Sites that load fast and convert
Custom builds and integrations
Bounded, auditable, human-approved
Being found, staying fast, staying safe
Most people have been trapped in a chatbot loop and have not forgotten it. The difference between a support bot customers tolerate and one they resent is not how clever it sounds — it is how quickly it recognises it is out of its depth and gets a person involved.
A well-built one handles the genuinely repetitive questions — where is my order, what are your hours, how do I reset this — and passes everything else on with the conversation already summarised, so nobody has to explain themselves twice.
Watch it hand over ↓The line
Drawing this line is the entire design job. Get it right and customers barely notice the automation. Get it wrong and you have built the thing everyone complains about.
The temptation is always to widen the left column. Resist it. A bot that handles sixty per cent of contacts brilliantly is worth far more than one that attempts everything and frustrates a third of the people who use it.
Expectations
Vendors quote containment rates in the nineties. Those numbers usually count a conversation as contained if the customer gave up. Drag the slider and look at the shape rather than the headline.
Assumes the bot resolves about eight in ten of the questions genuinely within its scope — a realistic figure for a well-grounded system, not a best case. The remainder escalate, which is the system working correctly rather than failing. The goal is never one hundred per cent. A bot that refuses to hand over is how you lose customers who were only mildly annoyed to begin with.
The stack
Nobody wants another inbox. The handover lands in the help desk your team already works in, and conversations sit alongside the ones that arrived by email.
Channels
Help desk & CRM
Commerce & scheduling
Models & runtime
Channels
The same knowledge and the same handover rules, surfaced wherever your customers already are. One system, several front doors.
The obvious one, and still the highest volume. Appears where it is useful rather than ambushing everyone on arrival.
Best starting point for most businesses
Where a large share of customers would rather talk. Conversations persist, so nobody re-explains anything a week later.
Strong for retail and trades
Not a chat window at all. Incoming email classified, enriched with account context, and answered or drafted for approval.
The quiet, high-value one
For simple status questions and appointment traffic, where opening an app is more effort than the question deserves.
Bookings, deliveries, service calls
Pointed inward instead of outward. Staff ask about policy, process and accounts without interrupting a colleague.
Often the easiest first launch
Calls answered and qualified out of hours, with a transcript and summary waiting for whoever picks it up next morning.
See Voice & Phone AI Agents
Interactive
This is a real conversation we have at the start of every support project. Switch rules on and off and watch six live conversations get routed. There is no correct configuration — there is only the one that matches your risk tolerance.
Turn everything off and the bot attempts all six, including the refund and the person who has already asked twice. That configuration is how support bots earn their reputation.
Escalation rules
handled by bot — escalated —
Honesty
Every one of these is a decision somebody made, not a limitation of the technology. Which means every one of them is avoidable.
The single biggest cause of resentment. If there is no obvious route to a person, customers stop trusting the whole channel — including the parts that worked.
A fake name and a stock photo. People work it out within two messages and feel handled. Say what it is; nobody minds a bot that is upfront.
Producing something plausible instead of admitting uncertainty. One confidently wrong answer about a policy costs more than fifty deflections saved.
The customer explains everything again to the agent. This alone undoes the entire time saving and reads as contempt.
A pop-up three seconds after the page loads, before anyone has read anything. It is the digital equivalent of being followed around a shop.
Endless button trees where nothing matches the actual question. If it cannot handle free text, it is a phone menu wearing a chat window.
Asking for an order number that was supplied two messages earlier. Small, and it destroys confidence faster than almost anything else.
Deployed over documentation that is thin, contradictory or years out of date. The bot did not invent the confusion; it exposed it.
Nobody reads the transcripts, so the same failure repeats for months. Support automation is a system that needs tending, not an installation.
Scope
It begins with reading your existing conversations. Everything worth knowing about what to automate is already in them.
Reading a real sample of your past conversations to find what genuinely repeats, how customers actually phrase it, and where agents spend their time.
An explicit written list of what the bot will attempt and what it will always pass on, agreed before anything is built.
Connecting your policies, help articles and past resolutions so answers come from your material with a source, not from a model's general impression.
Order lookup, account status, booking availability. Without these it can only recite policy, which is a fraction of the value.
The rules, the summary format the agent receives, and the routing. This gets more attention than the answering does.
Wording that sounds like your business, including how it declines, apologises and admits it does not know.
Run against actual past conversations with known outcomes before a single customer sees it.
Volume, resolution rate, escalation reasons, and the questions nothing could answer — the last being the most useful of the four.
A scheduled read of transcripts after launch, because the first month teaches you more than the whole build did.
Questions
Yes, because we tell them. It is labelled clearly and never given a fake human name or photograph.
People are entirely comfortable with a bot that is upfront and gets them to a person quickly. What they object to is discovering they were misled.
The design goal is that it declines rather than guesses. Answers are grounded in your documentation with a confidence threshold, and anything below it becomes a handover instead of an attempt.
Every conversation is logged, so when something does go wrong you can see exactly which passage caused it and correct the source.
Not in any project we have been involved in. What changes is the mix — the repetitive volume drops, and the work that remains is the difficult, judgement-heavy kind that people are better at anyway.
Teams generally report the job improving. If your goal is specifically headcount reduction, we are probably the wrong fit and would rather say so now.
A first version handling a defined set of questions on one channel is usually weeks, not months. We start deliberately narrow.
The first few weeks of real conversations tell you more about what to automate next than any amount of planning would.
Yes, and that is where most of the value sits. Reciting policy is easy; answering "where is my order" from live data is what actually reduces volume.
It needs a connection to your commerce or CRM system, and identity verification appropriate to what it is exposing.
It keeps answering what it can and collects everything it cannot, so the queue on Monday morning arrives triaged and summarised rather than raw.
For many businesses this alone justifies the project — questions get answered at nine in the evening instead of waiting for morning.
Some, and it rarely needs to be as much as people fear. A handful of clear, current policy pages beats a large archive nobody has updated.
Part of the work is identifying which gaps matter, and the transcript analysis usually answers that in the first week.
It depends on where your team already works. If you have a help desk with a good API, the bot lives inside it. If not, we build the layer and connect it to your channels.
We avoid being locked to one vendor's proprietary builder — the knowledge, the rules and the transcripts should remain yours and portable.
Call back
Leave a number and a good time. We will call you back to talk about what you are trying to build — no charge for the conversation.
Or call us directly +1 (647) 385-5532