Build vs buy AI for a growing team (the SaaS subscription trap explained)

Quick math problem. You pay $99/month for a "AI-powered" tool. The tool wraps GPT-4o under the hood. The actual API cost to the vendor for what you use is somewhere between $4 and $9 per month.

You're paying a 11x markup. Sometimes worse.

Multiply that across the seven SaaS tools your 12-person business is paying for. That's somewhere between $400 and $1,500 per month going to vendors who are renting you the same underlying intelligence at a markup. None of them custom to your workflow. All of them generic.

This is the "wrapper economy." It's also the reason build-vs-buy math has changed in 2026.

The old answer

Five years ago, the build-vs-buy answer was easy: buy the SaaS. Custom software for a growing team meant hiring a developer for $150/hour, a six-month timeline, and a maintenance burden you couldn't afford. The SaaS was always cheaper, even at $99/month forever.

That math held when "custom software" meant a 6-month CRUD app build.

What changed

In 2026, custom doesn't mean six months. It means seven days. Productized AI agents — built on Claude Code, Lovable, and the rest of the agent stack — can be scoped, built, tested, and shipped in a week at fixed price.

The build cost dropped from $90,000 to $4,995. The maintenance burden dropped from a full-time engineer to a $497/month retainer (or zero, if you don't need it).

That changes the math.

The new breakeven

Here's the actual question: at what point does $4,995 once beat $99 per month forever?

If a SaaS tool costs $99/month, you're paying $1,188/year. Over five years, $5,940. Plus annual price increases, plus the usability tax of a generic tool that doesn't quite fit your workflow.

A productized AI agent at $4,995 with a one-time build cost pays back in 4 years and 3 months on direct cost alone. But that's not the real comparison.

The real comparison is fit. A productized agent built for your workflow, talking to your tools, using your data, eats the work. A generic SaaS that sort-of-fits your workflow gets used 60% of the time and ignored the other 40%. The 40% you ignore is where the SaaS markup quietly stops being worth paying.

When buy still wins

Three categories where the SaaS subscription still wins, no question:

1. Tools that are hard moats. Your CRM. Your accounting software. Your tax-prep tool. The vendor has spent 15 years building a product that does 1,000 things. You can't out-build HubSpot in a week. Buy.

2. Tools that need a network effect. Slack. Zoom. Calendly. The value is that everyone else uses them. You're not buying the software, you're buying the integration with your customers and team. Buy.

3. Tools you'll use for less than a year. A $99/month tool you'll cancel in six months is $594. A $4,995 build that you'll abandon in six months is $4,995. The risk-adjusted math favors buying when you're not sure you'll keep using it.

When build starts winning

Three categories where the math has flipped to build:

1. The 80% workflow that nobody else does the way you do. Your specific intake process. Your specific proposal format. Your specific pre-call brief. The thing you'd describe in a sentence and that no off-the-shelf tool quite handles. Build.

2. The "five SaaS tools, none of which talk to each other" problem. When you're paying for HubSpot, Calendly, Zoom, Loom, and a transcription tool, and the friction between them is a 15-minute manual workflow per call, building one agent that traverses all five is cheaper than another SaaS that promises to "integrate."

3. Anything that turns into recurring revenue or recurring cost. If automating it generates revenue you can measure, or eliminates a recurring expense, the one-time build pays back fast. The cleanest example: meeting-notes-to-CRM. Saves 8 hours/week of admin per partner. At $250/hour blended, that's $104,000/year per partner of recovered time. Against a $4,995 build cost, the ROI math is 20x in year one.

The two-question test

Before deciding, two questions:

Question 1: Does this exact workflow exist in 1,000 other businesses? If yes, buy the SaaS. If no, build the agent.

Question 2: Does the time saved exceed $5,000 of value per year? If yes, build it. If no, the SaaS subscription is fine.

Most growing teams get this wrong in both directions. They buy SaaS for things that should be custom (the unique 80% workflow), and they hand-do things that should be automated (the painful repetitive task that's worth $30k/year of time).

Where this lands for a 5–25 person business

Five SKUs that almost every B2B has and that productized agents handle well: inbox triage, proposal drafting, meeting notes to CRM, CRM data hygiene, and an AI receptionist. Each one's a fixed-price build at $2,495–$4,995. Each one replaces 1–3 SaaS subscriptions or a chunk of human admin time.

For most 5–25 person businesses, the right move in 2026 is build the 2–3 things that are unique to how you work, and keep buying the SaaS for the things that aren't.

The subscription trap isn't paying $99/month. It's paying $99/month for ten things, none of which actually fit, and never auditing whether one of them should be a $4,995 one-time build instead.

The hybrid stack most growing teams actually run

In practice, almost no 5-25 person business runs all-build or all-buy. The honest answer is a hybrid:

• Buy: HubSpot, Slack, Zoom, QuickBooks, Google Workspace. Anything with a network effect or a deep integration moat.
• Build: the 1-3 workflows that are uniquely yours and are eating the most time.
• Skip: the 4-6 SaaS tools you're paying for and barely use.

