AI Foundations for Small Business · Lesson 8 of 11

Build vs. Buy vs. Configure

Time: ~25 min reading + ~15 min worksheet Goal: Understand the three ways AI shows up in an organization, and how to recommend the right path without falling in love with building.

---

The Three Paths

Every AI opportunity boils down to one of three options:

Path 1: Buy

Use an existing product that already does the thing. Pay the subscription, learn the tool, move on.

Path 2: Configure

Start with a general-purpose AI platform (ChatGPT Team, Claude Team, Copilot, etc.) or a workflow tool (Zapier, Power Automate, Make) and tailor it with your company's instructions, files, and prompts. No custom code.

Path 3: Build

Commission or write custom software that integrates the AI API with the company's systems, data, and UI.

These are not equally expensive. They are not equally flexible. And — this is where consultants go wrong — they are not always better the higher up you go. Sometimes Buy beats Configure beats Build. Sometimes it's the other way around.

How to Choose

Five factors drive the decision:

1. Does a well-regarded product already exist?

If someone reputable has spent 3 years building a specific product for this exact problem, and there are happy customers in the same industry — buy it. You won't out-build them on a side project.

Examples where "buy" is almost always right:

  • Ambient clinical documentation (Abridge, Nuance DAX, Suki)
  • AP automation (Bill.com, Ramp, AvidXchange)
  • Meeting transcription (Otter, Fireflies, Fathom)
  • Customer service platforms with AI (Intercom Fin, Zendesk AI, HubSpot Breeze)
  • AI coding assistants (GitHub Copilot, Cursor)
  • Marketing content (Jasper, Copy.ai — though often Claude/ChatGPT + templates is better)

2. How unique is the workflow?

If the client's workflow is 80% like everyone else's, buy. If it's genuinely unusual — they have a niche, a special regulatory wrinkle, or their customers really are different — Configure or Build.

3. How tight does the integration need to be?

Embedded AI that shows up inside the user's existing tools beats switching to another app. If the client lives in a specific system (their proprietary ERP, industry-specific line-of-business app), integration is worth a lot.

4. What's the volume?

High volume (thousands of interactions a day) rewards Build. Low volume (dozens a day) doesn't justify it.

5. What's the budget, honestly?

Buying is capex/subscription. Configuring is time + subscription. Building is a project with real risk and real ongoing costs.

Cost Shapes

Rough ranges for a typical SMB engagement in 2026. Your mileage will vary.

Buy

  • Per-seat monthly fee ($10–$100/user/month typical)
  • Setup: often free or a few thousand dollars
  • Ongoing: subscription + some internal admin time
  • Total year one for a 50-person company: $10K–$75K depending on the product

Configure

  • Platform subscriptions (Claude Team $25/user/mo, Copilot $30/user/mo, Zapier plans ~$50–$300+/mo, etc.)
  • Consultant time to build custom GPTs, automations, templates — typically $5K–$30K for a package of deliverables
  • Ongoing: subscriptions + occasional tune-ups
  • Total year one: $15K–$60K

Build (custom API project)

  • Discovery + scoping: $3K–$15K
  • Build: $20K–$250K+ for a real project depending on scope
  • Ongoing: hosting + API costs + maintenance — typically 15–25% of build cost annually
  • Total year one: $30K–$300K+

The trap

Consultants who only know how to build tend to recommend building. Consultants who only know how to sell software recommend buying. Be the one who honestly picks the right answer for the client — you'll win more trust long-term.

When Each Path Wins

Buy wins when:

  • A mature product exists in the space
  • The client's problem is common in their industry
  • They want predictable monthly spend
  • They need something working next month, not next year

Configure wins when:

  • The client has a lot of general-purpose writing/knowledge work to accelerate
  • The workflow uses their specific voice, style, data
  • They want to iterate quickly
  • Budget is modest
  • This is where you (the consultant) often make the most impact per dollar

Build wins when:

  • The problem is specific and strategic
  • The client has volume and scale
  • Existing products don't fit
  • They want to own the IP
  • They're playing a long game (the custom system becomes competitive advantage)
The Mistake Everyone Makes

New consultants look at an SMB's pain and think "I should build them something cool." Clients love the idea — custom always sounds sexier than a subscription.

But the math rarely works. A $40K custom build solving a problem that a $50/month SaaS already solves leaves the client with:

  • A system to maintain
  • Technical debt
  • Dependency on you (or whoever's available)
  • Features the commercial product already has for a fraction of the price

The test: Before recommending a build, spend one full day searching for existing products. Is there something 70% fit? Often yes. Start there. Build only when you've ruled it out.

Hybrid Approaches — Often the Sweet Spot

Real engagements often combine all three:

  • Buy Copilot licenses and ambient meeting notes
  • Configure three Custom GPTs for specific company workflows
  • Build one targeted automation that ties the client's proprietary system to AI

This lets you land fast wins with buying and configuring, then earn the right to do the bigger build when trust is established.

What to Actually Recommend When

Quick decision guide:

| Situation | Start with | |---|---| | "We want to use AI but don't know where to start." | Configure. Roll out an approved platform (Copilot/Claude), train the team, build 3 Custom GPTs. Level 1–2. | | "We have a specific pain — doc processing, meeting notes, AP, etc." | Buy the best-fit SaaS. Don't reinvent the wheel. | | "We want a chatbot on our site that answers product questions." | Buy first (Intercom, Voiceflow, etc.). If truly specialized, Build. | | "We have a proprietary workflow where our edge lives." | Build. That's the IP. Go deep. | | "We want all our field techs to file reports faster." | Configure with voice notes + Extractor prompts. Build if deep FSM integration required. | | "We're worried about data / on-prem only." | Buy on-prem or private cloud, or Build on open-source models. |

Total Cost of Ownership (TCO) — Don't Get Blindsided

When pricing any path, include:

  • Initial costs — licenses, setup fees, consulting time
  • Ongoing subscriptions — include them for 3 years in the comparison
  • API / usage costs — especially for Build paths, token costs add up
  • Training / change management — real dollars, often skipped
  • Maintenance / admin — someone has to keep it running
  • Integration / data cleanup — often the largest hidden cost
  • Sunset risk — what happens if the vendor goes away, or the tech shifts?

