Pilot calculator: cost and timeline of a first deployment
Five process parameters — and the order of magnitude of the one-time investment in a first pilot becomes visible, along with its timeline. With those two numbers, a conversation with a vendor starts with a subject.
Your numbers
A scenario is one complete task with a clear input and a clear result: parse an email, assemble an answer from the rulebook, fill in a record.
Both data sources and destinations for the result count: CRM, email, a document folder, a messenger.
Those who will use the result daily and give feedback.
This is the launch: building the environment, scenarios, integrations, data preparation and acceptance. Monthly operation is calculated separately.
The typical size of a first deployment for a mid-sized company. The budget allows building the environment, connecting a couple of systems and bringing answers to a quality that users accept. Payback becomes visible when the pilot covers a process that costs hundreds of thousands of roubles a month.
- Pilot timeline
- 2 months
- Base build of the environment
- 250 000 RUB
- Scenarios and answer logic
- 360 000 RUB
- Connecting systems
- 240 000 RUB
- Preparing data and knowledge
- 252 000 RUB
- Answer quality checks
- 108 000 RUB
- Training people and acceptance
- 60 000 RUB
- Cost of a pilot week
- 127 000 RUB
- Share of work spent on data
- 19,8%
The estimate gives an order of magnitude for budget planning. A precise quote comes after the specific process and the state of the data are reviewed.
How the calculation works
A pilot is assembled from six line items. One is fixed, five depend on scale.
The base build of the environment is 250,000 RUB. It covers access rights, storage, an answer log and a dashboard where every decision the system makes is visible. This part barely changes with the size of the task: the environment is needed for one scenario and for five alike.
A scenario is valued at 180,000 RUB. The benchmark comes from the volume of work: roughly two weeks of an analyst and an engineer to describe the input, the processing rules and the result format. A scenario is a complete task. Parsing an incoming email and creating a request is one. Assembling an answer from the rulebook and sending it to a customer is already another.
Connecting a system is 120,000 RUB. Both sources and destinations count: email, CRM, a document folder, a messenger. The price holds at this level for services with a documented API. A homegrown accounting system without documentation costs more, and that comes to light during the review.
What moves the total the most
The state of the knowledge. The coefficient ranges from 1 to 1.9, and across three scenarios the difference between tidy rulebooks and scattered scans exceeds 400,000 RUB. In the breakdown the share of this line item is shown as a percentage — it reveals where the work actually sits.
Requirements for checking answers. An internal assistant for employees tolerates an error and a correction. An answer that goes to a customer requires a test set, confidence thresholds and a handover-to-human scenario. A regulated area adds logging on top of that. In the calculation the surcharge runs from 15% to 50% of the scenario cost.
The number of people in the pilot has a milder effect: 4,000 RUB per person covers training, feedback collection and acceptance. This line item rarely determines the total, but skipping it is noticeable — a system without users who have accepted it stays a demonstration.
Why the timeline is calculated separately
Money and time grow differently in a pilot. The amount rises linearly with each scenario, while the timeline rises more slowly: part of the work runs in parallel, and the second integration is cheaper than the first in calendar terms.
Hence a useful number in the breakdown — the cost of a week. It shows the density of the work. If a week comes out above 200,000 RUB, the pilot has a lot of parallel tasks built into it, and that pace requires dedicated people on the client side.
Under what conditions a pilot pays off
The first condition is that the process costs noticeable money right now. One-time investment is repaid out of monthly savings, so a pilot on a process worth 50,000 RUB a month takes years to pay off. A benchmark for fast payback: the monthly cost of the manual work is comparable to a quarter of the pilot budget.
The second is that the result has a verifiable marker. The request was created correctly or it was not, the answer matches the rulebook or it does not. A pilot with a measurable output produces a number for a scaling decision; a pilot with a vague output produces impressions.
The third is that the process rules can be written down. Where decisions are made by instinct and every case is handled individually, the rules first have to be extracted and formulated. That work is useful in itself, but it has to be built into the timeline.
How to use the number in conversation
Two figures — the amount and the timeline — turn the discussion into something concrete. A budget and a horizon are named to the vendor, and from there the scopes of work are compared rather than general promises.
A one-and-a-half-fold spread between estimates is normal: the coefficients here are averaged, and the state of the data is only visible up close. A threefold spread means a different understanding of the task — and that is the best reason to check what each side counts as a scenario. A quote noticeably below the calculation usually excludes data preparation or acceptance. It is worth asking where those line items are, and on what volume of documents they were estimated.
FAQ
What amount makes sense for entering a first AI pilot
The benchmark for a single process is 700,000 to 1,500,000 RUB of one-time investment over a 2–3 month timeline. The lower bound applies when there is one scenario, one or two systems to connect, and the rulebooks are already written. The upper bound applies when knowledge is scattered across correspondence and the answers go out to customers.
Why does data preparation cost separate money
A model answers based on what it was given. If the rules live in people's heads and the documents are scans, they first have to be extracted, checked and converted into a machine-readable form. In the calculation this is 90,000 RUB per scenario with a coefficient from 1 to 1.9 depending on the state of the sources. On scattered data this line item often comes out larger than the integrations.
What makes up the timeline
Three weeks go into the base build of the environment and access rights. Then a week and a half is added per scenario, four days per system, plus a surcharge for the state of the data. The total is bounded to a 4–26 week range: a pilot is never shorter than the base build, and anything longer than six months is best split into stages.
How does this amount differ from a monthly AI bill
This is the one-time investment: analysis, build, integrations, data preparation, training people. Tokens, servers and support are paid monthly after launch and are calculated by a separate calculator. Both numbers are needed for a decision: the one-time figure shows the entry cost, the monthly one shows the cost of ownership.
What to do if the amount exceeds the planned budget
Reduce the number of scenarios to one and the number of connected systems to a minimum. The first scenario delivers a measurable result and the data needed to estimate the rest. Such an approach cuts the entry cost by two or three times and preserves the main thing — testing the hypothesis on a real process.
Check it on your case?
We will look at your process and give numbers for it, not for averages.
Related tools
Updated: September 8, 2026