Exzel XExzel
AI Solutions

The tools we build to sell and measure better.

We are a sales and marketing consultancy that builds its own technology. This is what already works: watch the story and talk to the assistant.

The story uses the dental CRM we built, with sample data. The tools adapt to any business.

01

A CRM built around you

Each business gets its own installation and its own database, with its name, logo and color. Modules are switched on for what the business needs, and each person on the team sees only what concerns them.

What it does

  • Where each customer came from, what they were told and whether they booked, in a single record.
  • Scheduling with confirmations and reminders, plus the daily list of who to call back.
  • Roles: each person sees their own part; the front desk, for example, doesn’t see the finances.
  • Already built for dental clinics, optical stores, veterinary clinics, hospitals and medical practices, with what each sector needs: dental charts, prescriptions, vaccines, signed medical records.
  • More than 8,700 automated tests across its five versions.
01Dental clinics
02Optical stores
03Veterinary clinics
04Clinics and hospitals
05Medical practices
Another sector? Let’s talk.
02Live

An assistant that books

It works like a good front desk, at any hour: it understands the reason for the visit, offers open slots within business hours, asks for a name and phone number, and creates the record, the appointment and the reminder.

What it does

  • The rules are set by code, not by an instruction to the model: it gives no prices or diagnoses, even if the model tries.
  • If it spots an emergency, it doesn’t book: it sends the person to the ER and alerts the team.
  • It answers with our own models and, if they’re unavailable, with a backup model. Who wrote each reply is always logged.
  • This is the real assistant: whatever you write, it answers. In the demo no appointment is saved.

Illustrative example

C

Sample dental clinic

online · assistant

Hi, do you have a whitening appointment this week?

Of course! I have Thursday at 10:30 a.m. or Friday at 3 p.m. Which works for you?

Thursday, perfect.

Great. Could you give me your name and phone number to book it for you?

What happens in the business

at the same moment

  1. Reason understood · Whitening
  2. Calendar checked · 2 open slots
  3. Appointment created
  4. Reminder scheduled · 24 h before
03

Your own assistant

Ours runs Exzel’s day-to-day from a chat: we ask in plain words and it finds prospects, prepares messages, watches who replied and sends us a briefing every morning. On that same foundation we build yours, for the work that eats your time.

What ours already does

  • You talk to it in a chat, in plain language: “who should I call today?”, “draft the email for this client”.
  • Every morning, on schedule, it sends the briefing with what matters today.
  • It watches the inbox and alerts you as soon as a client replies.
  • It prepares messages and waits for your OK: it never sends anything on its own.
  • It runs on our own AI model, on our own machines.

What you can ask for · built to measure

  • Sort your email and send you a summary every morning.
  • Remind you of appointments and meetings.
  • Draft a reply to every email, so you only review it and hit send.

Illustrative example

Your assistant

online

“Every morning, summarize my email and remind me of my meetings.”

  1. Email summaryDaily · 7:00 a.m.
  2. Draft replyWhen an email arrives
  3. Meeting reminder15 min before
  4. Appointment reminder24 h before

Nothing goes out without your OK.

04Live

Live consultation transcription

With the patient’s consent, the system listens to the consultation and transcribes it live. At the end it drafts a structured note that the doctor reviews and corrects: nothing goes into the medical record without their approval, and nothing signs itself.

What it does

  • Asks for consent before starting and records it.
  • Can transcribe on our own machine, without sending the audio to outside services.
  • Audio is stored encrypted. If the patient withdraws consent, the transcript, the note and the audio are deleted.
  • Only staff allowed to document the consultation can use it.

Illustrative example

Live

DoctorHow long have you had the discomfort?

PatientAbout two weeks, mostly at night.

DoctorHave you taken anything for the pain?

PatientJust acetaminophen, and it helps a little.

Suggested note: the doctor reviews it, corrects it and decides whether it goes into the record.

05

End-to-end metrics

Meta tells you how many conversations were started. What happens next —whether someone replied, booked, showed up, paid— almost nobody measures. We do, channel by channel.

