Orders from email and PDF
Orders that arrive as free text become proposals ready to confirm in the ERP.
For example30 orders a day, 4 minutes each: 2 hours of typing become about 30 minutes of checking.
Automated checks
Connections between programs
Custom management systems
The method
Technology
Artificial intelligence in your company
We look at how your office works and separate the jobs where software is enough from the ones where AI really saves hours. Then we build it on the programs you already use: AI prepares, software checks, a person confirms.
$1 to $2 per 1,000 emails read: the AI cost in our order prototype
1 job at a time: we start from the one that costs the most hours today
0 writes to the ERP without a person confirming
The first number was measured on the Inbox to Order prototype, with test emails. The other two are how we work.
Free check
Tell us about a job that costs you time today. We reply in writing: whether AI helps, whether a rule or a connection is enough, and where it pays to start.
How to bring it in
Everyone has tried an AI chat. Then orders, invoices and customer requests kept arriving as before, because an AI that is not connected to your programs cannot do your work. To bring it into the work, you go like this, one step at a time.
Today’s doubtYou do not know where to start
Agents, RAG, models: plenty of acronyms, and nobody saying which job in your company would really change.
The step that clears itFree check
You tell us about a job that costs you time. We reply in writing whether it needs AI, a rule or a connection, and where it pays to start.
Free, in writingToday’s doubtFear of mistakes
An invented order or price costs more than the time saved. Without checks it is a real risk.
The step that clears itTrial on your data
A real sample, anonymised: emails, documents or requests. We count how many results come out right before building. Fixed price, deducted from the project if you go ahead.
The result is measured firstToday’s doubtDoubts about data
Do customer emails, contracts and price lists end up in an outside service? It is a question to ask first.
The step that clears itDeciding where the data goes
An outside AI service, also with servers in Europe, or a model on your own servers: it is chosen first, and the choice is written in the proposal together with what AI may do on its own.
Written in the proposalToday’s doubtAI sits apart from the work
A text is pasted into a chat and the answer is retyped into the ERP. The manual work is still there.
The step that clears itConnected to your programs
Built on the ERP, mailbox and folders you already use. Fixed price, delivery in stages. At first a person confirms everything, and every step is logged.
A person confirms16.4%* of Italian companies with 10 or more employees used at least one artificial intelligence technology in 2025. A year earlier it was 8.2%. Industry context. Source: ISTAT, Imprese e ICT, 2025 (in Italian)
Where AI automates the work
They arrive by email, sit in documents or in your data. The same rule holds in all of them: AI prepares, software checks, a person confirms.
Pick a job
Arrives by email
Sits in documents
Sits in your data
AI preparesSoftware checksA person confirms
Orders that arrive as free text become proposals ready to confirm in the ERP.
For example30 orders a day, 4 minutes each: 2 hours of typing become about 30 minutes of checking.
Invoices, delivery notes, order confirmations and supplier price lists, each in its own format, read and turned into lines.
For exampleA new supplier with a PDF never seen before: AI extracts the lines, rules check the totals and the differences go to a person.
Manuals, procedures and price lists queried with a question, with the page the answer comes from.
For example5 people, 4 questions a day, 6 minutes each: 2 hours a day spent searching PDFs.
Order status, shipments, availability and returns answered straight away. The real cases go to a person.
For example30 requests a day, 6 minutes each: 3 hours, and many are “where is my order?”.
Emails to info@ and orders@ sorted by type, with the data already extracted and a draft reply to approve.
For example80 emails a day, 20 seconds each just to understand what they are: almost half an hour before replying starts.
From the customer request to a draft with items, list prices and availability. The salesperson corrects and sends.
For exampleA 15-line request: 20 minutes of looking up codes and prices become 5 minutes of checking the draft.
Descriptions, attributes and translations prepared from supplier files, reviewed before they go on the shop.
For example400 new items a season, 5 minutes a page: over 33 hours of work that become a review.
“How many of this item did we sell in March?” asked in plain words, with the answer read from the ERP.
For exampleThe number you ask the Excel person for today arrives in a minute, with the data it comes from.
From sales history, the suggested quantities to reorder. The buyer always decides.
For exampleWith 800 items, the “below stock in two weeks” list is ready on Monday. Often a good rule is enough, and then we tell you.
An example you can look at: Inbox to Order, our prototype that reads orders from a mailbox and writes them to a demo ERP only after confirmation. There is a 53-second video and the code is public. Read how it is built
Chat or connected AI
They are different things and both are useful. The difference is what the AI is connected to.
The AI chat on its own
AI built on your work
To write an email or summarise a document an AI chat is perfectly fine, and costs a few euros a month. Connecting it to your programs pays when the same job repeats dozens of times a day.
Before you write to us
It makes sense if
You do not need it if
What clients say
You explained exactly how you would approach the problem. That gave me the confidence to choose you.
Verified review, anonymised for privacy.
Frequently asked
Starting from one job: the one that repeats the most and that a person does today by reading and retyping. First a free check to see whether AI is needed or a rule is enough, then a trial on your data to measure the result, then the build on the programs you already use, with a person confirming.
It is a program that uses an artificial intelligence model to read text and data, choose the steps to take and prepare a result: an order proposal, a reply, a line to record. Unlike a chat, it works connected to your programs and follows written rules.
A chatbot answers questions. An AI assistant answers using your documents or your data. An AI agent also carries out steps of work, such as reading an email and preparing the order. In practice two things matter: what it is connected to and what it can do without confirmation.
Yes, if it starts from one job. The usage cost of AI is low: in our order prototype it was $1 to $2 per 1,000 emails. The real cost is building the connections and the checks, and that has a fixed price agreed upfront.
You pay model usage directly to the provider, through an account in your name whose keys you hold: no markup from us. If the model runs on your own servers there is no usage cost, but adequate hardware is needed: on a machine that is too weak the model makes mistakes, and we tell you first. The software we build has no fee.
Only if an outside service is chosen, and then the service receives the text needed for that job. If data must not leave, a model running on your own servers is used, on adequate hardware. The choice is made at the start and written in the proposal.
That is why it does not work alone. Software checks every result against your data, doubtful cases stop, and a person confirms before anything is written or sent. Reliability is measured with a trial on your data, before building.
No. AI connects to what you have: ERP, ecommerce, mailbox, folders. If a program does not allow connections, we see it in the free check.
The one that fits the job. Reading an email takes a small, cheap model, long documents need a more capable one. Our order prototype uses Claude. If data must stay in the company an open model is used, on your own servers: it works if the hardware fits the model, and we check it with a trial.
Start with the problem
We tell you in writing whether AI really helps, or whether something simpler is enough.
Prefer to write directly? [email protected]