AI is not clever. That is why it works.
A plain English guide for business owners who have never used AI, written by the people who build it. No jargon, no acronyms, nothing to install. Read the first two sections and you will understand what this stuff actually is, and what it is worth to you. Read section 16 and you will be able to take apart any pitch you get, including ours.
About twenty minutes end to end, and built so you can stop after any section. Last updated 27 July 2026.
If you read nothing else
- AI does not think. It predicts the sensible next thing, very fast, all night, without getting bored.
- That makes it good at obvious work done at volume, and bad at judgement.
- It knows nothing about your business until someone connects it and tells it. That work is most of what you are actually paying an agency for.
- It can be confidently wrong. Building it properly means the wrong answer never reaches your customer.
- You can work out what it is worth to you before you spend anything. Section 02 shows you how, with your own numbers, and tells you when the answer is "do not buy this".
Contents
It predicts. That is the whole trick.
Everything else you have heard is built on top of this one idea.
Start a text with "sorry, running" and your phone offers "late". It is not reading your mind. It has seen that sentence finish that way thousands of times.
Now imagine the same trick, done by something that has read more writing than a person could get through in a thousand lifetimes. That is modern AI. Give it a customer's email and it can work out what they are asking for, check the real diary, and write a reply in the way your business actually speaks.
There is no thinking happening. Nothing in there wants anything. It is prediction, running at a speed and volume no person can match.
That sounds like a limitation. It is the point. Most of the work that falls through the cracks in a business is not clever work. It is obvious work that nobody got round to. The enquiry at 9pm. The fifth follow-up. The quote nobody chased. Something that does the obvious thing every single time, at three in the morning, on the four hundredth message of the month, without getting bored, is worth real money to a business your size.
"If you can explain the job to a new starter in one page, AI can probably do it. If it needs a year of experience and a gut feel, it cannot."
Work out your own number first
Ten minutes, your own inbox, no supplier in the room. If the number comes out small, you can stop reading and we have saved you a fortnight.
Step one. Open your enquiry inbox, your WhatsApp and your website form submissions for the last full month. Count the ones that arrived outside your working hours. Timestamps only. Do not read them.
Step two. Of those, count how many had a reply within an hour of arriving.
Step three. Work out what share of enquiries you normally win. If you do not track it, estimate, and estimate low.
Step four. The rest waited until you opened. You will still win some of them, so do not count them all as lost. Take the ones that waited, multiply by your win rate, multiply by your average job value, then take a third of the answer. That third is roughly what the waiting costs you each month.
Roughly what the waiting costs you each month, on a deliberately cautious assumption:
£850 per month
Nothing here is stored or sent anywhere. The sum happens in your own browser.
The one third is us being careful on your behalf. If you think you lose more than a third of the people who wait until morning, put your own fraction in. We would rather this number came out too small than too big.
40 enquiries out of hours, 6 answered within the hour, 25% you would normally win, £300 average job. The 34 that waited are worth £2,550, and a third of that is £850 a month. Those are made-up inputs to show the sum, not a client result. Put your own in.
The honest bit
Our entry product carries a build fee and a monthly one, both quoted at the audit. If your monthly number is not comfortably above the monthly fee, do not buy it. Not from us, not from anybody. Fix your Google Business listing and your reply times first, because both are free and both will do more for you.
If it only just clears it, remember the build fee has to earn itself back too, so give it a few months before you judge it. The useful thing about that gap is that you can measure the whole of it before you spend a penny. Any supplier unwilling to do this sum with you, out loud, before quoting, is hoping you do not do it at all.
Four things AI is not
Most of the fear and most of the hype come from these four misunderstandings.
It is not a separate system to learn.
There is nothing to install, nothing arrives in a box, and there is nothing new for your team to log into. It sits behind the website, the inbox and the WhatsApp number you already have.
It is not conscious.
It has no opinions, no motives, and no memory of you unless somebody gives it one on purpose. It will not develop a mood, and it will not take against a customer who annoyed it on Tuesday.
It is not connected to your business.
