JJamie Blair
← All posts
StrategyConsultingAIGuide

Build vs buy: when to use off-the-shelf AI and when to build custom

6 min read

A decision framework for choosing between off-the-shelf AI tools and custom-built solutions. Cost, control, time, and maintenance tradeoffs explained.

Monolithic AI core vs. raw components and tools, evaluating buying off-the-shelf or building custom AI solutions.

Quick answer

Buy first. Off-the-shelf AI tools work well for generic problems like ticket triage, email classification, and meeting transcription. Build custom only when you need full control over the model, your data is too sensitive for third parties, or no existing product fits your specific workflow.

Before building anything custom, the first question should always be: does a product already exist that solves this problem well enough?

Off-the-shelf AI tools have improved massively. Platforms like Zendesk, HubSpot, Notion, and dozens of others now include AI features baked into their products. The question isn't whether AI tools exist for your problem. It's whether the existing tools are good enough or whether your specific needs justify building something from scratch.

When to buy

The problem is generic

If your problem is one that thousands of other businesses have, a product probably already solves it. Support ticket triage, email categorisation, lead scoring, meeting transcription, document search. These are solved problems with mature products.

Examples of "just buy it" scenarios:

  • Meeting transcription: Otter, Fireflies, or the built-in features of Zoom/Teams
  • Email spam filtering: Already built into every email platform
  • Basic chatbot: Intercom, Drift, or Zendesk's AI features
  • Document search: Notion AI, Google Workspace AI, Microsoft Copilot
  • Social media scheduling: Buffer, Hootsuite (with AI-assisted content suggestions)

You need it fast

Off-the-shelf tools are available now. A custom build takes weeks or months. If the need is urgent and a 70% solution is better than no solution for the next two months, buy first and evaluate whether to build later.

You don't have proprietary data

Custom AI solutions shine when they're trained on your data: your ticket history, your customer patterns, your operational quirks. If the task doesn't depend on data unique to your business, a general-purpose tool usually performs fine.

Maintenance isn't your thing

Bought tools are maintained by the vendor. Updates, bug fixes, infrastructure. Custom tools are maintained by you (or the developer you hire). If you don't have the budget or inclination for ongoing maintenance, buying is lower risk.

When to build

The problem is specific to your business

No off-the-shelf tool handles "match delivery reports from these three specific courier APIs against invoices from these suppliers in this specific format with these business rules." That's your problem, shaped by your systems and your processes. Custom build.

You need control over accuracy

Off-the-shelf tools optimise for general performance across all their customers. A custom model trained on your data can be tuned for your specific accuracy requirements, your specific edge cases, and your specific failure modes.

Integration is complex

If the AI needs to connect to multiple internal systems, apply custom business logic, or fit into an existing workflow that doesn't match any standard product's assumptions, custom builds are often simpler in the long run than trying to force a product to do something it wasn't designed for.

Data privacy is critical

Some off-the-shelf AI tools process your data on their servers. For businesses handling sensitive financial, medical, or personal data, running models locally or on your own infrastructure may be a regulatory requirement. Custom builds give you full control over where data lives.

The economics work at your scale

If you're processing high volumes, the per-unit pricing of SaaS tools can exceed the cost of running your own solution. A document processing API that charges £0.01 per page is fine for 100 pages a month (£1) but expensive for 100,000 pages a month (£1,000). At high volume, building your own is often cheaper.

Note

A common middle ground: buy an off-the-shelf tool to validate the approach, then build custom once you've proven the value and understand the requirements better. This de-risks the custom build because you already know the solution works.

The comparison framework

| Factor | Buy | Build | |---|---|---| | Time to value | Days/weeks | Weeks/months | | Upfront cost | Low (subscription) | Higher (development) | | Ongoing cost | Recurring subscription | Hosting + maintenance | | Customisation | Limited to product features | Full control | | Data privacy | Vendor's servers | Your servers | | Maintenance | Vendor handles it | You handle it | | Accuracy on your data | Generic (good enough?) | Tuned to your data | | Switching cost | Moderate (data export) | High (rebuild) |

The hybrid approach

In practice, most businesses end up with a mix. You might use Zendesk for ticketing with their built-in AI triage, but build a custom integration that enriches tickets with data from your internal systems. Or use Xero for accounting but build a custom reconciliation pipeline that handles the complex matching logic.

The key is knowing where the off-the-shelf tool ends and where custom work begins. Usually it's at the point where your specific data, your specific business rules, or your specific integration needs don't match the product's assumptions.

Questions to ask before deciding

  1. Does a product exist that does at least 70% of what I need? If yes, start there.
  2. Is the remaining 30% critical or nice-to-have? If nice-to-have, the product is probably enough.
  3. How much do I spend on this process annually? If it's under £10,000/year, buying is almost certainly cheaper than building.
  4. Will my requirements evolve significantly? If yes, custom gives you room to adapt. Products evolve on their roadmap, not yours.
  5. Do I have budget for ongoing maintenance? If no, don't build custom. You'll end up with unmaintained software that breaks at the worst time.

Key Takeaways

  • Buy first for generic problems, fast timelines, and low maintenance tolerance.
  • Build when the problem is specific to your data, your systems, or your business rules.
  • The hybrid approach usually wins: off-the-shelf for the core, custom for the edges.
  • Factor in maintenance cost, not just build cost. Custom AI needs ongoing care.
  • Validate with a bought tool first, then build custom once you've proven the value.

Need help deciding?

Build-vs-buy is one of the first things I work through in an AI audit. If you're weighing up your options, get in touch and I'll give you a straight answer.

Related reading: