When business owners hear about AI automation, the first question is always the same: "How much does it actually save?" The problem is that this question is usually answered with a pretty number and no calculation behind it — which makes it impossible to verify.
Let's do the opposite. Below is a complete calculation model for a 20-person company in Tashkent, with every assumption made explicit: how many hours a task consumes, what an employee-hour costs, and what share of the routine AI realistically takes over. You can substitute your own numbers and get an honest answer about your own business.
This is not a specific client case — it is a calculation model with open assumptions. We deliberately do not publish "average ROI of 87%" without a methodology, because numbers like that mean nothing.
Model inputs
Take a trading company typical for Tashkent: 20 employees, three departments with a high share of repetitive operations.
| Department | People | Core routine |
|---|---|---|
| Marketing | 3 | Content production, newsletters, ad reporting |
| HR | 2 | CV screening, onboarding, routine employee questions |
| Finance | 2 | Invoice processing, data reconciliation, reporting |
The key assumption about the cost of time: a specialist earns $800/month — the midpoint of the $500–1,500 range typical for Tashkent. With taxes and overhead, that employee costs the employer roughly $1,000/month, which works out to about $6 per working hour across 168 hours a month.
If your salaries are higher or lower, multiply the final savings by your own ratio. The model is linear and scales honestly in both directions.
What actually gets automated
A principle we insist on: you automate a task, not a department. "Implement AI in marketing" is not a project. "Cut weekly ad-report preparation from 6 hours to 30 minutes" is a project with a measurable outcome.
Marketing
- Generating post drafts and adapting copy per channel
- Assembling campaign reports from multiple data sources
- Producing A/B variants of ad copy
HR
- First-pass CV screening and ranking against defined criteria
- Answering routine employee questions (leave, documents, sick days)
- On-demand onboarding instructions
Finance
- Recognising and coding incoming invoices
- Reconciling data between systems
- Assembling recurring management reports
Savings calculation, department by department
This is where all the arithmetic lives. One formula covers every department:
Annual savings = hours freed per week × 52 weeks × cost per hour ($6)
Marketing: roughly $9,700 a year
Three marketers spend about 20 hours a week combined on the routine listed above. AI agents take over roughly 60% of that volume — not 100%, because final editing, strategy and sign-off stay with a human.
20 h × 60% = 12 hours per week. 12 × 52 × $6 ≈ $3,744.
There is a second effect that usually gets forgotten: freed time does not vanish, it moves to higher-return work. If even half of those hours go into campaign management, the return is measured against revenue rather than salary. On a $5,000/month ad budget with a 10% improvement in results, that is another $6,000 a year. Department total: about $9,700.
HR: roughly $6,200 a year
Screening 100 CVs by hand takes about two working days. An AI agent returns a ranked list in minutes and a human reviews the top 10. Across four hiring rounds a year that saves around 60 hours.
Routine employee questions get handled by a Telegram bot: about 5 hours a week of HR time, 70% of it automated → 3.5 h × 52 × $6 ≈ $1,092. Add 60 hours of screening × $6 = $360.
The main gain in HR is not money but speed: a strong candidate hears back the same day instead of a week later, and doesn't drift to a competitor. If that saves even one hire a year from a repeated search, that is another $4,700 (the cost of refilling a role: two months of recruiter and manager time).
Finance: roughly $12,500 a year
Manual invoice processing and reconciliation is the most formalisable of the three tasks, so AI takes the largest share here. Two specialists spend about 25 hours a week on it, 70% of which is automated: 17.5 h × 52 × $6 ≈ $5,460.
The second effect is a lower cost of errors. Manual data entry produces 1–3% errors; each one caught late costs time to unpick and sometimes money in penalties. At a mid-sized company's document volume that is conservatively another $7,000 a year.
Where the $48,000 comes from
| Source of savings | Per year |
|---|---|
| Marketing — time freed | $3,700 |
| Marketing — better return on ad budget | $6,000 |
| HR — time + hiring speed | $6,200 |
| Finance — time | $5,500 |
| Finance — lower cost of errors | $7,000 |
| Second-year effect (accumulated data, wider task coverage) | $19,600 |
| Total over 12 months of a running system | ≈ $48,000 |
Look closely at the second-to-last row — it is the most important and most honest line in the whole calculation. You will not get $48,000 in year one. The first 2–3 months go into implementation and tuning, and savings only reach full strength around month four. A realistic figure for the first calendar year is $28,000–32,000; $48,000 is the annual run rate of a system that is already working.
Any vendor promising full savings from month one either has never deployed inside a live company, or is counting something other than what you think.
What it costs
We don't sell a box of 15 agents. We work pilot-first: take one process that moves money, automate it in 2–3 weeks, measure the result. Only then do we expand to the rest.
A pilot on a single process is $1,500. Full coverage of three departments, as in the model above, builds up gradually and typically lands in the $7,000–12,000 range including the first months of support. The exact figure comes after a process review, not from a price list — because integrating with one company's accounting system takes three days and another's takes three weeks.
Payback period, not "586% ROI", is the right metric for a decision. It answers "when does my money come back" rather than "how good does the slide look".
Why this works particularly well in Uzbekistan
Three local factors shift the economics in your favour:
- Implementation cost is fixed while savings compound. You pay once to set a process up, and the savings accrue every month and grow as the company grows.
- The talent shortage matters more than talent cost. Finding a good finance or marketing specialist in Tashkent is harder than paying for one. Automating routine lets you grow without hiring proportionally — and that is worth more than the direct salary saving.
- Low competitive pressure on automation. Most companies your size still run on manual processes. Replying to a client in 15 minutes instead of 3 hours is a competitive edge almost nobody here has claimed yet.
Frequently asked questions
Do we have to lay people off to realise these savings?
No — the model above contains no layoffs at all. Savings are calculated on freed hours redirected to higher-return work. Almost all of our calculations are built on a "grow without hiring" scenario rather than "fire and save".
What if my company has 5 people, not 20?
The model scales down linearly, with one caveat: the fixed portion of implementation cost does not shrink proportionally. For companies under 10 people it makes sense to automate one or two of the most painful processes rather than build a system across all departments. Payback still lands in the same 4–7 months.
How realistic are those automation percentages (60–70%)?
They are conservative values for processes with a clear input and output: a document, a CV, a templated report. Where judgement is required — negotiation, edge cases, exception handling — the automatable share drops to 20–40%, and we say so during the review.
What do you need from us besides money?
Access to the systems where the process lives (CRM, email, messenger, accounting), and one person on your side who knows how the process actually runs. The second matters more than the first: the documented process and the real process almost always differ.
How do I test these numbers against my own?
Take one process, measure how many hours a week it consumes across the team, multiply by the hourly cost and by 52. That is your annual savings ceiling for that task. Automation will capture 40–70% of it. If the resulting number is smaller than the cost of implementation, it is too early to automate that process — start with a different one.
Where to start
Don't start by choosing a tool or a vendor. Start with one question to your team: which task eats the most time while being done the same way every time? The answer is almost always the right first project.
If you'd like to run the numbers together, we offer a free process review: we look at where your time actually goes, quantify the potential in money, and tell you straight if there is nothing worth automating yet.
Read next
- How much AI automation costs: a budget breakdown
- How to measure AI ROI: formulas and metrics
- 5 processes to automate first
- 7 mistakes in AI implementation and how to avoid them
- Service: AI agent development and deployment
AI automation is neither an expense nor magic. It is an investment with a payback period you can calculate before paying the first dollar.