AI automation systems · for businesses that already have customers
A new enquiry sits for hours before anyone answers. A quote goes out and nobody follows up. Onboarding a client means the same copy-pastes every time. Monday starts with pulling numbers into a report.
I design and build the AI systems that take those jobs, with a person approving every step where judgement matters, and I measure what they release in hours and in cash, never blended into one flattering number.
SYSTEM Speed-to-lead
Where it leaks
When I map a business, the leaks tend to sit in the same four places. None of them is glamorous. All of them happen every week.
Someone asks, nobody answers for hours, and by then they’ve booked with whoever replied first.
Zone 1Speed-to-lead
Quotes, no-shows, past customers. Everyone means to follow up. It depends on someone remembering.
Payment comes in, then the welcome email, the contract, the intake form, the kickoff booking and the shared folder, each one typed out again.
Zone 3Client onboarding
Numbers live in several tools, someone copies them into a sheet, and by the time it’s done nobody has time to act on it.
Zone 4Reporting
If none of these sound like you, you probably don’t need automation yet. See “not a fit” below: I’ll tell you what I’d do instead.
What I build
I automate closest to the money first: the reply, the follow-up, the booking. Admin comes after. Every system below is a pipeline with a trigger, an AI step where it actually helps, a human checkpoint, a stop rule and a log. The AI writes and classifies. Your rules decide.
Every new enquiry answered in your voice, qualified and offered a booking.
A website form or an email becomes a logged lead, once, even if it arrives twice. AI reads it against your business context, sorts fit and urgency, drafts the reply in your voice and offers your booking link. Anything unclear, price-sensitive or upset goes to a person first. Instagram DMs and text messages can feed the same pipeline, within each platform’s rules.
Checkpoint Replies are drafted for your approval until they pass the test set we agree on.
The follow-ups you meant to send, sent on schedule, and stopped the moment someone replies.
Quotes that went quiet, booked calls that didn’t show, customers who haven’t been back and agreed to hear from you. Each gets a short, written sequence with a clear end. Any reply pauses the sequence and lands with you.
Checkpoint Nothing offers a discount, a refund or a new date without a person.
Payment in, and the welcome, contract, intake and kickoff happen without anyone copy-pasting.
The moment a client pays, the system sends the welcome, issues the contract, collects the intake form, books the kickoff and creates the client’s folder from your template. It chases missing items once, then tells you.
Checkpoint Anything missing or unusual goes to the account owner, not into a loop.
Your numbers pulled from every tool into one page, with a short brief you read before anyone else.
Leads, bookings, sales, delivery and cash, pulled on a schedule into one sheet or dashboard. AI writes a short summary with links back to the source numbers, so every sentence can be checked.
Checkpoint The brief comes to you first. It goes to your team only after you’ve read it.
An assistant that knows your business, answers from your documents and knows when to hand over.
Support, sales prep or internal knowledge. It’s grounded in a context pack about your business, tested against real questions before it goes live, and given written rules for what it must escalate. I onboard an agent the way I used to onboard a developer: a clear job, a test, then trust in steps.
Checkpoint Escalation rules are written before the build, and every answer is logged.
Agents, without the hype
Most “AI agents” fail for the same reason a new hire fails: nobody told them how the business works. So every agent I build starts with your company’s context, written down. Then it gets one clear job, a set of real test cases it has to pass, and rules for when it must hand over to a person. It goes live in steps, starting with work a person reviews.
The artefact
Every system starts as a one-page recipe you can read in two minutes and argue with. It names the trigger, each step, where AI is used, where a person decides, what stops it and what gets logged. If the recipe is wrong, it’s cheap to fix. If the build is wrong, it isn’t.
The free blueprint is this exact system, built out: every node in n8n and Make, the two prompts, the test leads, the follow-up sweep and the heartbeat.
Human checkpoints
The fastest way to lose a customer’s trust is an automation that confidently says the wrong thing. So before I build, we write down where a person has to stay in the loop. It’s the same pattern I built into Misclo, my own app in development: the system proposes a change, the person accepts it or undoes it, and nothing happens silently.
Proposed: Misclo phone screen: a late meeting leads the AI companion to propose a shorter home workout and a quicker dinner, waiting for the user’s OK. Accepted: The same Misclo screen after the change is accepted, with an Undo link.
Who builds it
I’m Aymane Hitmi. I was CTO of a US marketing agency, where I led a team of 12 developers, and as CTO my team delivered and maintained 28+ client platforms. Over my career I’ve managed 45+ digital projects, and I’ve built and shipped two software products of my own. Now I build with AI as my team, in public, and I hold it to the same standard I held people to.
How we work
Same discipline I used as CTO of an agency team, now with AI doing the build and me directing, testing and approving. You see each step before it replaces the last.
We look at how work actually moves through your business and rank the candidates by value, feasibility and risk. We take the baseline numbers now, before anything changes.
I write the recipe for each system: trigger, steps, checkpoints, stop rules, logs, tools, running cost and the test cases it must pass. Your words go into the spec verbatim, so nothing drifts.
I build in your accounts, against test data first. Nothing reaches a real customer until it has passed the tests we agreed.
It goes live with drafts-for-approval and stop rules switched on. We loosen checkpoints only where the log shows it’s earned.
We compare against the baseline: hours released, cash recovered, running cost. Then we keep it, change it or switch it off.
Measured honestly
Automation pitches love one big number that blends time saved, revenue “influenced” and a guess about the future. I don’t do that. I measure two things, separately, against a baseline we take before I build anything, and I put what the system costs to run next to both.
