Build vs buy vs wrap — the AI decision matrix
Buy when a vendor SaaS solves your specific problem well (no customization needed beyond configuration); wrap commercial LLM APIs when the workflow is unique to your business but the AI capability is general; build from scratch when the use case requires deep integration, custom workflows, or proprietary data that vendors cannot accommodate. Most business AI in 2026 is wrap, not build — the heavy infrastructure investment of build is rarely justified outside specific compliance or capability scenarios.
The longer answer
The build / buy / wrap decision matrix for AI is one of the most important early calls in any business-AI initiative, and the right answer is usually less ambitious than buyers initially assume.
Buy: vendor SaaS
The right answer when a focused vendor SaaS solves your specific problem well. Examples: Intercom Fin for customer-support AI; Gong for sales-call analysis; Harvey for legal research; Glean for enterprise search. Pros: lowest engineering cost, fastest time-to-value, vendor handles model upgrades and operational burden. Cons: limited customization, dependency on vendor pricing and roadmap, data flows to the vendor. The honest check: does the vendor's product actually solve your problem at 80% or higher, or are you trying to force-fit a tool that does not match your workflow?
Wrap: commercial LLM APIs with custom integration
The right answer when the workflow is specific to your business but the AI capability is general (text generation, summarization, classification, conversational interaction). Build a thin custom layer on top of Anthropic Claude or OpenAI APIs with your prompt templates, your data flows, your authentication, your observability. Pros: fast to build (4-12 weeks for most use cases), capability is current-frontier, low operational burden. Cons: per-token cost scales with usage; you are dependent on the API provider's pricing and capability roadmap. This is the dominant pattern in 2026 and where most production business AI lives.
Build: from-scratch model + infrastructure
The right answer in three specific scenarios. Compliance. When commercial APIs cannot satisfy your compliance posture (HIPAA-strict, FedRAMP, data residency). Capability gap. When the task requires fine-tuning on proprietary data that commercial models cannot capture through prompting alone (some legal, some scientific, some highly-specialized industry domains). Volume / cost. Genuinely high sustained volume (100M+ queries/month) where the API math no longer works. Build is the most expensive option (12-24 weeks minimum, plus ongoing infrastructure cost), and the right answer for fewer scenarios than buyers initially think.
The decision framework
Ask in order: (1) Does a vendor SaaS solve this at 80%+? If yes, buy. (2) Is the workflow specific to my business but the AI capability general? If yes, wrap. (3) Do I have a compliance, capability, or volume reason that wrap cannot satisfy? If yes, build. Most buyers stop at step 2; the ones who go to step 3 should be sure the reason is real.
Common follow-up questions
Can I start with buy and move to wrap later?
Often the right pattern. Use a vendor SaaS to validate that the AI use case actually delivers value, then build a custom wrap if the vendor product is the wrong fit at scale. The lesson is usually that wrap was right earlier than you thought.
Should I worry about vendor lock-in if I wrap?
Moderately. Most well-built wrap architectures abstract the LLM provider behind an internal interface, so switching from OpenAI to Anthropic (or vice versa) is a 1-2 day engineering exercise. The lock-in to a specific provider is less than to a SaaS vendor.
When should a small business build from scratch?
Almost never. Build is the right answer for buyers with specific compliance constraints or genuinely unique capability requirements. For mid-market businesses without those constraints, wrap is the right answer 90%+ of the time.
If this answer is useful and you have a real engagement in mind, the contact form routes directly to the principal — James Henderson is the single engineer who scopes, writes, and supports every engagement end-to-end.