A voice AI agency running ten clients on $1,500/month retainers is billing $180,000 a year. At a 40% target margin, that is $72,000 in operating profit before owner salary. Most agencies at this stage believe they are somewhere in that range.
Most are not.
The actual margin, once provider fees, maintenance hours, call volume overages, and support time are counted honestly, is usually closer to 15 to 25%. On $180,000, that is the difference between $72,000 and $27,000.
The gap follows a pattern. It is not random, and it gets wider as the agency adds clients.
What Does the P&L Look Like on Paper?
A typical agency models it something like this: take the monthly retainer, subtract the direct provider cost, and call the remainder margin. For a $1,500 retainer with $200 in Vapi or ElevenLabs charges, that looks like an 87% gross margin before overhead. The agency feels profitable. The model says it is.
The model is missing most of the costs.
Direct provider fees are only one line item. The rest are costs the agency either tracks loosely or does not track at all: the hour spent troubleshooting a missed call event, the afternoon fixing a webhook that stopped delivering after a provider update, the time spent pulling a usage report when a client asks why their bill looks different this month.
These are real hours. They just do not show up on an invoice, so they do not show up in the margin calculation.
Where Does the Margin Actually Go?
Track the actual hours for one month across ten clients and the picture changes quickly.
Provider costs are not fixed. A $200 monthly provider estimate assumes a predictable call volume. It often is not. One client running a campaign will send call volume up three to five times in a week. If the retainer does not have an overage clause, that spike comes out of the agency's margin, not the client's budget. A 5x spike on a $200 monthly estimate is a $1,000 line item the agency was not expecting.
Support time compounds. At five clients, managing issues takes maybe two to four hours a month per client. At fifteen clients, that does not stay linear. New clients ask more questions. Mature clients surface edge cases. Something breaks for one client at the same time something else breaks for another. Four hours per client per month across fifteen clients is sixty hours. At even a modest $75/hour internal cost, that is $4,500 a month in support and maintenance time that most agencies are not billing for.
The integration layer needs maintenance. Every time a provider updates its webhook schema, changes a field name, or alters how call events are structured, something downstream breaks. For an agency running multiple clients across one or two providers, fixing it once is trivial. Fixing it across fifteen clients, verifying that each client's automations are still working, and confirming that no call data was lost in the window takes most of a day.
The integration tax is what this kind of work is called. The first time it happens, it costs a few hours. By the twelfth time, it is a permanent line in the operating calendar.
Onboarding is underpriced. Most agencies charge a setup fee of $500 to $1,000 for a new client. The actual time to get a client live, including discovery, configuration, testing, and the first two weeks of issue resolution, is often fifteen to twenty hours. At $75/hour, that is $1,125 to $1,500 in cost against a $750 average setup fee. The agency is subsidizing every new client acquisition without realizing it.
What Changes When an Agency Reaches 15 Clients?
Five clients is a manageable operating system. Fifteen is not the same system at three times the size. It is a different system that nobody designed.
At five clients, one person can hold the context for everything: which client uses which provider, which automation receives which call type, which client is sensitive about response time. At fifteen, that context does not fit in one person's head. Work falls through the gaps. A client's call volume spikes and nobody notices until the invoice arrives. A webhook issue affects three clients and the fix gets applied to two of them.
The agency is still billing the same retainers. The cost to deliver those retainers has gone up significantly. The margin has compressed, quietly, over several months.
Building out the pricing structure correctly helps, but pricing is only part of the answer. The other part is the operating model: how the delivery system is built so that the fifteenth client does not cost more to manage than the fifth.
What Actually Fixes the Margin Gap?
The agencies that hold their margins at scale have usually done one of three things.
They made provider cost exposure predictable. This means overage clauses in client contracts, usage monitoring that flags spikes before the billing cycle closes, and a clear policy on what happens when a client campaign sends volume outside the defined scope.
They standardized delivery. Each client does not get a custom setup. There is a template: provider configuration, webhook routing, automation structure, reporting format. The fifteenth client takes the same setup path as the fifth. Support questions have documented answers. The hours are capped because the delivery model is capped.
They separated client operations structurally. When each client lives in its own operational lane, a problem with one client is a problem with one client. It does not require checking whether the same issue is affecting five others. It does not require tracing which automation belongs to which client. The cost of maintaining the integration layer drops when the structure enforces separation instead of relying on someone to keep track.
Frequently Asked Questions
What margin should a voice AI agency target?
A sustainable target for a well-run voice AI agency is 30 to 40% net margin, accounting for provider fees, support time, and infrastructure maintenance. Most agencies achieve this number only after tightening their delivery model, adding overage clauses to contracts, and either building or buying infrastructure that does not require per-client custom maintenance.
How much do provider costs affect agency margins?
Provider costs typically run $100 to $300 per client per month depending on call volume and provider choice. The risk is not the baseline cost. It is the spike. A client campaign that drives five times the normal call volume will add $400 to $1,000 in unexpected provider fees if the retainer does not account for overages. Over ten clients, one or two spikes per quarter is a structural margin risk.
What is the integration tax and how does it affect agency profitability?
The integration tax is the cumulative cost of maintaining custom infrastructure around a voice AI provider stack. It includes time spent adapting to provider updates, routing calls across client automations, debugging webhook failures, and keeping each client's data correctly separated. It starts small and grows with client count. For most agencies, it becomes a meaningful cost driver somewhere between client eight and client twelve.
Voxfra handles the operating layer around voice AI providers, keeping each client's calls, data, and automations structurally separated so agencies are not paying the integration tax on every new retainer. See how agencies use Voxfra.