Three Jobs the Catalogue Prices Twice — and What the Higher Bill Buys

Two shelves, one problem

Somebody in the office spends three hours a day retyping supplier invoices. Somebody else answers the same six customer questions all week. A third person opens competitors' websites every morning and copies prices into a spreadsheet. Each of those problems has a price in the VIT MARKET catalogue. Each of them has two prices, because each appears once in the automation section and once in the artificial-intelligence section, described in near-identical terms and quoted 2.14, 3.33 and 6.84 times apart.

This is the question the catalogue's own navigation cannot answer, because the two listings never appear on the same page. Read side by side, though, the paired descriptions say fairly precisely what the higher price buys — and in one of the three pairs, VJOURNAL would argue it buys something a buyer may not need.

A disclosure first. VIT MARKET is VITON13's services catalogue and VJOURNAL is VITON13's newsroom, so nothing here is asserted from memory. Prices, durations and difficulty labels are read from the catalogue's own fields; midpoints are floor plus ceiling divided by two; the ratios are computed from those midpoints. The pairing of listings is our reading of their published descriptions, not a claim the catalogue makes anywhere.

The two shelves, measured against each other

Both sections hold exactly twenty-five listings. That is where the symmetry ends. The automation section's median midpoint is $340; the AI section's is $1,350, a ratio of 3.97. Automation floors run from $50 to $400 with a median of $150; AI floors run from $300 to $1,500 with a median of $700. The highest ceiling anywhere in automation is $1,500 — a figure twenty of the twenty-five AI listings exceed.

The gap is widest at the entrance. Six automation listings have a floor of $100 or less, and four can be bought whole, ceiling included, for under $300. The cheapest AI listing starts at $300. In other words, four complete automation services cost less than the AI section's opening price.

Speed splits the same way, and for a structural reason. All twenty-five AI durations are denominated in working days. The automation section quotes nineteen listings in days and six in hours, the fastest at two to four hours for $60 to $200. Nothing in the AI section is available in hours at all; its fastest listing is three to six working days.

One thing the section labels do not tell you is which technology you are buying. Four of the twenty-five automation listings mention AI in their own text: a Telegram digest that arrives with an AI summary and a first-line AI answering machine name it in the title, while two others name it only in the delivery estimate — the build is done with a model, the delivered thing is a scraper and a lead-intake bot. Sixteen of the twenty-five AI listings do not carry the letters AI in their title at all. The shelf is a price and delivery bracket, not a statement about what is under the hood.

Pair one: invoices out of PDFs, at 3.33 times

The automation listing is digitisation of invoices from PDF to table without manual entry: $350 to $1,000, three to six days, tagged Advanced, with the duration string noting that half the time is calibration on the client's real documents. The output it promises is a table of positions, quantities, prices, supplier and document number, ready for loading into an accounting system.

The AI listing is recognition of invoices and contracts into structured data: $1,000 to $3,500, ten to eighteen working days, also Advanced. Its output is described as clean JSON or a row in an accounting system, itemised down to counterparty, tax number, document number, date, line items, totals and VAT — and it takes contracts, not only invoices.

Midpoints of $675 and $2,250 give a ratio of 3.33. The floors are 2.86 times apart and the ceilings 3.5 times apart, so the whole band shifts rather than stretching. On the listings' own text, the extra money buys three things: contracts as well as invoices, named tax and VAT fields rather than a generic table, and a machine-readable output rather than a spreadsheet a person still has to import.

VJOURNAL's reading: if a human opens the file before it reaches the ledger, the $675 version is already doing the job. The $2,250 version is priced for the case where nobody opens it.

Pair two: the same customer questions, at 2.14 times — with the labels inverted

The automation listing is a first-line AI answering machine built on the company knowledge base: $341 to $1,200, three to five days. Its description claims it answers 60 to 80 per cent of typical questions — how much delivery costs, how to return something, where an order is — in the company's own words, and hands over to a person when unsure. That percentage is the listing's claim; VJOURNAL has not verified it and it cannot be checked against anything in this repository.

The AI listing is a knowledge-base chatbot for the website, built as retrieval-augmented generation: $800 to $2,500, five to ten working days. Its description promises answers strictly from the client's documents, price lists and FAQ, a link to the source document in every answer, and a handover to a live manager when confidence is low.

Midpoints of $770.50 and $1,650 give a ratio of 2.14 — the narrowest gap of the three pairs. Read the two descriptions closely and the difference is not really the answering. It is the citation. One of them shows the customer where the answer came from; the other does not say it does.

There is a detail here worth pausing on, because it will mislead anyone shopping by label. The cheaper listing carries the catalogue's hardest difficulty tag, Advanced. The dearer one is tagged Experienced. Across this pair the difficulty label runs backwards to the price, which is a useful reminder that difficulty in this catalogue describes the work, not the invoice.

