The build vs buy vs partner question for AI is mostly decided by one variable: does your company have someone who can build and maintain the AI system correctly; while continuing to run the operations it is being built into?
For most $10M–$25M non-tech companies; the honest answer is no.
The choice is not between three equally viable options. It is between one option that produces generic AI capability; one that produces nothing that is not already available off-the-shelf; and one that builds the company-specific system the investment requires.
That framing is direct because the decision deserves a direct answer rather than a false equivalence of three viable paths.
This article describes what each path produces; who each is right for; and how to identify which path matches the specific constraints of a $10M–$25M non-tech company.
The build path: what it requires and who it is actually for
What the build path actually requires
The build path; developing the company’s AI capability internally without an external partner; requires four things that are rarely present together in a $10M–$25M non-tech company.
Requirement 1: A technically capable AI builder
Not just someone who uses Claude well; someone who can design workflow specifications; build prompt architectures that produce consistent outputs; configure automation tools (Make; Zapier; or code-based); and diagnose quality problems when they appear.
For most $10M–$25M non-tech companies: this person does not exist internally. If they do exist; they are already fully occupied in another role.
Requirement 2: Dedicated time that does not displace operations
The build path takes 6–12 months of steady effort; not 6–12 months of someone’s 10% availability.
A partly-built AI system is worse than no AI system because it produces inconsistent outputs that calibrate the team to expect poor AI performance.
If the internal build must be done by someone who is also running 80% of their capacity in their operational role; the build will not be completed at the pace required and the quality will reflect the attention available.
Requirement 3: Deep operational context in the builder
The person building the AI system needs to understand how the company actually operates; not the official version; but the operational reality. The decision rules; the client archetypes; the edge cases; the judgment calls that recur frequently.
This knowledge is usually in the founder or senior operations leader; not in the technically capable person who might be hired to build the system.
This mismatch is the most common reason internal builds produce generic AI systems: the builder has the technical skills but not the operational knowledge; the operational leader has the knowledge but not the technical skills.
Requirement 4: A 6–12 month horizon without the pressure of visible results
The build path takes time that produces no visible results in months one through three. The context pack; workflow documentation; and context loading are not impressive-looking deliverables.
A company under pressure to show AI results quickly; to a board; to clients; or internally; will not stay the course through the invisible foundation phase.
Who the build path is actually for
A $10M–$25M company with all four requirements is rare but exists.
The company that should consider the build path has:
- A founder or COO who is personally technically capable and can dedicate 15–20 hours per week to the build
- Deep operational context in the builder
- A twelve-month horizon without pressure for visible results
- Operations that are already well-documented
For most companies reading this: one; two; or three of these are present; not all four.
The buy path: what it produces and where it runs out
What the buy path is
The buy path is purchasing commercial AI solutions; either general-purpose (Claude Teams; ChatGPT Team) or purpose-built (HubSpot AI; an AI-powered accounting tool; a commercial AI customer service platform); and using them without the foundational build work that makes them company-specific.
This is where most companies start; the tool-first approach.
Where buy is appropriate
High-volume; low-judgment; well-defined use cases:
| Use case type | Example | Why buy works |
|---|---|---|
| Invoice processing | Commercial AP automation | Structured inputs; defined outputs; no company voice needed |
| Meeting transcription | Otter; Fireflies | Structured input; defined output format; commodity task |
| Email filtering | Built-in email AI | Rule-based; no company-specific judgment |
| Embedded software AI | HubSpot AI; Notion AI | Augments existing tool; low specificity requirement |
Team productivity for AI-fluent individuals:
Claude Pro or ChatGPT Plus for individuals who need general-purpose AI assistance. These produce value for AI-fluent individuals even without company-specific configuration; the user provides their own context.
Where buy runs out
Commercial solutions without custom configuration produce generic outputs for company-specific tasks.
The proposal workflow; the client communication; the specific decision logic of the company’s operations; none of these are addressed by the commercial solution out of the box.
Commercial solutions also do not produce the improvement loop; the shared context; the workflow documentation; and the ownership structure that make the investment compound. They produce a tool. The strategy layer converts the tool into a system.
The partner path: what it produces and who it requires
What the partner path produces
An engaged AI partner builds the things the buy path does not include and the build path requires significant internal capability to produce:
- AI Foundations: context pack; voice guide; decision rules; and workflow documentation built from the company’s actual operations
- The shared workspace: configured; loaded; and working before the team is trained on it
- The trained team: each team member trained on their specific role workflows using real current work
- The improvement loop: adoption tracking; maintenance cadence; and a trained AI system owner who maintains it after the engagement
- The first automated workflows: trigger-based workflows that run without human initiation; built from proven manual workflows
The partner’s primary contribution is the expertise to build these correctly and the time that the company’s operations cannot provide.
What the company must contribute
The partner path requires genuine contributions from the company:
- Founder or COO time for context pack development (the partner structures the work; the company provides the knowledge)
- Team member time for training (60–90 minutes per person for their specific workflows)
- AI system owner capacity (3–5 hours per week during the active engagement)
- Operational access (the partner needs to understand how the company works)
The partner does not replace the company’s contribution; it structures and accelerates it.
The cost-outcome comparison
| Build | Buy (unconfigured) | Partner | |
|---|---|---|---|
| Typical investment (year 1) | $80,000–$180,000 (dedicated AI hire) | $3,000–$15,000 (tools) | $60,000–$150,000 (engagement) |
| Foundation built | Depends on internal capability | Not included | Included |
| Team trained | Depends on internal capability | Not included | Included |
| System owner | Must be hired or designated | Not included | Trained during engagement |
| Improvement loop | Depends on internal capability | Not included | Installed in Phase 1 |
| Time to running workflows | 6–12 months | Immediate but generic | 4–8 weeks |
| Risk | High if internal capability is insufficient | Low but outcomes are limited | Medium; depends on partner quality |
The combined approach: the actual optimal path for most companies
The three paths are not mutually exclusive. The most effective AI strategy for most $10M–$25M non-tech companies combines all three.