Out of the box it knows nothing about your prices, your diary, your policies or your team. Everything it knows about you, somebody has to connect or write down. That unglamorous work is most of what an agency is actually for, and it is where the quality comes from.
It is not always right.
It can produce a confident, well-written, completely wrong answer. This is the most important line on this page. Building AI properly is mostly the work of making sure the wrong answer never reaches your customer. Any supplier who skips past this is selling you a problem.
Computers learned to read messy English
Nothing about your business changed. What changed is what software can understand.
Software has followed instructions for fifty years, as long as the instructions were exact. Tick the box. Click the button. Fill the field. The moment a human went off script, a person had to step in. That person was you, or whoever was nearest the inbox, or nobody.
Here is the sort of enquiry that lands on a dental practice's website at half past ten at night:
"hi do u do invisalign and how much roughly, im in Hackney"
No punctuation. A brand name. A vague question. A location. Old software sees nonsense and hands it to a human. Current AI sees a price enquiry about a specific treatment from someone local, and can answer it, ask the one question that decides whether they are worth a slot, and offer Tuesday at four.
That is the whole shift. Not intelligence. Understanding mess.
On your website, in WhatsApp, in your inbox
It goes where your customers already type. It does not go near your phones.
People imagine a machine answering the telephone. We do not do that, and we will say so plainly, because for most businesses it is the wrong place to start.
Your customers already type at you. They fill in the form on your website at half past ten at night. They message your WhatsApp on the way home. They email. They send a photo of a leaking boiler. Meeting them there is the job of an AI agent on your website and WhatsApp. That is where the work piles up, and that is where AI sits: on the page, in the app, in the inbox, and inside the documents you already keep.
If your problem genuinely is the telephone, we will tell you we are not the right firm for it. That answer costs us a sale and saves you a year.
What it is good at, and what it is not
Anyone can list the strengths. Choosing correctly is the part that matters.
Good at
- Reading two hundred enquiries and sorting them by what each person wants
- Doing the same job the same way, every single time
- Being awake at 11pm on a Sunday, and on Boxing Day
- Never forgetting the fifth follow-up
- Finding one line inside three hundred documents
- Writing the notes up, the same day, in your format
Not good at
- Decisions with no precedent to copy
- Anything where being wrong is expensive and nobody checks
- Work that has never been written down by anyone
- Reading a room, or a relationship
- Judgement about people
- Being accountable. Accountability stays with you, always
How to choose, in one rule
Score each task on two things: how often it happens, and how much damage a wrong answer does.
Often, and low damage, is where AI pays first. Answering enquiries. Chasing quotes. Sorting documents, which is AI document automation. Confirming appointments. Updating records.
Rarely, and high damage, is where you keep a human, permanently. The complaint from your largest client. The clinical call. The legal opinion. The price exception.
Often, and expensive to get wrong, is the interesting one, and it is where most of the money is. That 10:30pm question about treatment prices is exactly it. You can automate that, but only with a hard boundary: it answers from your written prices and policies and nothing else, and it hands over the second the question goes outside them. If a supplier will not tell you where that boundary sits, that is your answer.
Start with the first. Never start with the second. A supplier who wants to begin with your hardest, rarest, most sensitive task is either inexperienced or selling you a demonstration.
Every word they will use at you
Keep this open the next time somebody pitches you. The third column is the one that matters.
