Minimal access
A scenario that reads has no right to write.
Projexo Digital
Less data entry, without losing control
Spotting what is repeated, copied over and lost between two people — then handing it to a machine, while keeping judgement on the human side.
1process mapped before choosing any tool at all
0scenarios without a fallback or human takeover
APIofficial Meta API: the only route that does not put your number at risk
Diagnosis
Six time drains that do not show up in the accounts
They cost nothing identifiable. That is what makes them invisible.
The same questions
Opening hours, address, price, lead time. Two minutes each: half a day a week.
What answers itA bot that answers common questions
See how
Copy and paste
A name typed into the mailbox, retyped into a spreadsheet, then into invoicing. A chance to get it wrong every time.
What answers itA contact created once, which feeds the following tools
See how
Forgotten follow-ups
A quote never followed up. Nobody keeps an up-to-date list in their head.
What answers itA follow-up that goes out at the agreed time
See how
Quotes and invoices re-entered
The same information typed three times. The most profitable process to automate.
What answers itA scenario that reuses the information already entered
See how
Scattered data
One customer in four tools, with four versions of their number.
What answers itA single source, linked to your tools
See how
Out-of-hours messages
A request at 9 pm waits until the next day. Often, the fastest one wins the business.
What answers itAn immediate acknowledgement, with the stated response time
See how
Use cases
What can be automated, and what stays with you
Not all these automations are equal and they are not done at the same time.
| Process | What the machine does | What remains human |
|---|---|---|
| Acquisition | Records, assigns, acknowledges receipt | Qualification and the commercial response |
| Appointments | Proposes, confirms, reminds, handles cancellation | Special cases and emergencies |
| Qualification | Three closed questions, sorting, routing | What deserves a visit or a quote |
| Quote | Pre-fills, sends, tracks opening | The price, the terms, any judgement call |
| Payment | Issues the link, records, updates | Goodwill gestures, disputes, refunds |
| Follow-up | Written reminder, and stops as soon as the customer replies | The phone follow-up, and knowing when to stop |
| After-sales | Identifies the file, answers common cases | Any complaint |
| Reporting | Gathers the figures at a fixed date | Reading them and making decisions |
Three different things often called "AI"
- Deterministic automation — "when this happens, do that". Predictable result, verifiable line by line. What suits a quote or an invoice.
- The rule-based chatbot — a tree written in advance. It never steps outside its frame: a quality when the answers commit the company.
- The assistant using a language model — it understands unplanned wording and rephrases from a validated base. More flexible, but probabilistic: it is framed, tested and monitored, and never decides on its own.
Customer-relations WhatsApp bot
Answering straight away, without answering just anything
Common questions, appointments, follow-ups, tracking. It is not connected to any POS.
The bot, facet 1
What it handles
FAQ validated by you, appointment booking and reminders, qualification with closed questions, sending documents, quote follow-up, order tracking.
Hello, what are your opening hours on Saturdays?
From 9 am to 6 pm. Would you like to book an appointment?
Yes, Saturday morning if possible.
Two slots are free. You will receive confirmation by message.
The bot, facet 2
Handover to a human
At the customer's request, after two misunderstandings, or as soon as a sensitive subject comes up. The bot announces that it is handing over and passes on the entire conversation.
That is not what I was asking.
I did not understand your request. I am handing over to an advisor.
They will receive our whole conversation and reply to you as soon as we open.
The bot, facet 3
Hours and consent
It introduces itself as an automated service, says when a human will take over, and only sends messages on the company's initiative with consent that has been collected and kept.
Hello, you are talking to an automated service.
An advisor will pick up the conversation tomorrow from 9 am.
All right, thank you.
The bot, facet 4
What it does not do
No access to the POS, no order taking, no medical, legal or financial advice, no price negotiation, no handling of complaints.
Can you do something on the price?
I do not deal with prices. I am passing your request on to an advisor.
You will be called back; they will decide with you.
A bot lives: your offers change, platforms change their rules. Without updates, it starts answering off the mark. Maintenance is not a comfort option.
Not to be confused
Two WhatsApp bots, two different offers
One talks. The other sells. Taking one never obliges you to take the other.
Workflows and integrations
Anatomy of a scenario
It is the fourth element that separates a reliable automation from a makeshift job.
What can be connected — and under what conditions
Contact managers, shared spreadsheets, payment solutions, e-mailing tools: Stripe and Brevo are among them, as are Google spreadsheets. But an integration is not promised on a logo, it is verified:
- The tool exposes a usable connection interface.
- Your subscription plan gives access to it — this is often where things get stuck.
- The necessary data is genuinely accessible and editable.
