Build vs. buy

AI tooling exists in three tiers. They aren’t competing options — they’re stacked, and each one fits a different shape of work. Knowing where the work belongs is the difference between fast progress and a six-month build of something the team didn’t need.

The off-the-shelf tier

ChatGPT, Claude.ai, Gemini, Copilot, Notion AI, Cursor. Sign-up-and-use products. Mostly subscription-based, sometimes free.

What they’re for: individual productivity. Drafts, research, code, conversations. Anyone on the team can be using one within five minutes.

This tier fits when:

• The work is generic — drafts, research, code, brainstorming — and doesn’t need to know anything specific about the company’s data or workflow.
• The user is the one prompting and reading the output. Nothing runs unattended.
• The point is to learn what AI does for a given kind of work before deciding what to build around it.

This tier runs out when:

• The model needs to know things about the business that aren’t in its training — customer data, internal documents, pricing rules. (That kind of grounding is the job of tools and memory.)
• The work needs to be embedded in a system — triggered by an event, integrated with other tools, run unattended.
• Governance, audit logs, or access controls become non-negotiable. (Where your data goes covers what each surface implies.)

Most companies under-invest at this tier. Broad team access to ChatGPT or Claude.ai for everyday work is one of the highest-leverage moves available. Single-digit dollars per person per month, and the work changes — the slow drafts get faster, the research that used to take an afternoon takes twenty minutes. That alone often justifies the line item before any larger build is even on the table.

The vertical SaaS tier

Tools built for a specific job, with AI inside. Sales tools that handle outreach. Support tools that draft tickets. Marketing tools that generate campaigns. Recruiting platforms that screen CVs. Hundreds of them, more every week.

What they’re for: a specific repeated workflow, where the vendor has packaged the AI, the integrations, and the prompts into a usable product. The work that would take an internal team weeks of plumbing is — when the fit is real — already done.

A healthy vertical SaaS at this tier:

• Fits the workflow the team actually does, not a generic shape of it.
• Integrates with the systems the team already uses — CRM, helpdesk, doc store — without weeks of glue.
• Has its own product surface (not just an API key wrapped in a UI) and improves on it over time.

A thin vertical SaaS at this tier:

• Is a generic chatbot painted with industry vocabulary. Most of what it does, a ChatGPT Team account could do for a tenth of the price.
• Locks the workflow in. Data, prompts, history, integrations belong to the vendor. If the subscription ends, the work starts over.
• Ships glossy demos for the standard case and breaks at the edges where actual operations live.

The market at this tier is genuinely uneven. Some vertical SaaS is excellent and saves months. Some is a wrapper at premium pricing. The diagnostic isn’t the marketing — it’s whether the product does meaningfully more than a smart prompt on a generic tool would do.

The custom build tier

This tier uses the platform layer — APIs from OpenAI, Anthropic, Google, or self-hosted models on a cloud — combined with internal code, internal data, and an internal interface. The result is a system that exists nowhere else, because the workflow it serves exists nowhere else.

This tier fits when:

• The workflow is core to how the business operates and gives the company an edge in how it runs.
• The data and processes involved are specific enough that no off-the-shelf vendor would model them well — and probably never will.
• Ownership matters. The code, prompts, data, and integrations stay in-house and don’t depend on someone else’s roadmap.
• The volume justifies the build cost. A workflow that runs a hundred times a day for years pays back differently than one that runs ten times a quarter.

This tier doesn’t fit when:

• The workflow is generic and an off-the-shelf tool already handles it.
• The motivation is “we should be doing AI,” not “this specific work demands it.”
• No one inside the company is going to own the system after the project ends.

Custom build sits at the bottom of the stack: most flexible, most expensive, most demanding of internal capacity. It’s where the most distinctive work happens, and also where the most money gets burned when the work didn’t need to be there.

How the tiers stack

The three tiers aren’t a ladder where everyone climbs. They’re a stack where most companies should sit on multiple rungs at once — broad off-the-shelf access for the team, a few well-chosen vertical SaaS products for repeated workflows that fit a vendor’s shape, and custom build reserved for the small number of workflows that are genuinely core.

The two recurring mistakes:

Jumping to custom for generic work. A founder hears about AI, scopes a six-month build for a workflow ChatGPT could handle. Six months and a budget later, the team is using an internal tool that’s worse than the off-the-shelf product they could have rolled out in a week.

Buying premium vertical SaaS for work the team would shape better in-house. A leader signs a five-figure annual contract for an AI tool that turns out to be a wrapper. The team gets locked into a workflow they didn’t design and can’t change.

Neither mistake is rare. Both are expensive in the same way — months and budget spent at the wrong tier of the stack, on work that belonged a tier above or below.