BASELINE (before build, from your own records)
minutes per task × tasks per month ÷ 60 = hours per month now
enquiries per month · median time to first reply · calls booked
your close rate (calls → sales) · your average sale
AFTER 30 DAYS
hours released = hours before − (hours after + review time)
booking rate = calls booked ÷ enquiries (before vs after)
extra bookings = (rate after − rate before) × enquiries after
cash recovered = extra bookings × your close rate × average sale,
with each sale listed
running cost = tools + AI usage + care plan
BASELINE PERIOD
Same months last year when you have them; otherwise flagged as a
short baseline. With few enquiries, a difference can be noise, and
the report says so.
REPORTED AS
hours released (in hours, never converted to a dollar "saving")
cash recovered (in your currency, with the sales listed)
running cost (in your currency)Put in your own numbers. The calculator shows what the current way costs you, how many extra sales a month a system would need to cover its running cost out of the profit on those sales, and how long the one-off build fee would take to recover at a number of extra sales you choose. It doesn’t predict an uplift, because I can’t know that before we measure.
Everything is calculated in your browser. Nothing is sent, stored or put in a link. This is arithmetic on your inputs, labelled as an estimate, not a forecast.
I don’t promise hours saved or revenue. I promise the measurement will be honest, and that you’ll see the working.
Investment
Every build is a fixed scope with acceptance tests agreed before work starts. Software and AI usage are billed to you directly by the providers, at their prices.
Automation
From $850
One system end to end, such as lead capture → AI qualification → CRM → reply draft → booking.
Automation
From $1,500
One grounded assistant for support, sales prep or internal knowledge.
Automation
From $2,300
Three connected systems, typically speed-to-lead, client onboarding and a weekly numbers brief.
If you’d rather not own the upkeep, I can look after the systems I’ve built: monitoring, fixes, model and API changes, and a monthly report on health and running cost.
Final price depends on your tools and how many systems need to connect. It’s fixed in writing before I start.
Stack stance
I’m not tied to one platform. I pick the tool that fits what you already use and what your team can live with after I leave. Software and AI usage are billed to you by the providers, at their prices. I’ll estimate them in the spec so there are no surprises.
Chosen per business, written into the spec with the reason.
You see the usage and pay the provider directly.
I connect to what you have before I suggest anything new.
Never in prompts, documents, spreadsheets or chat.
Personal data is trimmed before it reaches a model wherever the job allows.
documented so another builder could pick it up, and handed over in full.
Built on myself first
Someone comments a keyword on one of my reels, gets a DM with a link and lands on a free tool, with no email wall. That’s a speed-to-lead system, and it’s mine. I’m building it the way I’d build yours: a written recipe, test mode first, checkpoints, a log, and numbers I’ll only publish once they’re real.
Being built for the launch of this site No results to report yet, and I won’t invent any.
Free, ungated
The Speed-to-Lead blueprint is the recipe I’d use, written so you can build it in n8n or Make: the flow, the qualification prompts and the ROI sheet that keeps hours and cash in separate columns. Pair it with the Ops Map to decide which system to build first.
All three are tools from the OPERATE track of AI Business Build Week, where I build them live next to you for five days, free.
Needs an automation tool: n8n can be self-hosted, and Make has a free plan with usage caps. No keys or passwords live in the files.
The scorecard method that ranks your top three AI opportunities by value, feasibility and risk.
The context pack every agent and automation should start from.
System blueprint, qualification prompts, ROI sheet.
Questions
I don’t promise either. I build systems aimed at the moments closest to your revenue and the tasks that repeat, and I measure what actually changes: hours released and cash recovered, separately, against a baseline taken before the build. If a system doesn’t earn its running cost, I’ll say so.
No. I build it, document it and walk you through the runbook. If you want to own the upkeep yourself, I’ll show you how. If not, a monthly care plan covers it.
You do. It’s built in your accounts, with your API keys, on your data. You get the recipe, the runbook and full access. Another builder could pick it up without me.
It stays in your tools. The AI only sees what each step needs, and personal data is trimmed before it reaches a model wherever the job allows. I use API accounts, which by default aren’t used to train models, and I’ll note the provider’s retention settings and data processing terms in the spec. If your data needs to stay in the EU, I check which providers and regions can do that before I choose the tools. I’m not a lawyer: if your industry has specific rules, we build to what your adviser tells us.
Only after it has earned it. New systems draft replies for your approval, and anything involving money, promises, complaints or sensitive topics always goes to a person. Customer-facing replies say they’re AI-assisted, and in some places, including the EU, that’s a legal requirement.
Every system logs each run and alerts you when it errors or hits a stop rule, and the runbook says what to do. If you’re on a Partner plan, I fix it.
It depends on how many tools have to talk to each other and how clean your current data is. The spec names the timeline before you pay, and I won’t quote one before I’ve seen your setup.
Usually, yes. I start with what you already use. If a tool can’t be connected safely or reliably, I’ll tell you on the call and suggest the smallest change that fixes it.
Yes. Take the free Speed-to-Lead blueprint and build the first system yourself. Or book the 15-minute call and I’ll tell you the one system I’d build first, even if the answer is “build it yourself”.
You talk to the person who designs and builds the system, not an account manager. I ran an agency’s engineering team, so I know the process an agency uses. I just run it with AI, and I show you every step.
Your move
Have it built
Fifteen minutes with me. Tell me what you sell, where enquiries come in and which tools you use. You’ll leave knowing the first system I’d build in your business, or that you don’t need one yet.
The call ends in one of four answers: do it yourself with my free tools, the audit, a direct build proposal, or an honest “not a fit”.
Rather learn it live? AI Business Build Week, OPERATE trackStart on your own
The recipe, the qualification prompts and the ROI sheet. Free, no email wall. Pair it with the Ops Map to decide which system to build first, and the Business Brain for the context pack every agent should start from.
Prefer email? contact@labgenz.com