Your AI agent is quoting prices you never set. Let's fix that today.
We deploy and audit AI agents that answer customers for restaurants, clinics, real-estate offices and home-services businesses. Almost every agent we open has the same four holes — and none of them need a new model to close.
No signup on the audit tool. Paste your agent's instructions, get the five failure points it's carrying into production.
Four failures. Every single time.
These are the things that break after go-live. They rarely show up in a demo, because a demo is a friendly conversation and a customer is not.
- It invents a price, discount or policy. Ask an LLM for a number it was never given and it produces the most plausible-sounding one. Your customer cannot tell the difference until they arrive.
- It escalates, and the customer has to start over. The agent hands off a bare "customer needs help" — so the human reopens the conversation blind and the customer retypes everything.
- Nobody tested it with a hostile message. Friendly tests pass on every prompt. "What's the real price if I pay cash?" is the one that finds the edge.
- A missed call becomes someone else's sale. The reply goes out four hours later, after the lead has already booked with the next business on the list.
The fix is not a better model.
It is four small pieces of operational discipline. All four are documented free on this site, with the actual text you can copy.
A 3-section facts file
[FACTS], [RULES], [ESCALATE IF] — one plain-text file and one load-bearing rule that turns your agent from a confident guesser into a careful reader.
20 hostile messages
Grouped by failure type — price fishing, policy invention, authority pressure, off-topic traps, data fishing — each with a pass criterion. Twenty minutes of testing buys the confidence that day one won't produce a refund you have to honour.
Read the guide → HandoffA 5-field escalation payload
Contact, summary, what was tried, what's open, suggested reply. With those five fields a human answers in about ten seconds — because they stopped reconstructing context from scratch.
Read the guide → RevenueA 3-touch recovery sequence
The exact timing, the word-for-word messages, and the discipline rules that stop it sounding like spam. Answer the question first; one question per message; never write "just following up".
Read the guide →Free audit tool: find your agent's five holes in 60 seconds
Paste the instructions, FAQ or policy text you gave your agent. We run it against the same five failure classes we check in paid engagements and tell you what is missing — and what to write to close it.
Agent Readiness Audit
Checks for: a quotable source of truth · an explicit "quote only what's listed" rule · escalation triggers that stop answering before they hand off · a handoff payload · a hostile-test plan. Returns a prioritised fix list, not a score.
Run it on your agent →Guides, by the failure you're actually having
Twenty field notes. No email required, no gated PDF, no "request access". If a guide solves it, you never need to buy anything.
Your agent quoted a price you never set
The gap-filling problem, the eager-helpfulness problem, and the one file that fixes both.
8 min read → ChecklistThe pre-launch agent audit
Twelve checks we run before an agent is allowed to talk to a paying customer. Grouped into source of truth, guardrails, handoff, data and monitoring.
9 min read → Industry playbookRestaurant booking agent
Set-menu minimums, deposit rules, private hire, and why it must never answer an allergen question.
8 min read → Industry playbookClinic & dental scheduling agent
Insurance, new-patient paperwork, no-show fees — and the hard line where it stops being an admin tool.
8 min read → Ops playbookGetting the budget approved
Why "it saves time" gets rejected, and the four numbers that survive a sceptical owner's questions.
7 min read → All twentySee every guide
Handoffs, refund disputes, missed leads, tool choices, receptionist limits, channel choice, law-firm intake and more.
Browse all →If you'd rather not write it yourself
The guides give you the method for free. These are the fill-in versions — the same structures, already written out, so you edit instead of invent.
Facts-File Template Pack
The fill-in version of the three-section facts file, structured per industry. The method is free in the guides — this is the same file already written out, so you edit instead of invent.
See the pack →Missed-Lead Recovery Kit
The intake sheet, the three-touch sequence with the timing already set, and the "don't lose a lead" checklist. Described free in the missed-lead guide.
See the kit →AI Ops Playbook — 32 lessons
The full deployment path for an owner or ops lead doing it themselves: scoping, source of truth, guardrails, launch, handoff, and the metrics that justify it.
Look inside →All toolkits are sold through our Whop store. One-time purchase, no subscription. No earnings claims — these are operational templates, not a money-making system.
How we work
Independent practice. We deploy agents and we audit agents other people deployed. The guides on this site are the same material we use on client work, which is the only reason they're worth reading.
We tell you when you don't need an agent
A saved reply template and a shared inbox fixes more "AI" problems than an LLM does. We say so in the guides, and we say it on calls.
We automate intake, not judgement
Bookings, hours, first replies, status questions — yes. Refunds, complaints, anything emotional or high-stakes — a human decides, every time.
No affiliate filler, no vendor pitch
We're not affiliated with any platform we mention. Where a comparison exists, you get the trade-off, not a referral link.
Start with the thing that's costing you money this week
If your agent already made a promise it shouldn't have, that's an urgent problem. If it hasn't yet, you have about a week. Either way, the facts file takes twenty minutes.
Fix the invented-price problem →