A clean TCO comparison done over 3 years beats the "look at the low sticker price" pitch almost every time.

When the Right Answer Is "Not Yet"

Sometimes the answer to "what AI should we use?" is "none, right now." Reasons:

  • Their data is a mess (clean it first)
  • No clear workflow pain — they're asking because of hype
  • Budget doesn't match the realistic project size
  • Team can't absorb any more change right now
  • Leadership hasn't aligned on what they want

Saying "not yet" when it's right buys you more trust than forcing a sale. Good consultants say this often; bad ones never say it.

---

Key Takeaways
  • Three paths: Buy, Configure, Build. Each has its sweet spot.
  • Buy if a good product exists — don't reinvent.
  • Configure is the SMB sweet spot for most early AI work.
  • Build only when the problem is specific, strategic, and big enough to justify it.
  • Hybrid engagements win — start where you can move fast, earn the right to do more.
  • TCO over 3 years is the honest comparison, not sticker price.
  • "Not yet" is sometimes the best answer. Use it.

---

Next up: Module 9 — Sizing the Opportunity. How to put real numbers on an AI project so the owner can say yes.

---

Time: ~15 min Goal: Practice the build/buy/configure call.

---

Exercise 1: Pick the Path (10 min)

For each situation, recommend Buy, Configure, Build, or Hybrid, and justify in one sentence.

  • A 20-person medical billing company wants to automate extracting info from incoming EOBs (Explanation of Benefits documents) from dozens of payers, each formatted differently.

→ Path: ______________ → Why: ______________

  • A 60-person B2B SaaS company wants their entire sales and marketing team to write with consistent brand voice and stop starting everything from scratch.

→ Path: ______________ → Why: ______________

  • A 45-person home services company wants to add AI to their ServiceTitan dispatch workflow — the scheduler should recommend daily routes based on traffic, technician skills, and past job duration.

→ Path: ______________ → Why: ______________

  • A 15-person law firm wants to automate intake — website forms in, initial conflict check, initial client summary, meeting scheduled.

→ Path: ______________ → Why: ______________

  • A 120-person regional bank wants an internal Q&A system that answers staff questions from their 2,000-page policy manual, enforcing role-based access control.

→ Path: ______________ → Why: ______________

  • An 8-person startup wants to "add AI" to their product to keep up with competitors.

→ Path: ______________ → Why: ______________

Exercise 2: Build the TCO (5 min)

You're recommending a Configure engagement to a 30-person company. Lay out a realistic 3-year TCO. Make reasonable assumptions and write them down.

| Line Item | Year 1 | Year 2 | Year 3 | |---|---|---|---| | AI platform subscriptions (per user) | | | | | Consulting / setup (your fee) | | | | | Workflow automation tools (Zapier/Make/PA) | | | | | Training | | | | | Maintenance / tune-up | | | | | Total | | | |

Assumptions: ______________

---

Answer Key / Discussion

Exercise 1:

  • Buy. This is a well-defined, mature problem — many vendors already do EOB extraction (Inovalon, Waystar, Thoughtful AI, various clearinghouse tools). You won't out-build specialists for reasonable money. Configure might come in if no off-the-shelf fits, but start with buy.
  • Configure. A handful of Custom GPTs/Projects + good templates + training = fast ROI. Not worth building custom software for this.
  • Hybrid, leaning Buy. ServiceTitan has its own optimization features, plus add-on products. If those don't fit, Configure with automation tools. Only Build if they're at real scale and have clear competitive advantage in routing.
  • Hybrid (Configure + Buy). Intake is a common problem — look at tools like Clio, MyCase, Lawmatics (law-specific). Configure a Custom GPT for the intake summary narrative. Don't build from scratch.
  • Build (or heavy Configure with an enterprise product). RBAC + 2,000 pages + staff Q&A = a real RAG project. Some enterprise products do this (Glean, Microsoft Copilot Studio with SharePoint integration). Mid-size banks often need custom because the security requirements are specific. Budget tens of thousands.
  • Not yet. "Competitors have AI" isn't a use case. Run a real discovery conversation (Module 10) to find actual pain before spending money.

Exercise 2 — sample TCO for a 30-person configure engagement:

Assumptions: 30 users on Claude Team ($25/user/mo = $9K/year), Zapier team plan (~$50/month = $600/year), one-time consulting package of three Custom GPTs + two automations for $18K, $3K training budget, ~$4K/year maintenance.

| Line Item | Year 1 | Year 2 | Year 3 | |---|---|---|---| | Claude Team for 30 users | $9,000 | $9,000 | $9,000 | | Zapier team plan | $600 | $600 | $600 | | Consulting / setup | $18,000 | $2,000 | $2,000 | | Training | $3,000 | $500 | $500 | | Maintenance / tune-up | $3,000 | $4,000 | $4,000 | | Total | $33,600 | $16,100 | $16,100 |

3-year total: about $66K. For a 30-person company where you can plausibly save 10–15% of everyone's writing time, the math usually works. Show the owner this math.

---

Check yourself: If an owner says "everyone says we need AI, what should we do?" — can you walk through the Build/Configure/Buy decision in 15 minutes and leave them with a clear recommendation? That's the skill.

Previous: Data, Privacy, and Risk (The Boring Stuff That Matters) · Next: Sizing the Opportunity