The words everyone uses
| Word | What it means | Why you would care |
|---|---|---|
| AI | A broad label for software that handles work which used to need a person. On its own it means almost nothing | If a supplier only ever says "AI", ask them what it does in a sentence with no acronyms in it. If they cannot, there is nothing underneath |
| Model, or LLM | The engine that reads and writes text. The ones behind ChatGPT and Claude are the best known | You do not buy one directly. They sit underneath whatever you buy, the way an engine sits under a van |
| ChatGPT | A website where you type and it answers | Useful to you personally. It is not connected to your business, so it cannot see your diary, quote your prices or book anyone in |
| Prompt | The written instructions it works from | It is a job description. Vague instructions get vague work, exactly like a new starter |
The words that describe what you might buy
| Word | What it means | Why you would care |
|---|---|---|
| Chatbot | The older kind. A script with buttons that breaks the moment a customer asks something the script did not expect | Most people's bad experience of "AI" is a chatbot. Test any supplier's on a deliberately awkward question before you sign anything |
| AI Agent | Software that does not only answer, it acts. Checks the real diary, writes the record, sends the confirmation | This is the difference between something that tells you the answer and something that gets the job done. Ask which one you are being sold |
| AI Superagent | Several specialist agents wired together and given one job description | This is our name for it, not an industry term, and you should treat any supplier's coined word the same way: ask what the parts are, and which of them you are actually paying for |
| Automation | A fixed sequence that runs when something triggers it. No understanding required | Cheaper and steadier than AI, and the right answer more often than people expect. A good supplier will sometimes tell you that you do not need AI at all |
| Integration | The plumbing that lets two systems read and write to each other | Without it, AI is guessing. With it, it is looking at your actual availability |
The words that tell you if they know what they are doing
| Word | What it means | Why you would care |
|---|---|---|
| Hallucination | When it states something confidently and it is wrong | The real risk. Managed by limiting what it is allowed to answer from and checking before anything reaches a customer. Not by hoping |
| RAG | Answering from documents you have given it, rather than from memory | The reason it quotes your price list instead of inventing one. Ask for it by name if you like, or just ask "where does it get its answers from" |
| Training data | What the engine learned from, before it ever met you | Business tools are normally set up so your information is not added to it, but that is a setting, not a law. Get it in writing |
| API | A socket that lets one piece of software plug into another | Worth knowing only so you can ask the useful follow-up: which of my systems are you plugging into, and who owns that connection if we part company |
Find your symptom, not your technology
Nobody wakes up wanting AI. They wake up annoyed about one of these.
| What you notice | How to check what it is costing you | What we would put in place |
|---|---|---|
| 1. Enquiries arrive at night and at weekends, and get answered the next working day | Ask your last three customers how many other firms they messaged that evening, and who came back first. Most owners are surprised by the answer | A website and WhatsApp AI Superagent that replies in under a minute at any hour, answers the question, and offers a time |
| 2. The same six questions, over and over | Count how many of this week's replies you could have copied and pasted from something already written down in your business | Answers drawn from your own price list and policies, with anything unusual passed straight to a person |
| 3. Quotes go out and nothing comes back | Look at your open quotes and count how many have had a second chase. Then a third | A follow-up sequence that runs on its own and stops the second they reply |
| 4. Information lives in PDFs, old emails and one person's head | Time yourself the next time somebody asks for a document from two years ago. Then work out what that job costs when the person who knows where it is happens to be away | Document and data intelligence. Ask a question in normal English, get the answer with the source next to it |
| 5. Reporting is an evening job at the end of the month | Check how long ago the thing you fixed last month first started going wrong | Reporting that assembles itself and arrives on a Monday |
| 6. Booked appointments, then no-shows | Count last month's, then multiply by what an empty slot costs you. That one is simple arithmetic | Confirmation and reminder sequences on the channel they used to book |
| 7. The same information typed into three systems | Watch yourself do it once and time it. Every rekey is also a chance to get it wrong | Automation between the systems, so it is typed once |
Rows one, two and four need AI, because in each of them something has to understand a sentence nobody wrote in advance. The other four are plumbing: fixed sequences that fire on a trigger, no comprehension required. Plumbing costs a fraction of the price. If you are in one of those four, say so on the call and we will quote you for plumbing.
Audit, build, run
We are an AI solutions agency. Three things, in that order, and we do all three ourselves.
Audit.
We go through how your business actually works, find where the money and the hours are leaking, and write it down with a commercial case for each recommendation. You own the report whether or not you build anything with us. The Vybzz AI Audit.
Build.
Most of the work is writing down how your business actually answers a customer, in your words and at your prices. The software comes after that. You approve what it is allowed to say before anybody outside your business sees it.