Partner for the foundational build (Phases 1 and 2)
An embedded partner builds the AI Foundations and trains the team. This produces the context pack; workflow documentation; shared workspace; trained team; and installed improvement loop.
Cost: the Phase 1 and 2 engagement investment.
Outcome: a working AI foundation and trained team in 8–12 weeks.
Buy for high-volume commodity workflows
Commercial tools handle the high-volume; low-judgment; well-defined workflows where off-the-shelf outputs are acceptable.
Cost: $50–$200/month for the relevant tools.
Outcome: specific high-volume tasks handled without custom configuration.
Build the specific custom components no commercial solution handles
After the foundation is in place; the company’s own AI system owner (trained by the partner engagement) builds specific custom components for workflows that are highly specific to the company’s operations.
Cost: the AI system owner’s ongoing time (3–5 hours per week).
Outcome: the custom components that make the AI operation fully company-specific.
The combined path: cost and outcome summary
Total first-year investment for a 10–20 person company:
Phase 1 + 2 partner engagement: $60,000–$120,000
Commercial tool subscriptions: $3,000–$6,000/year
AI system owner time (20% of role): $15,000–$30,000 equivalent
Total: ~$80,000–$150,000
Minimum return (conservative): 2–3 hours/person/week recovered
For 10 people at blended $75/hour: $78,000–$117,000/yearThe combined path produces return above investment in year one for most $10M–$25M companies; with compounding returns in years two and three as the system improves.
Common questions on build vs buy vs partner
”What if we have a technical co-founder: does that change the build path assessment?”
Yes; significantly. A technical co-founder who can dedicate time to the AI build and has the operational knowledge addresses Requirement 1 and potentially Requirement 3 simultaneously.
The build path is more viable with a technical co-founder. Run the four requirements against their actual availability and capacity; not their theoretical availability.
If they are running product and engineering at full capacity; the time requirement (Requirement 2) may still be the binding constraint.
”Is there a company size below which the partner path is not economically justified?”
Below five people and below $5M ARR; the partner path for the full four-phase engagement may exceed the return in year one. A Phase 1 Foundation-only engagement is typically appropriate and economically justified at this size.
The minimum viable partner engagement: Phase 1 only (context pack; workflow documentation; initial training). For a 3–5 person company; this is the investment that produces adoption without the overhead of a full four-phase engagement.
”What happens to the investment if the partner relationship ends: do we own what was built?”
Yes; entirely. The context pack; workflow documentation; trained team; shared workspace configuration; and running workflows all belong to the company. The partner builds into the company’s own accounts and systems; not a proprietary platform.
The partner relationship ends; the system continues. The AI system owner the partner trained maintains it independently. That is the design goal of the engagement; not a contingency.
”Can we start with buy and upgrade to partner later?”
Yes; and many companies do.
The tool-first experience provides real value and produces the inputs (use case clarity; team AI fluency; knowledge of what AI does not handle well) that make a subsequent partner engagement more effective.
The risk of staying in buy mode too long: the tool-first ceiling arrives and the team calibrates to “AI doesn’t work that well for our business.” That calibration is harder to reverse after twelve months than after three.
”What does a bad partner engagement look like: what are the warning signs?”
Four warning signs during the engagement evaluation:
- The proposal does not specify which workflows will be documented and built; only that “your workflows will be transformed”
- The engagement ends at system launch rather than at a confirmed acceptance rate target
- There is no named system owner in the proposal; the maintenance is described as “ongoing support”
- The cost is 100% upfront with no accountability milestones; or 100% at the end with no payment triggers tied to measurable outputs
An engagement structured to produce adoption has specific deliverables; specific acceptance rate targets; and a system owner trained before the engagement ends.
”How do we evaluate whether a partner has the operational understanding to build effectively for our specific company?”
Three specific questions:
- “Have you built AI foundations for companies in [our industry/function]? What did the context pack for that company contain that was specific to that industry?”
- “What is the most common edit type you see in a client’s adoption log in the first month; and what does that tell you about the context pack quality?”
- “Walk me through how you would identify the highest-priority workflow for our company to document first.”
A partner with genuine operational experience in your context type will answer these specifically. A partner with only technical capability will answer generically.
Want to know specifically which combination is right for your company’s constraints?
Build vs buy vs partner is not a three-way competition. It is a question of which path matches the company’s constraints; and for most $10M–$25M non-tech companies; the honest answer is a combination.
- Partner to build the foundation correctly
- Buy for commodity workflows
- Build the custom components that only internal expertise can produce
The company that tries to build everything internally without the full internal capability required produces the most expensive failure mode. The company that buys without building the strategy layer produces the tool-first ceiling. The company that partners for the foundation and combines it with targeted buying and specific internal build produces the AI operation the investment was meant to create.
Path one: run the four build requirements against your actual capacity today. Be honest on each. If any two of the four are not present; the build path is not viable for your situation. The combined path is.
Path two: bring in a partner for the foundation. Phos AI Labs is the partner path for the foundational build; designed for companies that have identified they need external expertise for Phase 1 and Phase 2 and want the combined approach. We have run 400+ AI engagements. Clients include Zapier, Coca-Cola, Medtronic, Dataiku, and American Express. Thirty minutes, no deck. Start here.