Skip to content
Shop integration & processes

iPaaS or Custom Build: Integration Platform vs. Middleware

iPaaS or custom middleware for ERP-shop integration: a five-year cost example with break-even, GDPR and Data Act rules, an exit checklist and a scoring matrix.

22 min read iPaaSMiddlewareArchitekturBuild-or-BuyIntegration

In short and in plain language

Your shop needs to work with your inventory system and accounting. Then a question comes up. Do you rent a ready-made platform? Or do you build your own?

A ready-made platform costs little at the start. But you pay fees all the time. Your own build costs more at first. But it fits your work exactly.

It also matters where your data lives. With your own build, it stays in your hands.

You can also mix both. The post helps you choose.

We look at your programs in a neutral way. Then we build the link that fits.

When an online store needs to work with an ERP, inventory system, accounting or CRM on a permanent basis, a fundamental question comes first: do we buy a ready-made integration platform (iPaaS) with a connector toolkit, or do we build a tailored middleware ourselves? This is not a matter of taste, because it shapes total cost, data sovereignty, time to go-live and dependence on a single vendor for years. The need is real: according to Bitkom, 61 percent (Bitkom) of companies with 20 or more employees barely use their data or do not use it at all. An integration layer that reliably moves data between the store, the ERP and other systems is a precondition for changing that. This article is a build-or-buy decision guide as of November 2026: it places iPaaS and custom development side by side, works through a five-year example with its break-even point, shows what the GDPR and the Data Act require for the data path and for switching providers, and closes with a scoring matrix and three worked cases.

Key takeaways

  • iPaaS means buy: a rented platform with ready connectors, monitoring and operations. Custom development means build: a middleware that mirrors your processes exactly and runs without a platform license.
  • In the worked example with assumed figures, the custom build recovers a head start of 40,000 euros in year four because the volume-based license rises every year. With flat volume, iPaaS stays cheaper over five years.
  • With personal data, an iPaaS platform is usually a processor: a contract under Art. 28 GDPR, authorization of further processors and, for third-country transfers, the conditions of Art. 44 et seq. GDPR belong before the selection.
  • 59 percent (Bitkom) of the companies surveyed name lock-in effects as the biggest obstacle to switching cloud providers. From January 12, 2027, switching charges are no longer allowed (Art. 29(1) Data Act), but the rebuild effort remains.
  • Business Central online receives two major releases every year, in April and October (Microsoft Learn). Monitoring deprecations through the Deprecation and Sunset header fields lets you plan rebuilds instead of being surprised by them.
  • A scoring matrix with four criteria – process idiosyncrasy, data sensitivity, change frequency, volume – assigns each data flow on its own. Mixed results point to the hybrid path, and an exit checklist keeps both paths open.

Build or Buy: The Core Decision

Build-or-buy is one of the oldest questions in IT and particularly consequential in the integration space. Buy here means: an integration platform as a service (iPaaS) provides ready connectors for common systems, plus graphical mapping tools, flow control, logging and operations. You configure instead of program, and pay a recurring fee for it. Build means: a dedicated middleware is developed that mirrors exactly your data models, special cases and processes – as code that runs in your environment or with an operator of your choice, with usage and modification rights governed by the development contract.

Neither answer is fundamentally right. The core decision depends on how many systems interact, how idiosyncratic your processes are, how important data sovereignty is and how quickly the first version must stand. A retailer with standard workflows often moves faster with a connector toolkit; a business with deep manufacturing logic, complex pricing rules or regulatory requirements frequently needs tailored middleware. Because system landscapes change over time, it pays to make the decision not just for today but for the coming years. Anyone planning an early customer-data sync into the CRM, for example, should factor that flow in from the start.

Why the Question Matters

On average, companies in Germany with 20 or more employees run 47 percent (Bitkom) of their IT applications from the cloud. Each of these applications brings its own interfaces, update cycles and data formats. The more systems interact, the more expensive each additional point-to-point connection becomes – and the more important the choice of the right integration layer.

