hands to the vendor — the contract is where you confirm how much of it — and which a company that builds has to buy instead, either in a person or in a support agreement. If that appears in neither plan, the solution did not get cheaper. The cost merely became invisible.
What to do over the next fortnight
You do not need a developer to reach the decision. Five steps are enough, and your own team can do four of them.
- Write out the process step by step, with the person who performs it daily. Mark the steps that belong to your company alone.
- Count how many people would use the system every day and how many would only look at it. Licence models often price the two the same.
- List the integrations that are genuinely mandatory and check which have ready connectors. That list settles most of the argument.
- Ask the vendor about export, notice period and the cost of leaving — in writing, before signing.
- Name the person who will own the system after go-live. If there is no name, neither option is worth starting.
If the process is largely conventional, the answer is the off-the-shelf product, and the money goes on implementation and training. If part of it is the company’s genuine expertise, still start with the smallest useful custom version. The common mistake is not choosing wrongly; it is taking too large a first step.
If an outside view would help, our custom software and CRM page sets out how the process and data map is produced, along with the written recommendation that justifies the conclusion in either direction. The automation and action plan modules of the digital assessment are built around these questions.
Sources
Hungarian Central Statistical Office (KSH) — Number of active enterprises by size category
In 2024 Hungary had 33,189 active enterprises in the 10–49 size band, 5,076 in the 50–249 band and 1,027 with 250 or more.
The bands count “persons occupied”, which includes working proprietors, regularly working partners and unpaid family workers. BD-SBS harmonised methodology; agriculture and public administration are excluded.
https://www.ksh.hu/stadat_files/gsz/en/gsz0071.htmlLast reviewed: 1 September 2026
European Commission — Hungary — 2025 Digital Decade country report
In 2024, 57.4% of Hungarian SMEs reached at least a basic level of digital intensity, against an EU average of 72.9% and the EU 2030 target of 90%.
Digital Intensity Index, 2024 reference year, small and medium-sized enterprises.
https://digital-strategy.ec.europa.eu/en/factpages/hungary-2025-digital-decade-country-reportLast reviewed: 1 September 2026
Eurostat — Use of artificial intelligence in enterprises (isoc_eb_ai)
In 2025, 10.4% of Hungarian enterprises with 10 or more persons employed used at least one artificial intelligence technology, compared with an EU average of 20.0%.
Enterprises with 10 or more persons employed, NACE Rev. 2 Sections C to N excluding Section K and Division 75, including Group 95.1. The financial sector, agriculture and mining are outside the scope.
https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table?lang=enLast reviewed: 1 September 2026
European Commission — Data Act explained
The Data Act has applied since 12 September 2025. The European Commission states that switching charges between data processing services — including data egress charges — are removed entirely from 12 January 2027, while between 11 January 2024 and 12 January 2027 providers may still charge the costs they actually incur in relation to switching.
Applies to data processing services, that is, cloud and edge services.
https://digital-strategy.ec.europa.eu/en/factpages/data-act-explainedLast reviewed: 1 September 2026
European Commission — Legal grounds for processing personal data
A controller must have — and document — a valid legal basis for processing personal data.
https://commission.europa.eu/law/law-topic/data-protection/information-business-and-organisations/legal-grounds-processing-data_enLast reviewed: 1 September 2026
Supabase — Row Level Security
PostgreSQL row level security defines, per table and in policy form, which rows a given role may access.
https://supabase.com/docs/guides/database/postgres/row-level-securityLast reviewed: 1 September 2026
Supabase — Securing your API
A grant and a row level security policy are two separate layers: the grant decides whether a role may reach the table at all, the policy decides which rows. Adding policies does not revoke existing grants, and tables and views exposed through the Data API without RLS can be accessed by any role with matching grants.
https://supabase.com/docs/guides/database/hardening-data-apiLast reviewed: 1 September 2026
Supabase — Securing your data
The documentation states that secret and service role keys are never safe to expose because they bypass row level security, so they belong on the backend only; roles should be granted the least privileges necessary for the application to function.
https://supabase.com/docs/guides/database/secure-dataLast reviewed: 1 September 2026
Supabase — API keys
A publishable key is designed for browsers, mobile applications and shipped code, while a secret key belongs only on servers, APIs with their own authorisation checks and background services. Replacing a leaked secret key is a procedure: create the replacement, roll it out everywhere, confirm, and only then delete the old one — a deletion that cannot be undone.
https://supabase.com/docs/guides/getting-started/api-keysLast reviewed: 1 September 2026
Supabase — Database migrations
Schema changes are timestamped SQL files applied in order, with a separate migrations table recording what has been applied. The documentation’s golden rule is never to change the remote database directly: doing so bypasses the migration history and makes the next deployment fail with sync errors.
https://supabase.com/docs/guides/deployment/database-migrationsLast reviewed: 1 September 2026
Supabase — Backups
The platform provides automatic daily backups whose retention depends on the plan, with point-in-time recovery available as a paid add-on. Database backups do not include objects stored through the Storage API, because the database holds only metadata about them.
https://supabase.com/docs/guides/platform/backupsLast reviewed: 1 September 2026