Blog · AI for small and midsize businesses
Buy, configure or build: choosing an AI solution for an SME
The right delivery approach depends on the workflow and who will maintain it. Buying a product, configuring an assistant and building software each create different obligations. Compare them against one business test and define when an option should be rejected before the project grows.

In this article
Buying AI software can be fast when its workflow already fits. Configuring an assistant can suit a bounded knowledge task. Custom development can address specific integrations, but creates a larger maintenance commitment. Choose the approach after defining the task, required controls and acceptable result.
Compare the work each option leaves behind
| Approach | Best fit to investigate | Reject or reconsider when |
|---|---|---|
| Packaged software | A common workflow with features the team can test directly. | Essential permissions, records or export needs cannot be met. |
| Configured assistant | A bounded task using approved sources and manageable review. | Required workflow control or integration exceeds what configuration can reliably enforce. |
| Custom application | A specific process needs controlled integration and tailored interaction. | The business cannot fund ongoing ownership or the task is still undefined. |
These are decision criteria, not claims about every product. Confirm actual capabilities with documentation and a test. A product advertised as configurable may still require custom integration; a custom interface may rely on managed platform services underneath.
Use the same test cases
Give each approach identical representative inputs, difficult cases and acceptance criteria. Include permissions, missing information and the manual fallback. Compare complete handling time, quality, operating effort and what the business receives at exit. A prototype's attractive interface should not substitute for evidence.
In a hypothetical professional-services firm, the task is to draft answers from approved internal procedures. Existing software may already provide adequate search and review. A configured assistant may improve question handling without a new application. Custom development becomes worth investigating only if the firm needs capabilities such as a controlled connection to its internal case system that simpler options cannot meet.
Account for ownership after setup
For packaged software, someone still manages users, approved use and updates to business information. For a configured assistant, assign source maintenance, instructions and regression testing. For custom software, also assign code maintenance, dependency updates, monitoring and operational support.
Ask where the authoritative information lives and how it can be exported. Confirm what happens if the supplier changes a feature or the company changes providers. Have the appropriate adviser review contractual claims; operationally, test export and handover rather than accepting a broad promise of portability.
Make the decision reversible where possible
Start with a limited scope and record the conditions for continuing. A configured pilot should not become a critical system without support and access review. Likewise, avoid building a custom application merely because configuration revealed a process rule nobody has resolved.
Deloitte's framework connects suitability with complexity and available capabilities. The NIST generative AI profile supports evaluating operating conditions throughout the lifecycle. Apply those principles to the actual alternatives, without assuming one approach is always safer or cheaper.
Write a short decision record: selected approach, evidence, exclusions, accountable owner and exit conditions. Bring it to an AI solution assessment if you need an independent review of the tradeoffs.