What iPaaS Platforms Deliver

An integration platform as a service takes recurring tasks off your plate. Instead of building each interface from scratch, you pick a prebuilt connector, connect it with credentials and model the data flow in a graphical interface. Authentication, retries on error, logging and an operations dashboard are already included. For widespread systems such as a SAP integration or a DATEV integration, proven building blocks exist that considerably shorten the path to the first working connection.

The price of this convenience is standardization. iPaaS connectors cover the common fields and workflows; as soon as your processes deviate from the standard, you hit the toolkit's limits. Special fields, idiosyncratic pricing logic or multi-step approvals can sometimes only be modeled with workarounds. On top of that, your data flows through the vendor's platform – a point that must be checked carefully for personal data and the processing location. Anyone planning an omnichannel stock view with the POS should verify whether the toolkit even delivers the required real-time events.

Fast Start

Prebuilt connectors considerably shorten the path to the first working connection and lower the initial effort.

Connector Toolkit

Common systems can be connected and flexibly combined through configuration instead of programming.

Operations Included

Monitoring, retries and logging are part of the platform and do not have to be built yourself.

Limits on Special Logic

Deviating fields, complex pricing rules or multi-step approvals frequently exceed the standard toolkit.

Data Flow via Vendor

Payload data passes through a third party's platform – processing location and data processing must be checked.

Recurring License

Costs are ongoing and often grow with data volume and the number of connections.

When Custom Development Is the Better Choice

A tailored middleware pays off where the standard is not enough. That is the case when your business logic is idiosyncratic – for instance with customer-specific prices, complex bills of materials, batch and serial numbers or multi-step approvals. It also speaks for a dedicated solution when data sovereignty is decisive for legal or strategic reasons, because data and processing remain entirely within your environment. And finally when you have to connect a system for which no ready connector exists.

Custom development does not mean reinventing every wheel. Proven open building blocks for queues, event processing and logging form the foundation on which the individual connectors and transformations sit. The effort is higher than clicking together a toolkit, but the solution mirrors your processes exactly and can be extended step by step. Across our integration projects one pattern shows: the third system to connect usually arrives sooner than planned, which is why a robust architecture pays off early.

  • Your processes deviate clearly from the standard and can only be modeled in the toolkit with workarounds.
  • Data sovereignty and processing location are decisive for legal or strategic reasons.
  • You need to connect systems for which no ready connector exists.
  • The integration is business-critical and should remain operable independently of a license model.
  • High or strongly fluctuating data volumes make ongoing, volume-based license costs uneconomical.

Calculating Total Cost of Ownership Honestly

The sticker price is the smaller half of the truth. What matters is the total cost over the usage span – the total cost of ownership. With iPaaS, upfront costs are low because connectors and operations are ready; in return, license fees run permanently and often rise with data volume, the number of connections or executions. With custom development there are higher initial costs for design and build, followed by mostly plannable costs for operations and further development, without a usage-based license.

A Bitkom survey published in June 2026 shows that running costs do not stand still: for 64 percent (Bitkom) of the companies surveyed, operating costs for cloud solutions rose over the past year. For the calculation, that means a license belongs in the model with a projection, not with the price of the first contract year. The same applies to hosting, maintenance and adapting to new versions of the connected systems in a custom build.

An honest calculation accounts for both sides over several years. Whoever compares only the first month almost inevitably picks the toolkit; whoever factors in five years and growing volumes often sees the point at which the dedicated solution becomes cheaper. The calculation also includes soft factors: how quickly is the first version live, how expensive is an additional connection, how high is the maintenance effort? Our transparent pricing and a joint review of your system landscape provide a solid basis here instead of gut feeling.