Run.
We manage it monthly, watch what it gets wrong, and refine it. Software you have to maintain yourself is not a solution, it is homework.
You can start anywhere in this. Most businesses start with one thing that is annoying them.
- AI Superagents. A full role or a whole department, covered.
- Website and WhatsApp Superagents. Answering where your customers already type.
- AI apps, software and websites. Built for your business rather than bought off a shelf.
- AI document and data intelligence. Answers pulled out of your own paperwork.
- AI automations. The plumbing between the systems you already pay for.
- AI lead response and follow-up. Speed to the first reply, then the chase.
- AI reporting and insight. Reporting that writes itself.
- AI audits. The paid diagnostic, and the front door to all of the above.
One job description, five specialists
Our biggest offer, explained without a single acronym.
An AI Superagent is not one clever thing. It is five specialists working as one, holding a single job description at a time. That is what we mean by managed AI agents that hold a whole role.
- 1 Enquiry. Reads what came in, on whatever channel it came in on.
- 2 Qualifying. Asks the one or two questions that decide whether this is worth your diary.
- 3 Booking. Looks at real availability and puts them in it.
- 4 Follow-up. Chases the ones who went quiet, and stops the moment they reply.
- 5 Reporting. Tells you every week what came in, what converted and what it could not handle.
Every chatbot answers. This one closes. A chatbot's job ends when it has replied. This one is not finished until there is a name in your diary, or a note telling you why there is not.
It holds one remit at a time: one role, or one department. Not everything at once. A thing that claims to do everything does none of it reliably, and you would be right to distrust anyone who tells you otherwise.
Five things we will not pretend
Every one of these has cost somebody money elsewhere.
It does not touch your phones.
Nothing we build answers a call. We work where people type: your website, WhatsApp, email, forms and documents. If the telephone is genuinely your main problem, we are not your firm, and we will say so on the first call.
It does not write whatever it likes.
You approve what it is allowed to say: the prices, the policies, the material it answers from, and the questions it must hand to a person. You sign that off before it meets anybody. You can then read every conversation it has. What you are approving is the boundary, not each individual sentence, and any supplier who tells you they can pre-approve every sentence does not understand what they are selling.
It does not fix a broken process.
Automating a process that does not work gets you a broken outcome faster and more often. If the audit finds that, we will say so, before you spend anything on a build, even though it is the more expensive answer for us.
It does not take on accountability.
It hands over to a person the moment it is unsure, and it tells you it has done so. The responsibility for your customer stays exactly where it is now.
It does not learn your business by itself.
Somebody has to write down how you actually work, including the parts that live in your head. We do most of that with you in the audit. It is the least glamorous half of the job and it is the half that decides whether the result is any good.
The questions people are too polite to ask
Straight answers.
Will it say something daft to a customer?
It can, which is why it is built not to. It answers only from material you have given it, inside limits you set, and hands over to a person when it is unsure. You approve those limits before it goes live, and you can read every conversation after. Ask any supplier what happens when it does not know the answer. If they say it always knows, end the meeting.
Is my data safe?
UK GDPR applies to you and to us. We tell you in writing which providers process your data and where it sits, and we configure it so your conversations are not used to train public models. Ask every supplier for the same in writing. A vague answer here is the answer.
Will it replace my staff?
Honestly: in a business your size, no. It replaces tasks, not people. In a five person firm nobody is sitting idle, so what changes is which work gets done at all, and how late in the evening it stops. If your plan is to remove a role, tell us at the start so we scope it honestly. We will not pretend that happens by itself.
Do I need to be technical?
No. If you can explain your job to a new starter, you can brief this. Nobody on your team needs to learn a new system. That is the point of it.
Can customers tell it is AI?
Often, yes. So we tell them. It introduces itself as an assistant and offers a person at any point. Pretending otherwise is a bad trade: you lose far more when somebody catches you than you ever gain by hiding it.
What if it breaks?
