The AI Loadout

Best AI setup for an automation agency

Updated · What changed: First published with a workflow, voice receptionist and retainer-first operating model.

By The AI Loadout editors · How we pick

The best AI automation agency setup is n8n for workflows, Retell for the voice receptionist, Claude Sonnet 5.5 for building and testing, and The Boring Consultant persona to keep the offer narrow. Use Ollama or LM Studio for local drafts and private client notes. Sell one measurable outcome, price the maintenance honestly, and link the receptionist work to a call, booked appointment or qualified lead that the client can see.

SlotPick
Tooln8n
ModelClaude Sonnet 5.5
SkillRetell AI voice receptionist
PersonaThe Boring Consultant

Free, no signup needed.

The loadout (why each pick)

Tool: n8n

n8n gives an agency control over triggers, credentials, branching and logs, and it can be self-hosted when a client needs the data on its own infrastructure. Make is the easier visual alternative. Start with a repeatable workflow template and keep client-specific changes in documented configuration, not scattered custom nodes.

Model: Claude Sonnet 5.5

Use Claude or GPT to map the client's process, write the first workflow, test edge cases and turn call transcripts into a short change list. Give it the actual business rules and sample records. It should produce a test plan and a handoff document, not quietly invent a policy.

Agent skills

  • Retell AI voice receptionist: Retell gives you the phone layer, call flows, tools and transcripts for a custom receptionist. Vapi is a solid alternative when you want a more provider-neutral build. Either way, test transfers, after-hours behavior, consent language, emergency phrases and the exact booking handoff.
  • Existing AI receptionist offer: Use the site's AI receptionist setup as the client-facing baseline for call scripts, pricing models, trial calls and the limits of a ready-made receptionist. Do not promise custom behavior before testing the real number.
  • Ollama or LM Studio local build notes: Keep private client notes, workflow drafts and test fixtures local when a hosted model is not appropriate. Ollama fits terminal work; LM Studio fits desktop review. A local model still needs the same human review and test cases.

Bot persona: The Boring Consultant

This is an inspired-by operating approach, not a real person. It sells one outcome, keeps clients on retainers, documents handoffs and refuses custom one-offs that cannot be supported.

Comparison table

OptionBest forModel(s)PriceFree planLast checked
n8nTechnical agencies that want control or self-hostingYour choice of model and toolsFree self-hosted; Cloud plans from €20/mo yearlyYes
MakeVisual workflows and fast client demosAI modules and your providerFree with limits; paid plans from $9/moYes
Retell AICustom phone agents with transcripts and toolsYour choiceUsage based; check current pricingNo
VapiAffiliate link. We earn a commission if you buy, at no cost to you.Provider-flexible voice agentsYour choiceUsage based; check current pricingAffiliate link. We earn a commission if you buy, at no cost to you.No
Ollama or LM StudioLocal drafts and private test fixturesQwen 3.5, Gemma 4Free (your hardware)Yes

Prices come from the linked vendor or source page and can change. Check the vendor before you buy.

Copy this setup

Paste this persona into your tool's custom instructions or project instructions:

The Boring Consultant prompt
You are The Boring Consultant for a small AI automation agency. Sell one measurable outcome to one type of local business. Start every proposal with the current process, the baseline number, the target number and how the client will verify it. Prefer a monthly retainer around maintenance, monitoring, small changes and a documented handoff. Refuse custom one-offs that do not fit the template or that have no owner on the client side. For each workflow, list triggers, inputs, tools, permissions, failure paths, human approvals, logs, test cases and rollback steps. For voice receptionists, test greeting, identity disclosure, recording consent, emergency phrases, transfers, after-hours calls, booking, missed calls and escalation. Never promise that AI will replace staff, never hide usage costs, and never leave a client without the credentials, diagram and runbook needed to operate the system.
  1. Pick one local-business niche and one result, such as more answered calls that turn into booked estimates.
  2. Build one n8n workflow and one Retell or Vapi receptionist against a fake business before you sell it. Keep the configuration separate from the template.
  3. Use Claude or GPT to write the workflow, then run real-looking edge cases and inspect every tool call, transcript and failure path.
  4. Sell a setup fee plus a 200 to 300 GBP monthly retainer for monitoring, maintenance, small changes and reporting. Put usage charges in writing.
  5. Hand over a diagram, credentials map, test log, runbook and named client owner. Refuse the next custom request until it fits the supported offer.

Alternatives by situation

  • Plumber, roofer or HVAC company: Sell one result: fewer missed calls and more booked estimates. Start with after-hours intake, a texted booking link and a daily report. Do not let the agent diagnose the job or promise an arrival time.
  • Client wants a custom app: Price it as a separate product or decline. If the request cannot reuse the supported workflow, it will consume the retainer and make the next handoff harder.
  • Client asks for a cheap one-time build: Show the maintenance list. Credentials expire, APIs change, prompts drift, phone calls expose edge cases and someone has to watch failures. A retainer is the product, not an apology for unfinished work.
  • n8n versus Make: Use Make for a fast visual proof and n8n when self-hosting, branching or long-term control matters. Keep the client on the platform you can support at the price you quoted.
  • Voice calls need judgment: Transfer angry, urgent, medical, legal, payment and unusual calls to a person. Test recording consent and emergency handling for the client's location before launch.

What changed

  • : Page launched with n8n, Retell, a frontier model, local drafting and The Boring Consultant persona. Sources are listed under Sources.

Full changelog

FAQ

Do AI automation agencies make money?

Some do, but the durable offer is usually a narrow result with recurring maintenance, not a promise to automate everything. Reports from agency operators describe many quitting around month five because maintenance and churn overwhelm one-off project revenue.

What should an AI agency sell first?

Sell one measurable result to one niche. An AI receptionist that answers after-hours calls, captures the lead and reports booked appointments is easier to test and price than a general AI employee.

How much can I charge for an AI receptionist retainer?

A practical starting point is around 200 to 300 GBP a month for monitoring, small changes, reporting and support, plus any usage or setup fees you clearly disclose. The right price depends on call volume and the cost of a missed lead.

Should I use n8n or Make?

Use Make for fast visual builds and n8n when you need self-hosting, deeper control or a workflow you can support at volume. The better choice is the one your team can monitor and hand off.

Should an agency use a local model?

Use Ollama or LM Studio for private drafts, test fixtures and client notes when the client requires local processing. Keep the production model decision tied to quality, latency, cost and the client's data requirements.

Sources

Disclosure: links marked "affiliate" may earn us a commission, and sponsored slots are always labeled. Neither affects our picks. Read our methodology.

Related setups

Related skills and personas

Agent skills and bot personas that pair well with selling AI workflows and AI receptionists to local businesses.

Browse full directories:All Skills (>35)·All Personas (>15)

Want tested skills and personas installed in one command? See the AI Loadout packs, from $9.