Cost FactoriPaaS (Buy)Custom Development (Build)
Upfront CostLow, toolkit is readyHigher, design and build
Ongoing CostLicense, often volume-basedOperations, largely plannable
Additional ConnectionBook a connectorDevelop a connector
Special LogicWorkarounds or limitsDirectly modelable
Break-evenCheap for short usageCheap for long usage
DependenceVendor and licenseDocumentation and team

Worked Example: Five Years of iPaaS Versus Custom Build

The following example shows how to find the break-even point. All figures are assumptions for a store-ERP connection with three data flows (items and stock, orders, documents); they are neither prices of a platform vendor nor an offer from us. For iPaaS we assume 20,000 euros for setup, mapping and testing, a license of 12,000 euros in the first year that rises by 3,000 euros each year with growing volume, and 3,000 euros per year for internal upkeep of the configuration. For custom development we assume 60,000 euros for design, build and testing, followed by 9,000 euros per year for hosting, monitoring and maintenance.

Year (assumptions)iPaaS cumulativeCustom build cumulativeCustom build is
Year 135,000 euros69,000 euros34,000 euros higher
Year 253,000 euros78,000 euros25,000 euros higher
Year 374,000 euros87,000 euros13,000 euros higher
Year 498,000 euros96,000 euros2,000 euros lower
Year 5125,000 euros105,000 euros20,000 euros lower

The break-even point follows from a simple comparison: the custom build starts 40,000 euros more expensive but saves the difference in running costs every year. That difference is 6,000 euros in the first year and grows with the license to 9,000, 12,000 and 15,000 euros. After four years, a total of 42,000 euros has been saved, the head start is recovered, and after five years the custom build in this example sits 20,000 euros below the platform.

Just as telling is when the result flips. If volume stays flat and the license with it at 12,000 euros, iPaaS costs 95,000 euros over five years and remains cheaper than the custom build at 105,000 euros. If the custom middleware has to be rebuilt in year three for an assumed 15,000 euros because an ERP interface was deprecated, its five-year cost rises to 120,000 euros and the advantage shrinks to 5,000 euros; with iPaaS, such a rebuild is usually carried by the connector vendor. The decision therefore depends less on the starting price than on two questions: how fast does the volume grow that the license charges for, and how often do the interfaces change that the custom build must follow?

Data Sovereignty, GDPR and Lock-in

Data sovereignty co-determines the architecture. With custom development, payload data and processing logic remain in the infrastructure you choose; you decide on storage location, retention and access. With iPaaS, the data passes through a vendor's platform – frequently personal data from orders, customer accounts and invoices. The platform vendor is then usually a processor, and the controller may only use processors providing sufficient guarantees for appropriate technical and organizational measures (Art. 28(1) GDPR). Anyone mirroring customer data cleanly into SAP and a CRM should know the path of that data end to end.

Three points therefore belong before the selection, not after it. First, sub-processors: a processor may only engage another processor with prior specific or general written authorization of the controller (Art. 28(2) GDPR); for a platform that itself runs on third-party data centers and services, this list is often longer than expected. Second, third countries: a transfer of personal data to a third country is only permitted if the conditions of Chapter V are met (Art. 44 GDPR). Third, the end of the contract: the contract must provide that the processor, after the end of the services, deletes or returns all personal data at the choice of the controller (Art. 28(3)(g) GDPR). With custom development, the same questions arise toward the hosting provider and the operations partner, usually with fewer parties involved.

Lock-in Works in Both Directions

iPaaS ties you to a license model and a proprietary connector format; poorly documented custom development ties you to the developers who know it. Open standards, clean interfaces and a maintained process documentation noticeably lower the risk on both paths.

A pragmatic approach to lock-in is to limit it deliberately rather than ignore it. With iPaaS that means: secure export paths for configuration and history data, document data models and keep the dependence on individual proprietary functions low. With custom development it means: use standards instead of special routes, cut connectors modularly and document the solution so that it remains understandable in two years and to a different team. In both cases: whoever plans for the exit negotiates and works more freely.