Pair three: watching competitors' prices, at 6.84 times

This is the widest gap in the catalogue that VJOURNAL could find between two listings describing the same job. The automation listing is a competitor radar, a price and assortment parser writing into a spreadsheet: $80 to $300, four to eight hours, tagged Beginner, delivering a Google sheet that refreshes daily with name, price, availability and link, plus a changes-per-day tab. Its own duration string says the scrape is built with AI, and adds one to two hours for antibot debugging.

The AI listing is extraction of website data using AI instead of selectors: $600 to $2,000, five to twelve working days, tagged Experienced. Its pitch, in its own words, is a parser that does not break with every site redesign, because the data is read by a model that understands the page rather than by fragile CSS selectors.

Midpoints of $190 and $1,300 give a ratio of 6.84; the floors are 7.5 times apart. And notice what the higher price is not buying. Both listings already use a model. The dear one is not more capable at reading a page — it is more durable when the page changes. That is a maintenance product sold at the point of purchase.

VJOURNAL's reading, and it is a reading rather than a finding: a 6.84 multiple on durability only pays if the sources you are scraping actually redesign often, and nothing in the catalogue tells you how often that is. For three to five stable sites, the $190 listing plus an occasional repair is the arithmetic to beat.

The pair that is not a pair

One more comparison looks identical and is not, and getting it wrong is expensive. Both sections carry a CRM listing. In automation it is site enquiries straight into amoCRM or Bitrix24 without loss, $300 to $1,200 over two to four days: every form, call and message arrives as a deal with its source, its UTM tags and a routing rule, with phone-number duplicates merged. In the AI section it is implementation of AI in an existing CRM, $1,000 to $3,500 over ten to twenty working days: the model fills in the deal card after a call, summarises the correspondence, proposes the next step and flags deals at risk.

The midpoints are $750 and $2,250, a ratio of 3.0 — and the ratio is meaningless, because these are not two versions of one job. One gets records into the system. The other writes inside records that are already there. A company with no reliable intake buys the second listing and still loses leads before they arrive.

The catalogue never claims these are alternatives. It also never warns that they are not, which is exactly the kind of thing a category page cannot do and a comparison can.

What the catalogue cannot settle

The single most important limitation is that the two shelves do not measure time the same way. AI durations are all in working days. Automation durations are mostly in calendar days, with six in hours. When this piece reports three to six days against ten to eighteen working days, those are not the same unit, and the first-party dataset that underpins the comparison says so plainly: its eight-hours-per-day conversion is a constant the script declared rather than measured, and figures that depend on it are labelled accordingly.

Nothing in the catalogue reports accuracy on both sides of any pair, so no reader can tell whether the dearer listing is also the more correct one. Only one listing in the three pairs publishes a performance figure at all. Nothing records how many buyers of a cheap version came back for the expensive one, which would be the single most informative number here. And neither shelf prices what any of these systems cost to run once built.

The pairing itself is ours. The catalogue does not link these listings, does not describe them as tiers of one product, and might reasonably say they were built for different buyers. We paired them on the substance of their published descriptions, and a reader who disagrees with a pairing can check it against the same two texts.

The rule that survives the data

Across all three pairs the higher price is buying the same category of thing, and it is not intelligence. It is what happens when no human is standing between the output and its destination: structured fields instead of a spreadsheet, a source link instead of an answer, resilience instead of a script that needs fixing. Every one of those matters only when nobody is checking.

So the practical test is not which technology to buy. It is who reads the output next. If a person reviews it before it is used, the automation shelf is priced for you, and 3.33 or 6.84 times is being spent on assurance you are already providing in person. If the output goes straight to a customer, a ledger or another system, the AI listing's extras are the product, not the packaging.

Two guardrails from the data before signing anything. Do not let the difficulty label do your budgeting — the pair that answers customer questions has the cheaper listing tagged Advanced and the dearer one Experienced, and the catalogue's only Beginner-tagged AI listing carries a $750 midpoint, higher than five of the eight Advanced automation listings. And check that you are looking at a pair at all: the two CRM listings sit 3.0 times apart and do different jobs, so a buyer comparing them by price is comparing nothing.

Decision framework: automation versus AI for business

Use deterministic automation when rules and inputs are stable; consider AI when the work requires interpretation of variable language, images or context. Many useful systems combine both and route uncertain cases to a person.

The higher quote should buy evaluation, exception handling, data controls and monitoring rather than an AI label. Without those controls, added model complexity increases operational uncertainty.

Map each step as rule, judgment or approval; automate the rules first, test AI only on bounded judgment tasks, and retain a human gate wherever failure has material consequences.