If it stops working or starts failing, an alert reaches us automatically, not at the end of the month. Enquiries fall back to landing in your inbox the way they do now, so nothing disappears while it is down. The monthly management is for improving it, not for noticing it is broken.
Is this actually worth it for my business?
Possibly not, and that is a real answer rather than modesty. If your enquiries are few, steady, and you already answer all of them within the hour, you do not need us. Do the sum in section 02 and let the number decide.
We tried something like this before and it was useless.
Then it was almost certainly a script with buttons, dropped onto a website and never touched again. Two things separate that from this: it is connected to your real systems, and somebody manages it every month. Ask whoever you speak to next which of those two they are actually providing.
Published, before you call
The number you approve is the number you pay.
| What | Price |
|---|---|
| Foundation. Website and WhatsApp AI Superagent. Answers, qualifies and books, day and night. Live in about two weeks | Quoted at the audit. Both channels. Managed. No contract, leave any month |
| The Vybzz AI Audit. A review of how you work, written recommendations, and a commercial case for each one. Report in five working days. You own it either way | Quoted before you commit. Half the fee comes off your first build |
| Founder, Growth and Workforce. Superagents, one department or several, built around your systems | Scoped and quoted after the audit |
Foundation does not need an audit. If you already know that is what you want, we build it. The audit is for anyone who is not sure what to fix first, or who wants something built to their business. You will never be charged for both without agreeing to both.
The audit price moves with how much there is to go through, so you get the exact figure in writing before you pay anything.
The AI audit also carries a guarantee: if it does not identify at least ten hours a week of recoverable value, you do not pay for it.
We do not quote scoped work before the audit, because a number given before anyone has looked at your business is a guess, and you would be right to treat it as one.
Four stages, no surprises
From first conversation to something working.
- 1 A conversation. 30 minutes, no charge. You describe what is annoying you. We tell you whether AI is the answer, and occasionally that it is not. Nothing is prepared, nothing is presented.
- 2 The audit. Five working days. Skip this if you are going straight to Foundation. Otherwise we go through how the work actually flows, where enquiries come in, what happens after five o'clock, what gets rekeyed, what gets forgotten. You get a written report with a commercial case per recommendation, and you own it.
- 3 The build. Foundation goes live in about two weeks. Everything else is scoped and timed at the audit, so you know before you commit. You approve what it is allowed to say before anybody outside your business sees it.
- 4 Running it. Managed monthly. We watch what it handles, what it escalates and what it gets wrong, and refine it. You get a report you did not have to build.
Three reasons we would turn you down
Worth reading before you book anything.
- You want something cheap and off the shelf. Those tools exist, they cost a fraction of this, and for some businesses they are the right answer. We are not competing with them and we will not pretend to.
- You would rather not explain how your business works. We cannot build this without that, and any firm claiming otherwise is going to hand you something generic.
- You expect it to be finished on day one. In week one it handles the common enquiries and passes you the rest. The hand-overs are the point: each one shows us something it does not yet know, and we close those gaps every month. Expect the number of things it passes to you to fall steadily. Anyone promising near zero on day one has not done this before.
Four questions to ask any AI supplier
Including us. Especially us.
- 1What will this actually do in week one? Say it in one sentence with no acronyms in it. If the sentence needs jargon, there is nothing behind it.
- 2Where does it get its answers from, and what happens when it does not know? The second half of that question is the one that separates the professionals from the rest.
- 3What does this cost to run in month twelve, not month one? Build fees are easy to compare. Running costs are where the surprises live.
- 4If we part company, who owns the build, the data and the accounts? Ask before you sign, not after.
Print those four. Use them on us.
Find out what "nobody" is costing you
Right now, everything that arrives after five o'clock is handled by nobody, which is the gap AI lead response and follow-up closes. Nobody works for free, never calls in sick and never asks for a rise. Nobody is also the most expensive member of staff you have. Thirty minutes, no charge, no slides. Bring the number you worked out in section two.
Or email hello@vybzz.co.uk. Vybzz is the trading name of MZ Corp Limited, registered in England and Wales, company number 17351738.