How to advise a client on building versus buying AI

The honest answer is rarely build, and the harder honesty is the conflict of advising on a tool you also resell.

Technology
By Mark Choudhari · Jun 7, 2026 · 6 min read

Buy the commodity layer. Own the judgment inside it.
Made with Works

TL;DR

When a client asks whether to build, buy, or rent AI, the honest answer is rarely build. Building carries a five to six figure tail and 12 to 24 months to value, and most use cases do not justify it. Buy the commodity intelligence layer, reserve building for genuine data advantage, and name the conflict openly if you also resell the tool. The resolution is configured expertise: your judgment inside a layer you did not have to build.

In this article

What are the real costs and ROI timelines of building versus buying AI for a client

Building is the expensive, slow option, and the numbers are not close. The fixed tail on a custom build runs $50,000 to $100,000 for an MVP and $250,000 to $400,000 and up for a multi-agent system, with open-weight models running 10 to 12 times cheaper than frontier SaaS and the cost crossover where building pays off sitting near a million conversations a year. Below that volume, buying is simply cheaper.

Time-to-value widens the gap. One total-cost analysis puts building at 12 to 18 months to value against 2 to 4 months to buy, with upfront comparisons missing 60 to 80 percent of the real cost. A worked ROI case sharpens it further: a document-automation use case returned 93 percent ROI in year one when bought at around $280,000, while the same thing built carried a negative year-one return near $750,000, only catching up years later if the custom version added substantial yearly value. For most client use cases, build loses on cost and on time.

A document-automation use case returned 93 percent ROI in year one when bought, against a negative year-one return when built.
Just Think AI, Build vs Buy for Enterprise AI in 2026

What is the build, buy, or rent decision for AI, and which fits which client

Build, buy, or rent is the choice between commissioning custom AI, licensing a finished platform, or running a layer someone else builds and maintains. The decision turns on a few dimensions, and the canonical treatment of the framework lives in the agency retainer breakdown and one place, not eight tools; this piece applies it from the agency’s seat as an advisor, not from a founder deciding for their own business.

The dimensions that decide it are differentiation, data advantage, and total cost over three to five years rather than year one. The honest default most analyses reach is a blend: buy or rent the commodity intelligence layer, build only the client-specific data and routing on top, and reserve a full custom build for genuine, data-advantaged differentiation. The cleaner test is whether the capability is a core differentiator for the client or a supporting utility; supporting utilities should be bought, and only true differentiators justify building, with maintenance on anything custom running 15 to 20 percent of build cost a year.

When should you tell a client to build instead of buy

Tell a client to build only when the AI itself is the differentiator and they own data nobody else has. That is the narrow case where a custom build earns its tail: a genuine data advantage, a use case core to how the client competes, and enough volume to cross the cost line. Outside that case, building is a five to six figure commitment to a slower, more fragile path.

For everything else, the answer is buy or rent, and reserve the build budget for the evaluation work, not a team. The practical version of this is to build the evaluation harness, not the headcount: spend on knowing whether a bought layer works for the client, not on constructing one from scratch. Most client use cases are supporting utilities, which means most honest recommendations land on buy.

Is reselling a platform the same as renting AI, and what is the lock-in

Reselling a platform is one form of rent, and the lock-in is real but manageable if you name it. The central risk in any buy-or-rent path is lock-in at the AI layer: operational and run costs surpass development cost within 18 to 24 months, breakeven on a custom build sits near 33 months, and vendor lock-in at the AI layer is the central buy-side risk to weigh against the speed and cost advantages of buying.

The way to make rent safe is to keep the client’s data and the agency’s judgment portable, so the layer can change underneath without resetting what was built on top. The full treatment of reselling-as-rent and its lock-in lives in white-label partner economics. The point for the advisor is that rent is the right answer for most use cases as long as the lock-in is named and the portability is designed in, not assumed.

How do you advise a client honestly on build, buy, or rent when you also resell the tool

You name the conflict first, then resolve it by making your judgment the product rather than the markup. The conflict is real: if you recommend buy and you also resell the tool, you can look like you are pushing your own product. Pretending the conflict away is worse than stating it. The honest move is to say it out loud, show the client the same build, buy, or rent math you would want shown to you, and let the recommendation stand on the numbers.