Most owners find that the audit conversation reveals two SaaS subscriptions per year that should be canceled. At $99/month each, that pays for the build cost.

What to build first if you build something

Five workflows that almost universally pay back faster than other options:

1. Meeting Notes → CRM. Highest direct ROI for sales-driven businesses. Recovered hours convert directly to billable capacity.

2. Inbox Triage. Highest morale impact. Owners and managers stop dreading their inbox.

3. Proposal Drafter. Highest revenue impact for service businesses. Faster proposals = higher win rates against competitors.

4. CRM Hygiene Sprint. Highest data-quality impact. Makes every other automation work better.

5. AI Receptionist. Highest after-hours capture for service businesses with phone-driven leads.

The order depends on the business. A consulting firm starts with Meeting Notes. An agency starts with Proposal Drafter. A trades business starts with AI Receptionist. There's no universal first move.

When build still loses (be honest)

Three cases where the math actually favors buy, even in 2026:

1. You'd skip using it. A custom build that gets ignored is worse than a SaaS subscription you'd actually use. Build only if you'll deploy it.

2. The workflow changes every six months. If your sales process is still being figured out, building an agent for it bakes in assumptions that'll break. Buy a flexible SaaS, build the agent in year two.

3. The integration target is unstable. If you're on a niche CRM that gets acquired every 18 months, building on top is risky. Stabilize the underlying tools first.

Common questions

What about AI tools we already pay for? Audit them. Half of small-business AI SaaS subscriptions are unused after month three. Cancel the ones that aren't getting opened. The savings often pay for the productized build.

What if my needs change after the build ships? Productized agents are scoped to a workflow, not a frozen process. Small adjustments (new fields, new categories, new triggers) are part of the optional retainer. Big rewrites (different workflow entirely) are a new project.

What about open-source AI? Can we build this in-house? Technically yes. Realistically, a 12-person B2B doesn't have the engineering capacity to maintain a custom AI stack. The "in-house" path makes sense at 50+ people with at least one ML engineer. Below that, productized wins on TCO.

Honest case studies (with the math)

Three real-shape comparisons across small-business sizes:

Case A: 6-person consulting firm. - Currently pays for HubSpot ($1,440/year), ChatGPT Teams ($720/year for 3 seats), Calendly ($240/year), Zoom ($600/year). Total: ~$3,000/year on SaaS. - Builds a Meeting Notes → CRM agent at $4,995. Saves 6 hours/week of partner admin. - Year-one math: $4,995 build vs $3,000 SaaS still running (build doesn't replace, it adds). But recovered partner time = $93,600 at $300/hr blended. - Net: build wins by an order of magnitude.

Case B: 18-person marketing agency. - Currently pays for HubSpot Pro ($9,600/year), Asana ($3,600/year), Lavender ($1,800/year), Slack ($2,400/year). Total: ~$17,400/year. - Builds Proposal Drafter + Inbox Triage + Meeting Notes ($11,985 stack). - Cancels Lavender (replaced by Inbox Triage). Saves $1,800/year. - Recovers ~24 team hours/week = $375,000/year at $300 blended. - Net: build wins, plus modest SaaS cancellation savings.

Case C: 4-person law firm. - Currently pays for Clio ($4,800/year), MyLegalSoftware ($1,200/year). Total: ~$6,000/year. - Builds Inbox Triage + Engagement Letter Drafter ($6,990). - Doesn't cancel anything (Clio is the moat). Adds the agent. - Recovers 8 hours/week of partner time = $135,000/year at $325 blended. - Net: build wins on outcome math, doesn't reduce SaaS spend.

The pattern: at any size above 5 people in a service business, the productized build pays back inside year one even when it doesn't replace existing SaaS.

When to skip the build entirely

Three signs:

• Your team is under 5 people and you don't yet have a stable workflow
• You're still in product-market fit chaos and your operations change monthly
• Your current SaaS stack actually fits the work and you're not annoyed by it

In those cases, keep buying. Revisit in 6 months.

What we'd do this week if it were our business

Pragmatic version: if you're under 5 people and your workflow is still changing weekly, keep buying SaaS. Revisit in 6 months.

If you're 5-25 people with stable workflows, run a quick exercise:

Step 1. List every SaaS tool you're paying $50+/month for. Write the actual annual cost next to each.

Step 2. Open each one. Check the last-used date. Strike out anything you haven't opened in 30 days.

Step 3. For the survivors, ask: does this tool fit our workflow exactly, or does our team work around it? Strike out anything you work around.

The remaining list is your real SaaS spend. Add it up. Honest answer is usually 2-3x more than the headline-cheap monthly numbers suggested.

What this means for you: half of small-business AI SaaS spend is unused. The other half is partially-used. The recovered budget pays for productized builds that actually fit.

Where to start

A 30-minute audit call maps your stack, your workflows, and your 5-tool overlap. Output: a punch list of which 2–3 things to build and which to keep buying. No commitment, no deck.