What it does

  • Every request tagged with its source channel, through to the appointment and attendance.
  • Anything without a source is shown as “unidentified”, never as “organic”. A zero is shown as a zero.
  • On your website: visits, clicks to call and to WhatsApp, forms and bookings.
  • Profitability by channel, and questions in your own words: “which campaign brings customers?”.

Illustrative example

Conversation
Reply
Booking
Showed up
Paid

What Meta measures

What we measure

06

Content factory

Ten agents, each with its own job: spotting topics, planning, writing, reviewing, adapting to each network, designing and analyzing. They write backed by our own library so every claim has support, and a filter checks the sector’s regulations before anything goes out.

What it does

  • A library of 73 works and more than 63,000 passages, searched by meaning.
  • Narration with the person’s cloned voice, with their consent, and generated video.
  • A regulatory lock: in healthcare, for example, it checks ads against Colombian law before publishing.
  • Nothing is published without a person’s OK.
  1. 01Spots the topic
  2. 02Plans
  3. 03Writes with sources
  4. 04Checks the rules
  5. 05Designs and edits
  6. 06A person approves
07

Smart prospecting

A crawler that cross-checks public sources to find opportunities in a niche and a city: who advertises, what their website measures, how they take bookings, what their reviews say. Every data point carries its source, and what couldn’t be verified is flagged.

What it does

  • Google Maps, Meta’s Ad Library (only the business’s own page), Instagram, official registries and each business’s website.
  • On the website it detects tracking pixels, the booking system and whether their WhatsApp link is tracked.
  • It scores with formulas, not with a model’s opinion.
  • Missing data isn’t negative data: a website that’s down means “unknown”, not “doesn’t measure”.
  1. 01Public sources
  2. 02Signals
  3. 03Scoring
  4. 04Sample website
  5. 05Approved contact
08

Donor discovery

We built it for healthcare foundations seeking international cooperation: it finds organizations in other countries with a giving history from their public filings, estimates how likely they are to support the cause and explains that estimate signal by signal. Then it takes them through a pipeline, like any sale.

What it does

  • Starts from public filings: in the United States, the Form 990 that nonprofits file.
  • Map by region and entity discovery.
  • Propensity explained signal by signal, not a black box.
  • Donor pipeline, from first contact to donation.
  1. 01Public filings
  2. 02Organization profile
  3. 03Explained propensity
  4. 04Donor pipeline
09

Websites that build themselves

It takes what is already known about a business —its Google profile, reviews, services, photos— and generates a branded website, ready for review. It only uses verified data, each item with its source.

What it does

  • 30 different designs, with three variants each.
  • Includes the assistant and, if wanted, a story like the one on this page.
  • Nothing goes online without human approval.
  1. 01Google profile
  2. 02Verified data, with sources
  3. 03Design and branding
  4. 04Human review
  5. 05Published

In-house AI

Our own models, on our own machines.

We work with open models running on our own GPUs. That gives us the choice: when the data is sensitive, it stays in-house and never goes to an outside service.

Language

Reasoning, writing and extracting data from documents and web pages.

Vision

Sorting photos: discarding storefronts, detecting people and organizing an image bank.

Voice

Live transcription and narration with a cloned voice, always with consent.

Image and video

Original images and videos, no stock photos.

Semantic search

Finding the passage or image that says the same thing, even in different words.

How we work with AI

Four rules we don’t bend.

No made-up data

If something couldn’t be verified, we say so. Copy gets written; facts don’t.

The model proposes, the code decides

The rules that matter —no prices, no diagnoses, nothing published without permission— are enforced by code, not by an instruction the model could ignore.

A person approves

Nothing reaches a customer without sign-off from someone on the team.

Tested before delivery

Thousands of automated tests that must pass before anything is called done.

One of these tools in your business?

Tell us what you sell and how customers reach you. We’ll tell you what makes sense to build and what doesn’t.

Book a free call