AVA · Numbers before the conversation

Cost the process before you buy the automation

AVA works out what a manual process costs you today and what budget for automating it would pay back. It computes in your browser, on your figures, with no form and no price of ours. If it will not pay off, it says so.

For anyone deciding what to tackle first, or holding a quote and wanting to check it against arithmetic.

The calculation

Not sure where to start?

Describe what eats the time and we point you at a process we have economics for. Nothing invented: these are the same six processes.

Your numbers

Which process are we costing

An engineer opens a drawing or project file and hand-builds a bill of materials, a take-off or a quote from it

operations
min
min

We ask for two figures, not one: the spread is the honest answer. The range in the result comes from here.

₽ per month, before tax

120 000 × 1.302 payroll contributions ÷ 164 working hours = 953 ₽ per hour of employer cost

people

Assumptions

These are the places where we are not deciding for you. Change any of them and the result recalculates.

What a mistake costs
What state the data is in
Where data is allowed to go
How cautiously to count

Moves three things at once: the share automation removes, and the lower and upper bound of the running cost.

Automation removes: 77 % of the operationA person still checks: 15 % of what was automated

What comes out

The process costs today

80 h — 240 h per month

910 000 ₽ — 2 750 000 ₽ per year

Freed up

52 h — 155 h per month

590 000 ₽ — 1 800 000 ₽ per year

Running the system
15 000 ₽ per month
Capacity freedThis much human capacity returns to the team. These are hours, not money: what they become is your call.
0,31 FTE
Net savingafter running cost, at full effect
34 000 ₽ — 130 000 ₽ per month

Worth costing further

On your numbers the process pays for its automation even on a normal operation. What is left is to verify the share against real data — here it is an assumption.

Budget that pays back

  • in 6 months140 000 ₽ — 610 000 ₽
  • in 12 months290 000 ₽ — 1 200 000 ₽
  • in 24 months700 000 ₽ — 2 800 000 ₽

This is not our price. It is your ceiling: a project costing more will not pay back inside that horizon. Take the figure to any contractor — including not us.

₽ project budget

Looks like a solution

Document AI

Turning files and documents into structured data: extraction, classification, completeness checks — with a human confirmation before anything is written.

Go deeper

Optional. It appears only when you enter your own figures — we do not invent them for you.

Where the freed hours go

By default freed hours count as cost saved — hours are not yet revenue. If the freed time goes into sales, enter the margin per hour.

% of operations

We do not measure your error rate for you: automation checks every operation against formal rules, a person still re-checks part of it. Enter your current error rate and the cost of unpicking one and we compute it as your assumption.

How this is computed

(operations × minutes ÷ 60) × 77 % × (100 − 15) % × 953 ₽ − 15 240

Employer cost of an hour: salary × 1.302 ÷ 164. Running cost: 12 ₽ per operation for model calls plus a fixed monthly amount for supervision and infrastructure — it scales with volume, never with your payroll.

Year one is counted with a ramp to full effect: 40% · 75% · 100% · 100% by quarter. Running cost is paid from month one regardless.

What automation actually takes over

  1. Parsing the project file and extracting its elements
  2. Classifying elements by size group and type
  3. Counting quantities, areas and lengths
  4. Assembling the spec in the working format

What the share rests on

Anchored to a delivered CAD-to-spec project: on control objects the app matched the manual spec line by line — 784 of 784 and 189 of 189 panels, area within 0.001 m². Per object: 2–15 hours by hand versus 10–30 seconds processed.

What this calculation does not know

We do not put our own price into it

We publish no verified price range, and inventing one here would be a lie inside the very figure this page exists to keep honest. So it computes a budget ceiling instead: what a project may cost and still pay back within your horizon. Take that number to any contractor — it is arguably more useful against someone else's quote than against ours.

The automatable share is an assumption, not a measurement

It is the one place where our estimate enters the arithmetic instead of your data. The band under each process traces to a specific delivered project, and where that project did not measure the effect, we say so. The control is left open deliberately: put your own figure in if you have one.

Freed hours are not yet saved money

An hour that stops going into manual work becomes money only if you leave a vacancy unopened, cut a contract, or move that person onto work that earns. If the freed time simply dissipates, the saving stays on paper. The page shows both hours and rubles — what they become is your decision.

One remembered figure is worse than two

So we ask for a normal operation and a heavy one, and answer with a range. When the spread exceeds fourfold, the page says the estimate is no longer meaningful from memory and offers a measurement — which is more honest than printing the average.

What next

Let us measure it on your data

The figure above rests on one assumption — the share automation removes. Send us 30 anonymised operations from a month and we will replace the assumption with a measurement: the real share and the real time, returned as a file with the arithmetic and the method. Free. If it does not pay off, we will say so plainly — that happens regularly.

Which process are we costing: Specs and take-offs from project files
The process costs today: 80 h — 240 h per month
Freed up: 52 h — 155 h per month
Net saving: 34 000 ₽ — 130 000 ₽ per month
Budget that pays back in 12 months: 290 000 ₽ — 1 200 000 ₽
What comes out: Worth costing further
Capacity freed: 0,3–1 FTE

or email us directly: hello@xteam.pro

Questions

Why do you not show what a project costs?

Because we do not have an honest figure for this page. The same sentence describes wildly different amounts of work: "automate request handling" is both a two-week scenario on clean data and a six-month project across five systems. We give a range after a short conversation and confirm it after a diagnostic. The budget ceiling, meanwhile, you can compute here without us.

Where do the figures I type go?

Your figures go nowhere. The whole calculation runs in your browser, and no entered value is sent to the server until you press Send in the form at the bottom. We do not write them into the address bar either, which is why the URL never changes as you move the controls. Only anonymised interface events — which process is selected and which verdict came out — go to analytics, and only after you consent to analytics. Salary, volume, minutes and budget never reach the counters.

Why does the running cost not depend on salaries?

Because it is our cost, not yours: model calls scale with the number of operations, while quality supervision and infrastructure are a fixed monthly amount. An earlier version of this model defined running cost as a share of the saving, so raising the salary control tripled the hosting bill. That was wrong and we rewrote it.

What does "on the edge" mean?

That on a normal operation the saving does not cover the running cost, and on a heavy one it does. The decision then depends on which kind you have more of — and there is no answer to that from memory. Such a process is worth measuring first and discussing second.

What do you do with 30 operations if I send them?

We measure on them what was an assumption here: how long an operation really takes and what share can actually be removed. You get a file with the arithmetic and the method, so it can be checked without us. It is free and commits you to nothing. If it does not pay off, we write that — it happens regularly.

Is this the same AVA that lived on its own domain?

The same problem, rewritten from scratch. The old version demanded a name, email, phone, company and job title first, then opened a chat with a model that produced prose without a single figure. It is now the other way round: numbers immediately and without a form, conversation afterwards and only if you want one.