The resolution is configured expertise. When what you sell is your judgment built into the layer the client runs on, rather than a margin on a license, the recommendation and the reselling stop being in tension. The client is buying your expertise configured inside a tool you did not have to build, which is exactly the move configured with your expertise owns. This matters more now because the advisory relationship is under pressure: 60 percent of senior marketing leaders spent less on agencies in 2025 because of AI, and 82 percent of ANA members now run an in-house agency. Honest advice is the retention play.

How to recommend Buy without betting your margin on a tool you do not control

The bar any answer has to clear is this: recommend the Buy or Rent layer that ends the per-client rebuild, not one you will reconstruct for the next client. A custom build is a rebuild you maintain forever, the cost the agency AI tax already priced. So the tool worth recommending is the configured layer that runs across clients and carries your expertise inside it.

That is where Works sits in this decision. It is the Buy or Rent option built so the agency owns the judgment and the commodity intelligence underneath is bought, not built: 500-plus expert-grade workflows the client runs on, the apps they already use connected through 3,000-plus integrations, and new models absorbed inside the existing setup so nothing resets when the model pool changes. At $49 a month for the Pro tier against a $50,000 to $500,000-plus custom build, the economics of Buy stop being a compromise. Machintel runs Works across six teams rather than building per team; an agency recommending Buy is recommending the pattern Machintel itself chose.

Recommend Buy, not Build. Get early access, or DM us for the agency economics breakdown that shows the build, buy, and rent math side by side.

Advise the client the way you would want to be advised, then make sure the tool you point them to is not one you will rebuild for the next client. Buy the commodity intelligence layer. Own the judgment inside it.

Common Questions

When should I tell a client to build vs buy vs rent AI?

Recommend build only when the AI is a genuine differentiator and the client owns unique data, because that is the narrow case where a $50,000-plus tail and a 12-to-24-month timeline pay off. For supporting utilities, which is most use cases, recommend buy or rent the commodity layer and reserve any build budget for evaluation, not a team. The framework that decides it lives in the agency retainer breakdown.

What are the real costs and timelines of building versus buying?

Building runs $50,000 to $100,000 for an MVP and $250,000 to $400,000 and up for a multi-agent system, at 12 to 18 months to value, with upfront estimates missing 60 to 80 percent of true cost. Buying reaches value in 2 to 4 months and wins on year-one ROI for most use cases. Run costs overtake build cost within 18 to 24 months, and breakeven on a custom build sits near 33 months.

Should my agency build its own tool or run someone else’s?

For most agencies, run someone else’s configured layer and put your expertise inside it, rather than building and maintaining a tool. Building your own is the same per-client rebuild tax applied to your own business, and it commits you to maintenance forever. The cost case is in the agency AI tax.

How do I advise honestly on a tool I also resell?

State the conflict directly rather than burying it: you resell the platform you are recommending, and the client deserves to know. Then show the same build, buy, or rent math you would accept as a client yourself. The conflict resolves when your configured expertise is the product, not the margin on a license, because a client paying for your methodology configured inside the layer has a different value proposition than a client paying markup. The full resolution is in configured with your expertise.

Is reselling a platform the same as renting, and what about lock-in?

Reselling is a form of rent, and the lock-in risk is real, at the AI layer especially. Make rent safe by keeping the client’s data and your judgment portable so the layer can change underneath without a reset. The full treatment of reselling-as-rent and its lock-in is in white-label partner economics.

Get Started With AI

Are You Ready to Make AI Work for You?

Simplify your AI journey with solutions that integrate seamlessly, empower your teams, and deliver real results. Jyn turns complexity into a clear path to success.

