Build vs Buy vs Orchestrate: Choosing an AI Stack for an Operating Business
Article Hossein Narimani November 11, 2026 AI & Intelligent Systems

Build vs Buy vs Orchestrate: Choosing an AI Stack for an Operating Business

There is a third option, and it's the one most businesses should pick. The build-or-buy debate is an old frame designed for enterprise software. For an AI stack the dichotomy misleads, because it ignores which layer...

There is a third option, and it's the one most businesses should pick.

The build-or-buy debate is an old frame designed for enterprise software. For an AI stack the dichotomy misleads, because it ignores which layer the value actually sits in.

Three strategies in one sentence each

The third option is usually absent from the discussion, and it is exactly where competitive advantage lives.

Side by side

BuildBuyOrchestrate
Cost shapeHeavy capital, light operatingFixed, predictable operatingModerate capital, variable operating
Time to value3 to 9 monthsWeeks4 to 10 weeks
Lock-inNoneHigh — in data and in processLow per component, none in the decision layer
CeilingNoneThe vendor's roadmapThe best available component
Who maintains itYou, foreverThe vendorYou, only the thin connecting layer
Ownership of differentiationCompleteNoneComplete, where it matters

Read rows three and six together. Buying removes the maintenance cost and removes differentiation with it — your competitor can buy the same product and have exactly the same capability.

The orchestrate path: buy components, own the decision layer

The decision layer is the part of the system that determines what should happen in each situation, who owns it, and when it escalates to a human.

That layer is almost always specific to your business. Two clinics of identical size have different follow-up rules. Two services firms have different project acceptance criteria.

Components that should nearly always be bought: the language model, the database, authentication, message delivery, and standard interface. None of these is a source of differentiation and building them is waste.

What you write yourself is usually small — a few thousand lines, not a few hundred thousand. But those few thousand lines are the difference between your system and a competitor's.

Four questions that choose the answer for you

  1. Is this process your competitive advantage? If not, buy. Do not build accounting or payroll.
  2. Does a product cover more than 80 percent of the need? If yes and the remaining 20 percent is not critical, buy. If that 20 percent is your advantage, orchestrate.
  3. Where does the data have to live? Legal or contractual constraints frequently eliminate buying outright.
  4. Do you have long-term maintenance capacity? Building something you cannot maintain for three years is the worst available option.

Question four carries the most weight and gets the least honesty. A system one person built and that person has left is effectively bought — from a vendor that no longer supports it.

The hidden cost of all three

None of these three invoices shows the real cost.

Build: the hidden cost is maintenance, and not merely bug fixes. Every dependency needs updating, every external API change generates work, and every person who leaves takes context with them. A rule of thumb: roughly 20 percent of the original build cost, annually.

Buy: the hidden cost is process adaptation. Your team reshapes its process to fit the product, and that change is almost never audited. Some of those adaptations are harmless and some remove your advantage.

Orchestrate: the hidden cost is boundaries. Every additional component is a joint that can break, and debugging across five components is harder than debugging one. That cost is real and usually smaller than the other two.

The mistake that ruins all three strategies

Independent of your choice, one error spoils the outcome: deciding before writing the process down.

If you do not know what your process is, buying means accepting the vendor's without knowing what you gave up. Building means encoding ambiguity. And orchestrating means writing a decision layer that reflects no decision.

The order is always the same: write the process, decide which layer is genuinely yours, and only then talk about tools.

A worked example: a purchase that became orchestration

A mid-sized services company bought a platform for managing customer requests. It was a good product and went live in six weeks.

The problem appeared at month four. The product's prioritisation logic worked on submission time. This company's real logic combined contract type, customer history, and team capacity — none of which the product could express.

The fix was not a rewrite. They kept the platform as the record layer and the interface, and wrote a small service that computed priority and handed it back to the platform.

That service was about 900 lines and took three weeks. What it did was return the one part of the system genuinely specific to this company to their own ownership.

That is the common orchestration pattern: buy everything, write only the rule that is yours.

An expensive mistake in the evaluation

When comparing, almost everyone weighs build cost against a product's subscription price. That comparison omits two things.

First, time to value. Three months of delay in launching a revenue-bearing capability carries a cost that usually exceeds the entire price difference and appears on no spreadsheet.

Second, probability of failure. A finished product works with near-zero risk; an internal build has a meaningful chance of not reaching the intended outcome. A fair comparison divides build cost by probability of success rather than assuming certainty.

Migrating between the three without a rewrite

Today's decision should not be permanent, and designed properly it is not.

One rule keeps the next migration available almost always: never write decision logic inside the tool.

Every platform lets you define rules within it, and the temptation is real. But a rule living in a product's configuration is locked to that product. The same rule in a small external service survives a tool change untouched.

The usual migration path is predictable too: buy in year one for speed, orchestrate in year two once the constraints surface, and build only for components with no acceptable option.

Isn't orchestration more expensive than buying?

In year one, usually yes. Over three years, usually no, because the main cost of buying appears in its constraints rather than its invoice.

When is building from scratch right?

When the capability is itself your product, or when no acceptable component exists. For an operating business, that case is rare.

How do we avoid data lock-in?

Test the full export before signing — not that it is mentioned in the contract, but actually run an export once and see what comes back.

Closing

The right question is not build or buy. It is which layer is genuinely yours — and that layer is almost always smaller than you think and more important than what you spend your time on.

If you want that boundary drawn for your own stack, that is what intelligent business systems does.

Ready to apply this in your own product? Book a Strategy Call and get a clear roadmap for your next sprint.

Comments (0)

Be the first to leave a comment.
Login / Sign up