A client calls you on a Tuesday. They're tired of missing after-hours calls, and they heard you're "doing AI now." You want to say yes. The question is how you back that yes.
You've got three ways to add AI voice to your stack: build it yourself on raw APIs, resell somebody else's branded product, or run a true white-label platform under your own name. They look similar on a slide. They are wildly different businesses.
Most MSPs pick wrong because they only price the first month. Here's the math that actually matters.
The three paths, plainly
Build (DIY). You wire together an LLM, a text-to-speech engine, a speech-to-text engine, and a telephony layer. You own the prompts, the call logic, the dashboards, the billing. You're now a software shop.
Buy (resell). You sign up as a reseller or affiliate of an established AI voice product. You sell their brand, they handle the tech. You take a referral cut or a thin resale margin.
White-label. You run a platform that's already built, but it carries your brand, your domain, your pricing. The vendor stays invisible. You own the customer relationship and the margin; they own the infrastructure.
The difference between these isn't features. It's who owns the customer, who owns the margin, and who carries the 2 a.m. problem.
The comparison table
| Dimension | Build (DIY) | Buy (Resell) | White-Label |
|---|---|---|---|
| Control over product | Total | Almost none | High (your config, their core) |
| Gross margin | High if it works, after huge sunk cost | Thin (10-30% typical) | Strong (you set retail) |
| Time to market | 3-9 months | Days | ~48 hours |
| Brand | Yours | Theirs | Yours |
| Support burden | All of it, forever | Vendor handles end users | Shared: you do L1, vendor does platform |
| Compliance (TCPA, recording consent, data) | You own it entirely | Vendor's terms, your exposure unclear | Vendor handles platform; you handle client config |
| Upfront cost | $50k-$250k+ in dev time | ~$0 | Low setup, monthly platform fee |
| Ongoing eng cost | 1-2 FTEs minimum | None | None |
| Customer ownership | Yours | Often the vendor's | Yours |
| Exit / lock-in risk | You're locked into your own code | High - they can cut you off | Moderate - depends on contract |
Read that "ongoing eng cost" row twice. It's where DIY budgets go to die.
The real cost of Build
The per-minute API math looks cheap. Stitch together a model, a TTS voice, and transcription and you're looking at roughly $0.05-$0.15 per minute at the infrastructure layer for the raw components, before telephony. That's the number DIY enthusiasts quote.
It's also the number that's irrelevant.
The cost of build isn't the inference. It's everything wrapped around it:
- Prompt and flow engineering so the agent doesn't hallucinate appointment times or talk over callers.
- Telephony plumbing - SIP trunks, number provisioning, call transfer that actually works.
- A dashboard your clients will use, with call logs, transcripts, and recordings.
- Multi-tenant billing so you can charge 40 clients without a spreadsheet from hell.
- Latency tuning. Sub-second response or the call feels broken. This is genuinely hard.
- On-call. When it breaks at 2 a.m. during a client's lead rush, that's you.
Budget $50k-$250k+ in engineering time to get to a product you'd put your name on, then 1-2 full-time engineers to keep it alive. You're not adding a service. You're founding a software company as a side quest.
When DIY actually makes sense: you have in-house engineers with downtime, a genuinely unusual use case no platform serves, volume large enough to amortize the build (think 50,000+ minutes/month across clients), and you want the IP as an asset. That's a real scenario. It's just not most MSPs.
The real cost of Buy (reselling)
Reselling is the opposite trap: it's easy and it's a dead end.
You get to market in days with zero engineering. Fine. But you're selling someone else's brand into your own accounts. Every time a client logs in, they see the vendor's name, not yours. You've just paid to introduce your customer to a company that can disintermediate you later.
And the margins are thin. Referral and resale deals typically leave you 10-30%, often as a one-time or trailing commission rather than recurring revenue you control. You don't set the price, so you can't expand the margin. You don't own the billing, so you can't bundle it into your stack cleanly.
Reselling is fine as a tactical test. It's a bad foundation for a recurring-revenue line.
The real cost of White-Label
White-label is the middle path that usually wins for MSPs, and not because it's a compromise - because the economics line up with how you already run.
You skip the build. The platform exists, it's tuned, it handles latency and telephony and the dashboard. But it wears your logo, sits on your domain, and uses your pricing. Your client never sees the vendor.
You own the customer and the margin. You set retail, the platform takes a wholesale cut, and the spread is yours - recurring, every month. (We break the packaging down in the white-label AI voice pricing guide and the margin mechanics in the UCaaS margin post.)
The honest tradeoffs:
- You don't control the core roadmap. If you need a wildly bespoke capability, you wait on the vendor.
- Support is shared. You handle L1 - client questions, config tweaks. The vendor handles the platform. That's less burden than Build, more than Buy.
- You're tied to a vendor. Pick one with a fair contract and clear data ownership, or you've traded one lock-in for another.
For an MSP whose real product is the relationship and the trust, not the codebase, white-label lets you sell AI voice as your own without becoming a software company.
How to use this
Run your own decision in four steps.
- Forecast 24 months, not one. Add up dev time, ongoing engineering, and on-call for Build. Compare to the white-label platform fee over the same window. The cheap-per-minute Build illusion evaporates fast.
- Decide who must own the customer. If protecting the account relationship matters - and for an MSP it always does - reselling is out. That narrows you to Build or White-Label.
- Be honest about your engineering bench. No spare senior engineers? Build is a fantasy. Don't start it.
- Pressure-test the exit. For white-label, read the contract: data ownership, offboarding, what happens if you leave. For Build, remember you're locked into your own maintenance forever.
Quick gut check: if you can't name the two engineers who'd own the Build, you're not building. Go white-label.
FAQ
Isn't building cheaper long-term since I avoid platform fees? Only if you ignore engineering salaries, which are the actual cost. Two engineers maintaining a voice stack will outrun any platform fee. Build wins on cost only at large scale with an engineering team you already pay.
Can I start by reselling and switch to white-label later? Yes, and it's a reasonable test. Just know you may have introduced your clients to the vendor's brand first, which you'll have to overwrite. Cleaner to white-label from the start if you're serious.
What's the biggest hidden cost in DIY? Latency and reliability tuning, plus on-call. Getting an agent to respond in under a second and not break mid-call is where months disappear.
How fast can a white-label offering actually go live? Configuration and branding can be done in around 48 hours. The longer pole is your go-to-market - packaging, demo, first pilots - not the tech.
The Voxtell angle
Voxtell is built for exactly this decision. It's a white-label AI voice platform made for MSPs and telecom resellers - your brand, your domain, your pricing, with the infrastructure, latency tuning, and compliance handled for you. You can be live under your own name in about 48 hours, and we include sales enablement so you're not just handed a login and wished luck. You sell the relationship. We stay invisible.