When no direct connection exists, there is often an intermediate route: scheduled export, shared file, assisted entry. Less elegant, but it works.
The method, in eight steps
Map what really happens · Prioritise by frequency, time spent and risk of error · Prototype on a real case, outside production · Test the awkward cases above all · Deploy gradually, with rollback · Document for your team · Monitor the logs and the failure rate · Improve from the unforeseen cases.
Automating a poorly defined process only speeds up the disorder. We keep one or two processes, not eight.
Security and data
What concerns your customers' data
Accounts in your name
Access keys stored encrypted, Projexo a mere guest.
Role of data processor
You remain the controller; Projexo acts on instruction, with no other use.
Traceable consent
Collected and kept. Refusal is respected immediately.
A reachable human
No scenario locks a customer into an automated journey.
Logging
What was executed, when, with what result.
Before / after
A quote request, before and after
Sequences of tasks, not quantified gains: those would depend on your volume.
Before
Before
- The message lands in a shared inbox; nobody knows who is dealing with it
- Someone calls back to get what the customer has already written
- Contact details are entered two or three times
- The quote goes out; nobody knows whether it has been opened
- The follow-up depends on someone's memory
After
After
- The form asks the questions that decide feasibility from the outset
- The contact is created once and feeds the following tools
- An acknowledgement goes out straight away, with the stated response time
- The request is assigned, with a deadline
- The follow-up goes out at the agreed time and stops as soon as the customer replies
The examples on this page use no real customer data: no name, no number, no conversation content. The price remains decided by you.
The offers
How much does an automation cost?
Prices live in a single place on the site. The customer-relations bot has a displayed price; workflows are priced after mapping.
Before deciding
Ten questions that decide
Yes. The bot is built on the WhatsApp Business Platform, Meta's official interface. It is the only route that does not put your number at risk: unofficial tools that drive the consumer app expose the account to permanent suspension. In return, the API imposes rules — business verification, message templates subject to approval, a limited reply window after the customer's last message.
It is possible, but it is not trivial: a number attached to the API can no longer be used in the standard WhatsApp app on a phone. If that number is the one your team uses every day, it is better to dedicate another one to the bot. Transferring a number already used on WhatsApp means releasing it first, and the conversation history is not carried over.
Yes, at two levels: the verification of your business, which requires official documents, and the approval of each message template you want to be able to send on the company's initiative. These two steps depend on Meta, not on us: no timescale is promised here. It is not a simple instant hook-up, and that is why a WhatsApp project always begins with this formality.
Meta bills the use of its platform according to its own price list, separate from our fees: it varies by recipient country and by conversation category, and Meta updates it periodically. This cost is re-invoiced or paid directly depending on the configuration chosen, and it is priced in the quote from the list in force on that day. No amount is displayed here, because it does not depend on us.
Only if the tool exposes a usable connection interface and your licence allows it — two conditions checked tool by tool, before any promise. When no direct connection exists, there is often an intermediate solution: scheduled export, shared file, or assisted entry. We promise no compatibility before having checked it on your installation.
An automation can fail: a third-party service unavailable, an expired access key, a data format changed without notice. That is why every scenario is designed with a fallback: what fails is logged, an alert is sent, and the item concerned is placed in a queue to be handled manually rather than silently lost. No system runs without interruption, and we do not claim otherwise.
The handover to a human is planned from the design stage, not added afterwards. It is triggered at the customer's explicit request, after two consecutive misunderstandings, or as soon as a sensitive subject is detected — complaint, dispute, emergency. The bot then clearly announces that it is handing over, states the response hours, and the complete conversation is passed to the person taking over.
No. A scenario does not change by itself: what it knows how to answer comes from a knowledge base that you validate and that we update. When a language model is used to rephrase or understand a question, it works from this base on every exchange; it does not train on your conversations and does not pick up new knowledge as it goes. Any change to what the bot says is a decided, tested and traced modification.
Monitoring operation, fixing scenarios that fail, updating connections when a third-party tool changes its rules, and the content adjustments you request. Not covered: creating new scenarios, connecting a new tool, and the costs billed by third-party platforms. Without maintenance, an automation degrades — not all at once, but surely, at the pace of the updates to the services it uses.
The scenarios are built in accounts opened in your name where the platform allows it, and documented so that another provider can take them over. The data remains yours: Projexo acts as a data processor, on instruction, and does not use it for any other purpose. The precise terms of reversibility and return remain to be confirmed in writing.
Going further
What goes with it
Each area of expertise produces material that the others reuse. That is what makes a combined offer cheaper.
Let's map a single process
The one that costs you the most time. We write it down together, step by step. You leave with the diagram of what exists and an opinion on what deserves to be automated.