Everything we learned fixing AI agents that talk to real customers.
Each guide solves one specific failure — an invented price, a lost lead, a handoff that makes the customer repeat themselves. Written from client work. Free, no email required.
Your AI agent quoted a price it was never given. Here's how to stop it.
An agent that invents a price sounds exactly like one that doesn't. The fix isn't a bigger model — it's a plain-text facts file with one load-bearing rule, and a ten-second update habit.
8 min read →Test before launchTwenty hostile messages that find your chatbot's failure modes before a customer does.
Your chatbot passes every friendly test you throw at it — that's the problem. Here is the 20-message suite I run before launch, grouped by failure type, each with a pass criterion a non-technical person can judge.
8 min read →Handoff that holdsThe AI escalated perfectly — and the human still asked the customer to repeat everything.
Routing decides who answers. The handoff decides what they know when they open the thread. Five fields turn a six-minute reconstruction into a sixty-second reply, and stop the customer repeating themselves a third time.
8 min read →Start hereThe 12-check audit I run on every AI agent before it goes live.
An agent that has never been audited is the most common state I find — live, confident, and quietly quoting wrong prices. Here is the 12-point pre-launch checklist I run, grouped into five areas, each with a pass condition.
9 min read →Industry playbookYour restaurant AI booking agent needs a facts file before it takes one reservation
A booking agent that quotes the wrong set-menu price or invents a deposit policy costs you a table and a review. Here is the facts file that stops it before it takes a single reservation.
8 min read →Industry playbookYour real estate AI chatbot will improvise a price the seller never agreed to
An AI assistant answering listing enquiries will fill any gap you leave — and in real estate the wrong sentence is a licensing and fair-housing problem, not a typo.
8 min read →Industry playbookA clinic AI receptionist books appointments — it must never answer a medical question
A clinic receptionist agent should book appointments and answer administrative questions. The moment it reassures a patient about a symptom, it has crossed a line nobody can put back.
8 min read →Honest guideWhat an AI receptionist can do, and what it should never be allowed to touch
Yes, an AI can answer your business phone — for a specific, listable set of calls, and not for the ones that decide whether the customer stays. Here is how to draw the line.
8 min read →Ops playbookThe missed-lead recovery playbook: a three-touch sequence you can run today
A missed call or unanswered DM is not a lost lead yet. Here is the four-column intake, the exact timing, and the word-for-word templates that get them back — no AI agent required.
7 min read →Budget caseHow to calculate AI ROI from your own numbers — and get the budget approved
An owner or CFO does not want a productivity story. They want four numbers. Here is how to calculate AI ROI from your own costs and present a one-page case that survives scrutiny.
8 min read →Risk controlStop your AI agent from creating refund disputes
An agent that promises a refund it cannot authorize turns a support tool into a liability. Here are the six cases it must always escalate and the rule wording it can actually follow.
7 min read →ToolingZapier vs Make vs n8n: the three questions that actually decide it
License price is the wrong thing to compare. Who maintains it in six months, how many branches the workflow has, and where the data is allowed to live will decide it for you.
8 min read →How to write itWhat belongs in an AI agent's system prompt, and what must stay out
Most system prompts fail because they mix permanent instructions with facts that change weekly. Here are the five blocks that belong, the four that don't, and a full annotated template.
8 min read →Channel choicePick the channel before you pick the platform
The channel decides more about your agent's behaviour than the model does. Where your customers already are, what each channel is mechanically bad at, and why you should launch one, not four.
8 min read →EcommerceOrder status is the first query to automate, and the safest
"Where is my order" is the highest-volume, lowest-risk question a small shop gets. Here is what a WISMO agent needs, the three answers that cover most of it, and when to escalate.
8 min read →Field playbookAn AI agent for contractors that triages emergencies and captures quotable leads
Home services break normal chatbot designs: emergencies can't wait, jobs can't be priced blind, and geography matters. Here is the triage, intake and dispatch setup that works.
9 min read →Buying guideWhat an AI agent actually costs a small business — all four cost lines, not just the monthly fee
The monthly fee on the quote is one of four cost lines. Here is the full structure — including the two that show up on no invoice — and a worksheet to price it from your own inputs.
8 min read →Decide firstIs a chatbot worth it for your small business? Score it before you buy it
The answer depends on three conditions, not on the product. Here is the scoring sheet, the honest no, and the cheaper alternatives that win more often than anyone admits.
8 min read →MetricsContainment is not resolution — how to measure what your AI agent actually did
"Handled without a human" is the most flattering and least honest number in an agent report. Here is the difference, a transcript-sampling method, and how to report both numbers without lying.
8 min read →Regulated nicheThe law firm intake chatbot: describe the firm, book the meeting, stop there
An intake agent is a receptionist, not a lawyer. Here is what it may say, the lines it must never cross, a facts file you can adapt, and the disclaimer you should read first.
9 min read →