See AI for Real Business Impact in Action →

ai that powers your team 226d8ee5db
Recent Posts
Two identical stacks of coins on a balance sheet, one dissolving into vapor, the other growing taller across three columns.
Is Your AI Spend an Asset or an Expense
A stone foundation being laid course by course, each new layer sitting level on the one below it, in a bright open field.
Is Your AI Setup Getting Better, or Just Getting Older
A crowded desk of a dozen glowing app icons on the left resolving into a single clean workspace on the right, the same work, fewer objects.
How Your AI Stack Can Get Simpler as Your Business Grows

How to advise a client on building versus buying AI

The honest answer is rarely build, and the harder honesty is the conflict of advising on a tool you also resell.

Technology
By Mark Choudhari · Jun 7, 2026 · 6 min read

Buy the commodity layer. Own the judgment inside it.
Made with Works

TL;DR

When a client asks whether to build, buy, or rent AI, the honest answer is rarely build. Building carries a five to six figure tail and 12 to 24 months to value, and most use cases do not justify it. Buy the commodity intelligence layer, reserve building for genuine data advantage, and name the conflict openly if you also resell the tool. The resolution is configured expertise: your judgment inside a layer you did not have to build.

In this article

What are the real costs and ROI timelines of building versus buying AI for a client

Building is the expensive, slow option, and the numbers are not close. The fixed tail on a custom build runs $50,000 to $100,000 for an MVP and $250,000 to $400,000 and up for a multi-agent system, with open-weight models running 10 to 12 times cheaper than frontier SaaS and the cost crossover where building pays off sitting near a million conversations a year. Below that volume, buying is simply cheaper.

Time-to-value widens the gap. One total-cost analysis puts building at 12 to 18 months to value against 2 to 4 months to buy, with upfront comparisons missing 60 to 80 percent of the real cost. A worked ROI case sharpens it further: a document-automation use case returned 93 percent ROI in year one when bought at around $280,000, while the same thing built carried a negative year-one return near $750,000, only catching up years later if the custom version added substantial yearly value. For most client use cases, build loses on cost and on time.

A document-automation use case returned 93 percent ROI in year one when bought, against a negative year-one return when built.
Just Think AI, Build vs Buy for Enterprise AI in 2026

What is the build, buy, or rent decision for AI, and which fits which client

Build, buy, or rent is the choice between commissioning custom AI, licensing a finished platform, or running a layer someone else builds and maintains. The decision turns on a few dimensions, and the canonical treatment of the framework lives in the agency retainer breakdown and one place, not eight tools; this piece applies it from the agency’s seat as an advisor, not from a founder deciding for their own business.

The dimensions that decide it are differentiation, data advantage, and total cost over three to five years rather than year one. The honest default most analyses reach is a blend: buy or rent the commodity intelligence layer, build only the client-specific data and routing on top, and reserve a full custom build for genuine, data-advantaged differentiation. The cleaner test is whether the capability is a core differentiator for the client or a supporting utility; supporting utilities should be bought, and only true differentiators justify building, with maintenance on anything custom running 15 to 20 percent of build cost a year.

When should you tell a client to build instead of buy

Tell a client to build only when the AI itself is the differentiator and they own data nobody else has. That is the narrow case where a custom build earns its tail: a genuine data advantage, a use case core to how the client competes, and enough volume to cross the cost line. Outside that case, building is a five to six figure commitment to a slower, more fragile path.

For everything else, the answer is buy or rent, and reserve the build budget for the evaluation work, not a team. The practical version of this is to build the evaluation harness, not the headcount: spend on knowing whether a bought layer works for the client, not on constructing one from scratch. Most client use cases are supporting utilities, which means most honest recommendations land on buy.

Is reselling a platform the same as renting AI, and what is the lock-in

Reselling a platform is one form of rent, and the lock-in is real but manageable if you name it. The central risk in any buy-or-rent path is lock-in at the AI layer: operational and run costs surpass development cost within 18 to 24 months, breakeven on a custom build sits near 33 months, and vendor lock-in at the AI layer is the central buy-side risk to weigh against the speed and cost advantages of buying.

The way to make rent safe is to keep the client’s data and the agency’s judgment portable, so the layer can change underneath without resetting what was built on top. The full treatment of reselling-as-rent and its lock-in lives in white-label partner economics. The point for the advisor is that rent is the right answer for most use cases as long as the lock-in is named and the portability is designed in, not assumed.

