Process
Measure first: how a feasibility check works
Scoping an automation project before anyone has looked at the documents it must handle leaves real uncertainty about feasibility and scope. Our approach is to look and measure first, then propose. This article describes what we intend to do when you ask for a free feasibility check, what you get back, and when the honest answer is "don't build this".
What you send
One of two things:
- Up to 3 sample documents of the kind you want processed: invoices, receipts, quotes, forms. Remove or black out personal data we don't need, such as names of private individuals, bank details or health data.
- Or a short description of one process you repeat: what comes in, what you do with it, where it ends up, how often.
The contact form asks four questions: what you need, how often and how much, which tools you use now, and when you need it. The form stores nothing on the website; it prepares an email in your own email app.
What we do with it
1. Read the material. Text PDFs or scans, how many layouts, which fields, how clean the numbers are.
2. Run a quick test where we can. For documents, we run them through our existing demo pipeline (doc2data) or a small adaptation, and count, field by field, what came out right, what was flagged, and what was wrong without being flagged. Three samples are a small check: the counts describe those documents, and they are not a reliable estimate of general accuracy.
3. Identify the risks. For example: several layouts per supplier, multi-page tables, phone photos (untested in our demo), fields only a person can interpret, data you may not be allowed to share.
4. Propose how the real project would be measured. A test set made from more of your documents, kept apart from those used for building, and a pass mark agreed in advance. (The reasoning is in A practical guide to a reproducible feasibility study.)
What you get back
The intended deliverable is a short written reply, around one or two pages:
- what looks feasible, and what doesn't;
- what we counted on your samples, with the small-sample caveat;
- the main risks and how we would handle them;
- how success would be measured in the project;
- a written proposal (scope, deliverables and delivery date), or a clear recommendation not to build.
The check creates no obligation on either side.
How long it takes, and why we do it
We expect a check to take a few hours of our time (an estimate, not a track record). We do it because it is also how we avoid scoping blind.
When we would advise against building
- The documents are too varied or too poor for the accuracy you need.
- A simpler fix exists, such as asking a supplier for an export instead of a PDF.
- The build would not save enough work to be worth it.
- The work needs something we don't offer: 24/7 support, hosting, regulatory sign-off, or AI running on our own subscriptions.
- The task is ghost-writing academic work.
A reasoned "no" is more useful than a "yes" that fails later.
What happens to your samples
Our policy is that sample documents are used only for the feasibility check and deleted once it is done, are not sent to any AI service, and are not used to train AI. Samples arrive by email, so copies also exist in the email account until deleted. The Privacy page explains who our email provider is and your rights.
If you go ahead
A project would follow the same logic: a written agreement based on the proposal, building in small steps, measuring on your data against the pass mark agreed at the start, and a handover with source code you own, a plain-language guide and a short call. The five steps are on our How we work page.
Who does the work
Snello is one founder, Foysal, based in Italy, working with AI tools for drafting, coding and checking. Foysal reviews and is responsible for everything sent to you, including the feasibility reply.
Ask for a check
Send up to 3 samples or describe one process. Free feasibility check
Sources
- Snello company plan, approved process and contact design, 2 October 2026 (internal); see also How we work, Services, Contact and Privacy.
- doc2data demo evaluation reports, 2 October 2026 (synthetic; internal). Published results with denominators and limits: doc2data test evidence.
- company-researcher review, 3 October 2026 (internal).
Bring your workflow into focus.
Send up to 3 samples or describe one process. We reply by email with what is feasible and how we would measure it.