Almost anything can be automated. The problem is that budget and team attention are finite, and the first project decides whether there will be a second. Start with the wrong process and the company ends up with a working agent nobody uses — and the conclusion that "AI isn't for us".
Below are five tasks that, in companies of 10–150 people, almost always deliver the best return per dollar, plus the rule for choosing which one goes first.
Prioritisation rule: pick the process where all three are true — (1) a lot of manual time, (2) the steps repeat identically, (3) errors cost money. If all three line up, that is your first project.
1. Handling inbound enquiries
Why it ranks first. Every enquiry left unanswered for two hours is usually a lost client who messaged a competitor in the meantime. Here automation affects revenue directly rather than cost, which is why it pays back fastest.
What the agent does: replies instantly to the first message, asks 2–3 qualifying questions, answers routine questions about services, creates a CRM record, and hands a hot lead to a manager with the context already gathered.
Savings: 2–4 hours a day of sales manager time, plus a higher enquiry-to-meeting conversion rate purely from response speed.
Typical stack: Telegram / Instagram Direct + CRM (Bitrix24, amoCRM or your own) + a language model.
Don't ask the bot to close deals. In version one its job is to qualify and hand over. Trying to automate negotiation at the start is the most common way to sour the whole initiative.
2. CV screening and recruitment
Why it works. HR spends up to 80% of hiring time reviewing irrelevant applications. The task is formalisable: defined criteria, a document as input, and a clear output — a ranked list.
What the agent does: reads CVs, checks them against your criteria, ranks candidates, politely declines clear mismatches, and hands HR a short list with reasoning for each entry.
Savings: 15–20 hours a week during active hiring. Screening a hundred CVs takes minutes instead of two working days.
The side effect that often matters more than the savings: a strong candidate hears back the same day and doesn't accept a competitor's offer first.
3. Content production and adaptation
Why it works. Content is a high-volume, low-error-cost process: the draft always passes through a human, so risk is minimal while the time saved is substantial.
What the agent does: drafts posts in your categories and tone, adapts one piece across channels, produces A/B headline variants, and translates into RU/EN/UZ while preserving meaning.
Savings: content production speeds up 5–8×. Important: that's a saving on drafting, not editing — the final call stays with a person.
What not to expect: that the agent replaces strategy. It accelerates production; it does not decide what you should be saying to the market.
4. Financial reporting and reconciliation
Why it works. The most formalisable process of the five. Clear input (a document), clear output (an entry), unambiguous rules. Automation takes the largest share here — up to 70–80%.
What the agent does: recognises incoming invoices, codes them to the right categories, reconciles data across systems, assembles recurring management reports, and flags discrepancies.
Savings: document processing speeds up dramatically, but the main gain is a lower cost of errors. Manual entry produces 1–3% errors, and an error caught late costs time to unpick and sometimes penalties.
A human in the loop is mandatory here. Financial operations are the zone where the agent prepares and proposes, but an employee always confirms.
5. Internal knowledge base and FAQ
Why it works. In any company above 20 people the same questions get asked dozens of times a month: how do I book leave, where is the contract template, what terms does client X have. The answers exist, but they live in people's heads and chat threads.
What the agent does: answers employee questions from your internal documents, cites the source, and when the answer isn't in the knowledge base, says so honestly and routes the question to a person.
Savings: 5–10 hours a week for managers and senior staff who get interrupted constantly. New-joiner onboarding speeds up as a bonus.
More on the technology in our article on RAG systems.
Summary table
| Process | Time saved | Automatable share | Setup difficulty |
|---|---|---|---|
| Inbound enquiries | 2–4 h/day | 50–70% | Low |
| CV screening | 15–20 h/week | 60–80% | Low |
| Content | 5–8× faster | 60% | Low |
| Finance and reconciliation | 15–25 h/week | 70–80% | Medium |
| Knowledge base | 5–10 h/week | 60–70% | Medium |
Note that nothing reaches 100%. That isn't cautious phrasing — it's reality. Every process retains a share of exceptions and judgement that a human takes.
Frequently asked questions
Which of the five should we start with?
The one that hurts most right now. If sales can't keep up with enquiries, start with the first. If accounting is drowning in documents, start with the fourth. There is no universal "correct first" — there is your most expensive process.
Can we automate two processes at once?
Technically yes, practically we advise against it at the start. The first project also teaches your team how to work with agents. Two parallel rollouts double the load on people and lower the odds that either gets finished.
What if our process isn't on this list?
The list is frequency statistics, not a constraint. Run your process through the prioritisation rule at the top: manual time, repeatability, cost of error. If it matches, the process qualifies regardless of industry.
How long does the first process take?
Usually 2–3 weeks from sign-off to a working pilot. A day-by-day breakdown is in our article on implementing AI in 2 weeks.
What do we need to prepare on our side?
Access to the system where the process lives, and one person who knows how it actually works. The documented process and reality usually diverge, and the agent has to be taught reality.
Where to start
Pick the one process from the list that causes the most pain and price it: hours per week × cost per hour × 50 weeks. If the resulting number is meaningful to you, that's your first project.
Want to work it out together? We run a free process review: we find where the time goes and tell you straight if there's nothing worth automating yet.
Read next
- How much AI automation costs: a budget breakdown
- How AI automation saves businesses $48,000 a year
- 7 mistakes in AI implementation and how to avoid them
- How to measure AI ROI
- Service: AI agent development and deployment
The right first process isn't the most technological one — it's the most expensive one that gets done the same way every time.