Which Automation Pays for Itself First: Reading a Price List by Hours Saved Instead of Hours Billed

The only number that matters is not the price

Every automation in the VIT MARKET catalogue replaces a task somebody is currently doing by hand. That makes it the rare category where the return can be calculated before the purchase rather than argued about afterwards, because both sides of the equation are knowable: what the build costs once, and what the manual version costs every week.

The arithmetic is unglamorous and it is decisive. A price monitor with instant Telegram alerts is listed at $100 to $400 for 3 to 6 hours of work. If checking competitor prices by hand currently takes someone forty minutes a day, that is roughly thirteen hours a month. The build pays for itself inside the first month at almost any hourly rate, and every month after that is profit.

Now take the opposite end. Invoice and contract recognition into structured data is $1,000 to $3,500 for 10 to 18 working days. If the company processes thirty documents a month and each takes six minutes, the manual cost is three hours a month, and the payback period runs into years. The expensive item is not worse — it is simply aimed at a different volume, and buying it below that volume is buying a fact about someone else's business.

This piece sorts the catalogue by that logic rather than by price, which reverses the order in several instructive places.

Tier one: things that pay back in weeks

Four entries share a shape: low price, low hours, beginner-level difficulty, and a task that repeats daily. The competitor radar — a price and assortment scraper writing into a spreadsheet — is $80 to $300 for 4 to 8 hours across three to five sites, with another one to two hours for anti-bot handling. The price watchdog with instant Telegram alerts is $100 to $400. The marketplace export pulling orders and stock from WB and Ozon into a working sheet is $120 to $450 for one to two days across two platforms. The social autopilot, turning a content plan in a spreadsheet into scheduled posts, is $120 to $400 for one to two days across three platforms.

What unites them is that the manual version is not merely slow, it is unreliable. A person checking competitor prices misses the day they are ill. A person watching for a stock change notices it in the evening rather than at nine in the morning. The automation is not buying back an hour; it is buying back an hour that was already being paid for and not always delivered.

The anti-bot line in the scraper entry is worth reading closely, because it is where this tier's honesty lives. One to two extra hours is a realistic allowance for the sites that resist scraping, and a quote that omits it is a quote that will be revised. Sites change their defences; a scraper is a thing that needs occasional maintenance, not a thing that is finished.

For a business that has never automated anything, the correct first purchase is almost always in this tier, and not because it is cheap. It is because a fast, visible win establishes internally that this category works, which is what makes the second purchase possible.

Tier two: the plumbing between systems

The integration entries — Google Sheets to CRM two-way sync at $150 to $600 for one to three days, website leads into amoCRM or Bitrix24 at $300 to $1,200 for two to four days — solve a problem that is invisible on any org chart and expensive on every one.

Data that has to be retyped between two systems is data that will diverge. Not might: will. Somebody updates the spreadsheet and not the CRM, the CRM becomes the version nobody trusts, and within a quarter the company is running on a third source of truth that lives in one person's head. The cost of that is never booked as a line item, which is exactly why it survives so long.

The lead-capture entry is the one with the sharpest business case, because the loss is countable. A form submission that fails silently is a customer who contacted you and got nothing. The catalogue's note that amoCRM with OAuth takes longer while Bitrix24 via incoming webhook is faster is the sort of detail that separates a real estimate from a guess — the same deliverable has different hours depending on which system is on the other end.

Payback here is measured in recovered leads rather than saved hours. One additional closed deal a month usually clears the whole build cost, which makes this the tier where the return calculation is easiest to defend to whoever signs.

Tier three: bots that answer people

Two entries look similar and are not. The lead-intake Telegram bot — a form that writes into the CRM and the chat — is $150 to $500 for one to two days on a five-to-eight-step script, marked "needs experience". The AI first-line responder built on the company knowledge base is $350 to $1,200 for 3 to 5 days, marked "expensive and complex", with the catalogue noting a Fiverr median of $341 to $520 for chatbot projects.

The difference is that the first bot follows a script and the second one does not. A scripted bot has a finite set of paths, which means it can be tested exhaustively and it fails predictably. A knowledge-base bot answers questions nobody anticipated, which is the whole point and also the whole risk.

