What does an AI automation agency actually sell?
An AI automation agency makes its money from scoped installation fees and recurring care plans, not from marked-up software seats. The deliverable is usually a working workflow—intake, qualification, booking, follow-up or support—built on infrastructure the client owns. A $4,000–$15,000 install can take two to four weeks; a $600–$2,500 monthly care plan covers monitoring, updates and small changes. The agency is paid for design, integration and accountability. That is a services business with software-like retention, not a reseller.
Why do install fees beat retainer-only models?
Retainers alone rarely fund the build. A client paying $800 a month expects value from week one, before the integration is stable. Charging separately for discovery, configuration and handover aligns cash with work and filters out buyers who will not invest. A typical split is 60–70% of first-year revenue from installation and 30–40% from care. That mix lets a small team stay cash-positive while the recurring base grows. Discounting the install to win the retainer usually creates a delivery backlog and weak margins.
Where do delivery costs eat the margin?
Delivery cost is mostly senior time. Discovery, systems mapping, prompt design, testing and client training can consume 25–40 hours on a simple receptionist and 80–150 hours on a multi-system support agent. At a $100–$150 blended hourly cost, a $5,000 project leaves little room for rework. Agencies lose margin when scope is vague, when the client cannot supply clean data, or when every change becomes a free favour. Fixed scope, a named owner and a change-request rate protect the model.
How should an agency price ongoing care?
Price care against the cost of the human process it supports, not against model tokens. If an agent handles after-hours intake that would otherwise need a $1,200 monthly answering service, a $900 care plan is easy to justify. Include a defined support window, a monthly change allowance and a clear escalation path. Avoid unlimited prompt edits. Charge setup separately for new workflows. Review pricing every two quarters, because model and integration costs move and scope creeps quietly.
What breaks the unit economics at ten clients?
At ten clients, the founder becomes the bottleneck. Each account may need two to six hours a month for monitoring, fixes and reviews, plus unpredictable incidents. Twenty clients can consume a full-time engineer before revenue supports one. The fix is standardisation: reusable templates, one orchestration stack, documented runbooks and tiered support. Track gross margin per account, hours per install and repeat-incident rate. Fire accounts that generate more custom work than they pay for. Bain's 2026 warning about growing AI budgets without returns applies directly here.
What does a defensible 2026 agency model look like?
A defensible agency sells outcomes with visible scope: fewer missed calls, faster response, cleaner intake and documented handoffs. It publishes a baseline, agrees a target and reports monthly. It uses agentic patterns where they help—tool use, memory, escalation—and avoids autonomy where auditability matters. MIT Sloan's 2026 explainer is a useful client primer. Ownership matters commercially: when the buyer owns the code and data, the agency cannot hold the business hostage, and renewal depends on service quality rather than switching costs.
Frequently asked questions
How much should an AI automation agency charge in 2026?
Most small-business installs land between $4,000 and $15,000, with care plans at $600–$2,500 a month. Price by workflow scope, integration count and support level, not by hours alone. Charge discovery separately if the client's systems are undocumented.
What gross margin is realistic?
Aim for 55–70% gross margin on delivery once templates are reusable. First projects often run lower because discovery and rework are underestimated. Track margin after founder time, subcontractors and tooling.
Is a retainer-only model safer?
No. It delays cash and encourages unpaid build work. Use an install fee plus care plan, with a clear change-request rate and a defined support window.
Want the full playbook?
The free 2026 AI Revenue Playbook turns these shifts into a plan with costs, scripts and pricing templates.