What the Data Act Changes About Lock-in

Lock-in is not a side issue. In Bitkom's Cloud Report 2026, for which 603 companies (Bitkom) with 20 or more employees were surveyed, 59 percent (Bitkom) name lock-in effects such as difficult data exports or migrations as the biggest obstacle to switching cloud providers. The legal situation has shifted: since September 12, 2025 (Art. 50 Data Act), the EU Data Act has applied, and its Chapter VI governs switching between data processing services. According to the recitals, this covers infrastructure, platforms and software provided as a service (Recital 81 Data Act). A cloud integration platform can therefore be covered; whether a specific contract falls under it has to be checked case by case.

Four rules matter most for the choice between iPaaS and a custom build. The contract must state a maximum notice period for initiating the switch that may not exceed two months (Art. 25(2)(d) Data Act). Switching, or transferring all exportable data, must as a rule be possible within a mandatory transitional period of no more than 30 calendar days (Art. 25(2)(a) Data Act) after that. For data retrieval, a minimum period of 30 calendar days (Art. 25(2)(g) Data Act) applies after the transitional period. And from January 12, 2027 (Art. 29(1) Data Act), providers may no longer charge switching charges for the switching process; until then, only reduced charges are permitted (Art. 29(2) Data Act), capped at the costs directly linked to the switch (Art. 29(3) Data Act).

One exception is particularly telling for the choice between platform and custom build: the ban on switching charges and some technical obligations do not apply to data processing services where most main features have been custom-built to the needs of an individual customer and which are not offered at broad commercial scale via the provider's service catalog (Art. 31(1) Data Act). A tailored middleware that a service provider operates for you may fall precisely into this exception. The exit therefore belongs in the contract: rights to code and configuration, handover of documentation, support during the transition. Anyone who wants to go through the rules for ERP and store in detail will find them in the article on the EU Data Act and portable ERP data.

Time to Go-Live and Maintainability

Time to go-live – the time until the first productive connection – initially speaks for iPaaS. A ready connector is connected faster than one built from scratch. In early phases, where speed matters, that can tip the balance. Yet the first version is not the end: what matters is how well the solution can be maintained, extended and adapted to new requirements over years.

Here the picture shifts. If a connected system changes its interface, with iPaaS you wait on the vendor or catch up within the toolkit. With custom development you adjust the affected connector yourself – independent of a third party's roadmap. Conversely, with iPaaS the vendor takes over part of the maintenance that you carry yourself with a dedicated solution. Whoever compares both cleanly looks not only at the start but at the entire lifecycle of the connection.

Version Changes and Deprecations in the Connected Systems

The biggest maintenance item of an integration rarely arises in the integration itself, but in the systems it connects. Cloud ERP and store software change their interfaces on a schedule. One example: Business Central online receives two major releases every year, in April and October (Microsoft Learn), and a minor update in each of the remaining months (Microsoft Learn). Concrete breaks for integrations are announced in them: starting with version 30 in 2027 release wave 1, a Microsoft page can no longer be exposed as an OData endpoint (Microsoft Learn). Anyone who built a connection on such endpoints has to migrate it by then, whether it runs on a platform or in your own code.

The difference between the paths lies in who reads the announcement and who carries the rebuild. With iPaaS, the vendor maintains the connectors and you check your mappings after every connector update; with custom development, watching deprecations is part of operations. Two header fields help here: Deprecation tells a client that a resource will be or has been deprecated (RFC 9745 of the Internet Engineering Task Force), and Sunset names the point in time from which it is likely to stop responding (RFC 8594). If the other side sends these fields and the middleware writes them to monitoring as a warning, a deprecation shows up with the first call rather than the first error.

Monitoring, Alerting and Recovery per Path

Just as important as the build is what happens when a transfer fails. Every connection needs monitoring for failed and stuck messages, alerting that reaches the person in charge, and recovery without duplicates. The paths differ in who delivers this and how far it can be adapted to your processes.