The catalogue's breakdown of the five days is the most useful sentence in the entry: one day to clean the knowledge base, the rest to assemble and configure refusals. Both halves are counterintuitive to buyers. The knowledge base is usually assumed to exist already, and it usually exists as a folder of contradictory documents that nobody has reconciled — a bot trained on it will confidently repeat the contradictions. And configuring refusals, teaching the bot to say it does not know rather than invent, is the deliverable that decides whether this is an asset or a liability.

The rule of thumb: buy the scripted bot when the questions are known and the volume is steady. Buy the knowledge-base bot only when someone owns the knowledge base as a maintained document, because the bot inherits its quality exactly.

Tier four: where the money and the risk both climb

The AI-for-business group starts where automation stops being about hours and starts being about capacity. A retrieval-based site chatbot is $800 to $2,500 for 5 to 10 working days. A Telegram assistant for clients is $500 to $1,500. An inbound email autoresponder is $600 to $1,800. CV screening against a specific vacancy is $600 to $1,800. A voice bot answering inbound calls is $1,500 to $5,000 for 12 to 20 working days.

The voice bot is the clearest example of the shift. Nobody buys it to save the receptionist forty minutes. It is bought because calls arrive outside working hours, or arrive in bursts that no realistic headcount can absorb, and the alternative is not a slower answer but no answer. The return is measured in calls that were previously lost, and that number is either large enough to justify $5,000 immediately or it is not.

Document recognition — invoices, delivery notes and contracts into structured data at $1,000 to $3,500 for 10 to 18 working days — is where volume decides everything. Below roughly a hundred documents a month it is difficult to justify. Above a thousand it is difficult to avoid. The middle is where careful arithmetic earns its keep.

The content pipelines in the same group price honestly in a way worth noting: batch product-card generation is $400 to $1,500 to set up and then $0.10 to $0.50 per card. Publishing both numbers is the difference between a tool and a subscription, and a buyer should insist on seeing the second number for anything that runs per item.

The document conveyor, and why half the hours are formatting

One entry deserves singling out because its cost breakdown is so unusual. The closing-documents conveyor — invoice, act and UPD in one click — is $250 to $900 for two to four days, and the catalogue states that half of that time is spent perfecting the template layout to accounting's requirements.

That is not padding. Financial documents are a format problem before they are a software problem: a field in the wrong position, a total rounded a different way, a stamp block that shifts when the item list runs long, and the document is returned. Automating the generation of a document that then gets rejected saves nothing at all.

For a buyer, the practical consequence is to involve the accountant before the build rather than after it. The single most common cause of overrun in this category is discovering the real formatting requirements during acceptance testing, when they should have been the specification.

Three questions before any of it

How many times a week does this task actually happen? Not how annoying it is — how often. Annoyance and frequency feel identical from the inside and produce opposite answers. The most irritating task in a company is often monthly, and monthly tasks almost never justify a build above the cheapest tier.

Who currently does it, and what would they do instead? If the honest answer is that nobody would do anything else, because the person is salaried and not at capacity, then the automation buys reliability rather than time, and it should be justified on reliability. That is a legitimate case, and it is a different case.

What breaks when it breaks? Every automation is a dependency. A price scraper failing quietly for three weeks is worse than never having had one, because decisions were being made on stale data the whole time. Anything in the first two tiers should be bought with an alert on its own failure, which usually costs nothing extra to add at build time and is nearly impossible to retrofit cheaply.

A buying order that survives contact with reality

Start where the task is daily and the build is under $400: the radar, the watchdog, the marketplace export. These pay back inside a month, they are verifiable, and they teach the organisation what it feels like to have a thing run without anyone touching it.

Then close the leaks between systems: lead capture into the CRM first, because lost leads are countable money, then the two-way sync that stops the spreadsheet and the CRM from disagreeing.

Then, and only then, consider the bots that talk to customers — the scripted intake bot if the questions are known, the knowledge-base responder if and only if somebody owns the knowledge base. The four-and five-figure items in the AI group are worth their price at the volume they were designed for, and are worth nothing at all below it. The catalogue publishes the hours next to the money precisely so that a buyer can do that arithmetic before signing rather than after.