Search

What are you looking for?

Search our services, use cases and practical insights.

Enter at least 2 characters

Popular starting points

AI decisions

Should you buy, integrate or build an AI solution?

The decision is rarely a pure either-or choice. Many sustainable solutions buy standard foundations, integrate them with existing systems and build only the business-specific parts.

The short answer

Buying makes sense when a market product covers most of a standard process. Integration connects existing models or platform services with the organisation's data and workflows. Custom development is worthwhile for differentiating requirements that standard products cannot meet appropriately and that the organisation can operate over time.

In brief

  • The decision should start with user needs, not with a preference for a particular technology.
  • Buying reduces development effort but does not remove integration, data assurance or internal accountability.
  • Building does not necessarily mean training a model; it often means creating an application around existing models.
  • A modular architecture can use standard components while keeping business-critical logic under organisational control.

What do buy, integrate and build mean in practice?

Buying centres on a finished application with an interface, functions and an operating model. Integration uses existing models, cloud services or platform components and connects them to identity, data and processes. Custom development gives the organisation control of the application and business logic while the underlying model may still come from a provider.

In practice, the result is often a combination. Standard capabilities such as model access, authentication or document processing are sourced, while the specific workflow, quality rules and integrations are implemented for the organisation.

What do buy, integrate and build mean in practice?
RouteTime to valueAdaptability and skillsLifecycle and lock-in
BuyOften weeks when the process is largely standardBounded by the product; internal expertise focuses on adoption and the domain processProvider runs the core, but data export, price changes and product dependency remain
IntegrateUsually several weeks to monthsOwn data, roles and workflows wrap existing services; integration skills are requiredThe organisation owns connections and tests; lock-in depends on APIs and portable business logic
BuildUsually several months to dependable productionHighest freedom; product, AI, security and operating skills are requiredFull responsibility for change and support; lower product but possible technology lock-in
HybridFaster than a full build and more targeted than a pure productSource standard parts while retaining control of differentiating logic and testsKeep components replaceable and document ownership by layer

Which criteria determine the right route?

Do not assess only a demo's feature list. The real user need, necessary adaptation, data and system integration, regulatory constraints and the ability to operate and change the solution over several years are what matter.

  1. Step 1

    Standardisation

    How closely does the process match mature market solutions, and which deviations genuinely matter to the business?

  2. Step 2

    Differentiation

    Does the specific logic, data foundation or user experience create competitive value?

  3. Step 3

    Integration

    Which identities, data sources, business systems and approvals need to be connected?

  4. Step 4

    Operational capability

    Who can own quality, security, change, support and cost over time?

  5. Step 5

    Portability

    Can data, configuration, tests and interfaces be reused when a provider changes?

Which work remains with the organisation in every option?

Even a purchased product must be introduced into the business. Data classification, roles, quality criteria, training and integration into working practices cannot be outsourced entirely. Conversely, a custom solution should not recreate functions already available as mature standard services.

  • Define purpose, permitted use and accountable roles
  • Review data flows, contracts and access rights
  • Test quality on internal cases and monitor change
  • Support users and maintain incident and exit procedures

How can the decision be validated?

Compare two or three realistic solution paths using the same demanding cases. The test should show not merely that a function exists, but how well it integrates, can be controlled and can be operated. An architecture decision record should preserve assumptions, evidence and accepted trade-offs.

  • Test one small but difficult representative process
  • Estimate total cost and internal effort over several years
  • Assess export, termination, model change and outage scenarios
  • Reconfirm the decision after the pilot and before a larger commitment
Example from day-to-day business

Example: knowledge assistance for customer service

A finished assistant covers general search and chat but cannot represent complex product permissions. Instead of building everything, the company uses a managed model and search service. It develops the permission logic, source display and ticket-system integration itself. Model access and infrastructure remain replaceable components; business-specific logic and tests remain under its control.

What to remember

Buy mature standard capabilities, integrate them in a modular way and build only where proprietary requirements create demonstrable value and the organisation can carry the lifecycle.

Sources and further reading

These primary sources provide further detail on definitions, technical foundations or responsible use.

Content reviewed

Would you like to apply this to your situation?

Together, we clarify what makes sense for your process, data and systems – in plain language and without unnecessary complexity.

Discuss Your Project