Should you build the AI capability custom, buy an off-the-shelf tool, or borrow it through a partner relationship? It’s the choice most mid-market operators get wrong — usually by defaulting to buy because it feels familiar.
Here’s a framework for getting it right.
The three options
- Build: a custom AI workflow developed for your specific business, owned by your team or a partner, operated inside your environment.
- Buy: off-the-shelf AI software (or an AI feature of software you already use) that you adopt with minimal customization.
- Borrow: a partner relationship where senior expertise operates inside your business on retainer — they own the capability, you consume the outcome.
Most mid-market operators default to buy because it feels familiar. Most are wrong. And the cost of getting it wrong is real: according to RAND, more than 80% of AI projects fail — roughly twice the failure rate of IT projects that don’t involve AI.
When to buy
Buy when AI is infrastructure you’d rather not build:
- The capability is truly horizontal — every business needs it in roughly the same way (calendar scheduling, email categorization, document storage with AI search).
- Off-the-shelf options are mature and integrate with your existing stack.
- The cost case is clear, and you can swap if a better option emerges in 12 months.
- The capability is not a competitive differentiator — table stakes only.
When to build
Build when AI is capability you want to own:
- The workflow is specific to your business — your quoting, your estimating, your customer service patterns, your data.
- The capability is competitive differentiation — something you’d want to defend.
- You have or can acquire the engineering capacity (internal or partner) to maintain it.
- Off-the-shelf options exist but don’t fit your workflow — forcing your team to adapt to the tool instead of the other way around.
When to borrow
Borrow when AI is expertise you can rent better than you can hire:
- You don’t have time to wait 12 months for an internal hire to ramp.
- The judgment work matters as much as the engineering work — and you need someone who’s seen 30 similar engagements.
- The economic case for an in-house hire is shaky given your scale.
- You want speed, pattern matching, and operational accountability without the long-term hiring risk.
The mid-market default
For most $5–100M B2B service businesses, the right path is borrow first, then build, then sometimes buy:
- Year 1: Borrow. A senior partner on retainer gives you the judgment and the engineering without the long internal ramp. You learn what you actually need before committing to an internal hire.
- Year 2: Selectively build. Now that you know which workflows matter, selectively build the workflows that matter — the ones that are competitive differentiation. Keep the partner for senior judgment and engineering surge.
- Buy where you can. Common workflows — scheduling, calendaring, doc search, basic CRM — just buy. Don’t over-invest in building what’s becoming commodity.
The traps
Trap 1: Buy everything. You end up with 8 AI tools, none of which fit your actual workflows. License utilization looks good; nothing in the business changes — and then you quietly scrap them. The share of businesses abandoning most of their AI initiatives jumped to 42% in 2025, up from 17% the year before, per S&P Global Market Intelligence.
Trap 2: Build everything. You hire an internal AI team at $400K+ all-in, ramp them for 6 months, and produce 1–2 workflows in year 1. You could have gotten 3–5 with a partner at a quarter the cost.
Trap 3: Borrow then never build. You stay dependent on the partner forever. Some partners encourage this — it’s high lifetime value for them. Good partners actively work to make you more capable so you need less of them over time. Choose accordingly.
Trap 4: Buy and call it building. Subscribing to ChatGPT Enterprise isn’t building AI capability. It’s licensing a tool. The capability lives in the workflows, the integrations, and the team adoption — none of which a subscription provides.
What this means in practice
If you’re starting from zero, the highest-leverage move is to borrow for the first 12 months: a partner who’ll identify the right workflows, scope the builds, and ship the working solutions.
After 12 months, you’ll know enough about which workflows matter to your business that the build-buy-borrow decision is informed by actual data — not by speculation.
Most companies that hire an AI lead before working with a partner end up firing or replacing that hire within 18 months. The internal lead was hired against a hypothesis about what AI work needed doing; the actual work turned out to need different skills. Borrowing first surfaces what you actually need.
If you’re in the “we should probably hire someone, but we’re not sure for what” zone, borrowing is built for exactly that decision. Borrow first to surface what you actually need, then identify the right workflows before you commit. By month 6, you’ll know what role to hire — or whether you need to at all.