How do you advise a client honestly on build, buy, or rent when you also resell the tool

You name the conflict first, then resolve it by making your judgment the product rather than the markup. The conflict is real: if you recommend buy and you also resell the tool, you can look like you are pushing your own product. Pretending the conflict away is worse than stating it. The honest move is to say it out loud, show the client the same build, buy, or rent math you would want shown to you, and let the recommendation stand on the numbers.

The resolution is configured expertise. When what you sell is your judgment built into the layer the client runs on, rather than a margin on a license, the recommendation and the reselling stop being in tension. The client is buying your expertise configured inside a tool you did not have to build, which is exactly the move configured with your expertise owns. This matters more now because the advisory relationship is under pressure: 60 percent of senior marketing leaders spent less on agencies in 2025 because of AI, and 82 percent of ANA members now run an in-house agency. Honest advice is the retention play.

How to recommend Buy without betting your margin on a tool you do not control

The bar any answer has to clear is this: recommend the Buy or Rent layer that ends the per-client rebuild, not one you will reconstruct for the next client. A custom build is a rebuild you maintain forever, the cost the agency AI tax already priced. So the tool worth recommending is the configured layer that runs across clients and carries your expertise inside it.

That is where Works sits in this decision. It is the Buy or Rent option built so the agency owns the judgment and the commodity intelligence underneath is bought, not built: 500-plus expert-grade workflows the client runs on, the apps they already use connected through 3,000-plus integrations, and new models absorbed inside the existing setup so nothing resets when the model pool changes. At $49 a month for the Pro tier against a $50,000 to $500,000-plus custom build, the economics of Buy stop being a compromise. Machintel runs Works across six teams rather than building per team; an agency recommending Buy is recommending the pattern Machintel itself chose.

Recommend Buy, not Build. Get early access, or DM us for the agency economics breakdown that shows the build, buy, and rent math side by side.

Advise the client the way you would want to be advised, then make sure the tool you point them to is not one you will rebuild for the next client. Buy the commodity intelligence layer. Own the judgment inside it.

Common Questions

When should I tell a client to build vs buy vs rent AI?

Recommend build only when the AI is a genuine differentiator and the client owns unique data, because that is the narrow case where a $50,000-plus tail and a 12-to-24-month timeline pay off. For supporting utilities, which is most use cases, recommend buy or rent the commodity layer and reserve any build budget for evaluation, not a team. The framework that decides it lives in the agency retainer breakdown.

What are the real costs and timelines of building versus buying?

Building runs $50,000 to $100,000 for an MVP and $250,000 to $400,000 and up for a multi-agent system, at 12 to 18 months to value, with upfront estimates missing 60 to 80 percent of true cost. Buying reaches value in 2 to 4 months and wins on year-one ROI for most use cases. Run costs overtake build cost within 18 to 24 months, and breakeven on a custom build sits near 33 months.

Should my agency build its own tool or run someone else’s?

For most agencies, run someone else’s configured layer and put your expertise inside it, rather than building and maintaining a tool. Building your own is the same per-client rebuild tax applied to your own business, and it commits you to maintenance forever. The cost case is in the agency AI tax.

How do I advise honestly on a tool I also resell?

State the conflict directly rather than burying it: you resell the platform you are recommending, and the client deserves to know. Then show the same build, buy, or rent math you would accept as a client yourself. The conflict resolves when your configured expertise is the product, not the margin on a license, because a client paying for your methodology configured inside the layer has a different value proposition than a client paying markup. The full resolution is in configured with your expertise.

Is reselling a platform the same as renting, and what about lock-in?

Reselling is a form of rent, and the lock-in risk is real, at the AI layer especially. Make rent safe by keeping the client’s data and your judgment portable so the layer can change underneath without a reset. The full treatment of reselling-as-rent and its lock-in is in white-label partner economics.

Get Started With AI

Are You Ready to Make AI Work for You?

Simplify your AI journey with solutions that integrate seamlessly, empower your teams, and deliver real results. Jyn turns complexity into a clear path to success.

See AI for Real Business Impact in Action →

ai that powers your team 226d8ee5db