AspectiPaaS (Buy)Custom Development (Build)
MonitoringPlatform dashboard, scope per planOwn metrics per data flow
AlertingPlatform rules and channelsOwn thresholds and recipients per error type
RetriesBuilt in, logic predefinedGrowing intervals by your own rule
Undeliverable messagesPlatform error store, check exportOwn store with reprocessing
Recovery after outageDepends on the connector, test for duplicatesIdempotent by design via unique keys
LogsRetention per platformSet retention periods and content yourself

What matters for recovery is that a repeated message does not create a second order or a duplicate booking. Unique keys per transaction achieve that, as described in the article on idempotency and retry strategies. Decoupling data flows through a queue lets you supply a failed recipient later instead of losing orders; the fundamentals are covered in the article on event-driven architecture with message queues. With iPaaS, a test before the selection pays off: switch off a recipient on purpose, let messages pile up and check what arrives once it is switched back on.

The Hybrid Approach: Combining Toolkit and Custom Build

In practice, the decision is rarely black and white. Many viable integration projects combine both paths: standardized, non-critical data flows run via iPaaS connectors, while business-critical or idiosyncratic processes emerge as tailored modules. That way you combine the short time to go-live of the toolkit with the precision and data sovereignty of custom development. The art lies in assigning each data flow to the right tool.

The better question is not 'iPaaS or custom development', but: which data flow needs which path – and what does it cost over the entire usage span?

A principle from integration practice

The assignment follows three criteria: how idiosyncratic is the process, how sensitive is the data, and how often does the interface change? Near-standard, non-critical flows with a stable interface are good candidates for the toolkit. Idiosyncratic, sensitive or frequently changing flows are better placed as a dedicated module. This dividing line can also shift: what starts today as an iPaaS flow can later migrate into a dedicated component when volume, cost or requirements suggest it.

Exit Strategy: What Has to Move With You

Switching between the paths happens more often than the first decision suggests: a company starts with iPaaS and later moves a critical flow into its own middleware, or it replaces an old custom build with platform connectors. Of the companies using cloud computing, 26 percent (Bitkom) have switched cloud providers once and 8 percent (Bitkom) several times, according to Bitkom. Whether a switch succeeds is decided by the list of things that have to move with you.

  • Mapping definitions: export field assignments, conversions and value lists in a readable format, not just as a screenshot of the interface.
  • Business rules: record special logic for prices, taxes, pack sizes or approvals as text and test cases.
  • History and logs: define what moves along for evidence and troubleshooting and what is deleted once its retention period ends.
  • Credentials and certificates: issue new ones for the new path and revoke the old ones after the move.
  • Open queues: drain them before switching over and resolve messages in the error store one by one.
  • Webhooks and schedules: point callback URLs in the store and ERP and scheduled syncs to the new path.
  • Mapping tables: carry over cross-references between store and ERP numbers, otherwise duplicates appear.
  • Monitoring and alerts: set up metrics, thresholds and recipients before the first productive flow moves.
  • Contractual side: clarify notice periods, return or deletion of personal data and the rights to code and documentation in advance.

The sequence is: run alongside, switch over, switch off. The new path first processes a data flow in shadow mode without writing to the target systems, and takes over only once its results match those of the old path. When switching off, the law helps: after the end of the services, the processor has to delete or return personal data at the choice of the controller (Art. 28(3)(g) GDPR), and for data processing services a minimum period of 30 calendar days for data retrieval applies (Art. 25(2)(g) Data Act).

A Five-Step Decision Guide

  1. Record systems and data flows: which applications should interact, which data flows where, and which system leads per data type?
  2. Classify processes: how many flows are near-standard, how many idiosyncratic or business-critical?
  3. Clarify data sovereignty requirements: where may personal data be processed, and which obligations apply?
  4. Calculate total cost over several years: compare upfront, operating and extension costs for iPaaS, custom build and hybrid.
  5. Set the path and plan for the exit: choose the tool per data flow, plan standards and documentation from the start.

Scoring Matrix: Rating Each Data Flow on Its Own

The five steps become tangible when each data flow is rated on four criteria. Each criterion scores zero to two points; two points speak for the custom build, zero for the toolkit. The matrix is a decision aid from our project practice, not a standard: it does not replace checking the specific interfaces, but it makes visible why a flow ends up on which path.

Criterion0 points1 point2 points
Process idiosyncrasyStandard fields, no special logicIndividual special fields or rulesOwn pricing, bill-of-materials or approval logic
Data sensitivityNo personal dataCustomer and order data in the usual scopeSpecial requirements, for instance on processing location
Change frequencyInterface stable for yearsPlanned version changesFrequent changes to the process or the other side
VolumeLow and steadyGrowing but plannableHigh or strongly fluctuating

The total sorts the result: up to two points, a ready connector is usually the more economical choice; from six points, much speaks for a dedicated module; in between, a close look at total cost and timeline pays off. The key is to fill in the matrix per data flow and not for the whole project. A project with five flows thus yields five results, and that is exactly where a hybrid setup often comes from.

Three Worked Examples

The following cases are constructed examples, not customer projects. A retailer with a standard range synchronizes items, stock and orders; a manufacturer sells configurable products with bills of materials, customer prices and approval levels; a wholesaler connects its store, ERP, accounting and a marketplace.

Example and data flowPoints per criterionTotalPath
Standard retail: items, stock, ordersIdiosyncrasy 0, sensitivity 1, change 0, volume 12Ready connector
Manufacturing: bills of materials, customer prices, approvalsIdiosyncrasy 2, sensitivity 1, change 2, volume 16Custom middleware
Hybrid: document export and marketplaceIdiosyncrasy 0, sensitivity 1, change 1, volume 02Platform connector
Hybrid: pricing and availability across several warehousesIdiosyncrasy 2, sensitivity 1, change 1, volume 26Dedicated module

In standard retail, a ready connector is usually the faster path and stays economical over five years as long as volume does not rise sharply; it is important to secure the export paths for mappings from the start. In manufacturing, bills of materials and approvals could only be modeled in the toolkit with workarounds, while a dedicated middleware models them directly and grows with the product range. Which data flows typically arise there is shown on our page on ERP integration in manufacturing. In the hybrid case, both parts meet at a shared queue with unified monitoring: the first version is live quickly, and the business-critical flows stay independent of the platform vendor's connector roadmap.

This article is based on data from: Bitkom, press release „Digitalisierung der Wirtschaft: Fast jedes Unternehmen beschäftigt sich mit KI“ of March 11, 2026 (survey of 604 companies with 20 or more employees) and press release „Deutsche Cloud: 4 von 10 Unternehmen würden Abstriche in Kauf nehmen“ of June 17, 2026 on the Cloud Report 2026 (survey of 603 companies with 20 or more employees); Regulation (EU) 2016/679 (GDPR), Articles 28 and 44; Regulation (EU) 2023/2854 (Data Act), Recital 81 and Articles 25, 29, 31 and 50; Microsoft Learn, pages on the update cycles and deprecated features of Business Central online; IETF, RFC 9745 (The Deprecation HTTP Response Header Field) and RFC 8594 (The Sunset HTTP Header Field); plus our own project experience and the worked example marked as assumptions.

Related Articles

Shop integration & processes

REST API or Middleware: Which Integration Architecture Fits

REST API or middleware for shop integration: advantages and disadvantages of both architecture models compared with concrete decision criteria.

11 min read
APIs, middleware & architecture

Event-Driven Architecture and Message Queues

How events and message queues enable decoupled, resilient shop integrations: retry, partitioned ordering, idempotency and the dead letter queue explained.

14 min read
Shop integration & processes

EDI Integration: Coupling EDIFACT to the Shop 2026

Couple EDIFACT messages such as ORDERS, INVOIC and DESADV between trading partners and your shop: converter, AS2, SFTP, mapping and receipts in practice.

14 min read