f(AI) CustomOffer: a Personalized Commercial Proposal from 1C in Under a Minute

A CRM may contain a client’s industry, size, meeting history, product interest, and trial results—while the proposal still promises “an individual approach and a broad range of services.”

That gap led to f(AI) CustomOffer, developed for the largest official ConsultantPlus distributor in St. Petersburg and the Leningrad Region. A finished, formatted, personalized proposal now takes under a minute instead of 20–45 minutes of manual preparation. Ten weeks passed from kickoff to production launch; the acceptance documents are signed and commercial operation is beginning.

Why so much data still produces a generic proposal

Managers previously prepared proposals from templates, accounting for dozens of parameters and avoiding commercial errors; one calculation could take up to an hour.

1C already held industry, business size, arbitration and public-procurement information, relationship history, trial results, hotline requests, used services, attended events, and requested consultations—across about ten tabs. After a small internal investigation before every letter, attention naturally went to amounts and details, not subtle recipient motivation. Arguments therefore often remained generic despite an existing selection playbook.

Giving that context to AI sounded attractive. First, though, we had to define “perfect” for each recipient and identify who was responsible for every figure.

Three people in one company may be buying three different kinds of value

An executive asks what the purchase does for the business; a lawyer asks how it helps their work; an accountant asks which tasks become easier. The product is one, but a shared benefit list makes every reader find their own relevance.

CustomOffer selects arguments by recipient role, industry, and available client facts. The implemented project has separate catalogs for an executive, lawyer, and accountant; the model never receives another role’s arguments. The executive gets an economic rationale; lawyers and accountants get professional-task benefits without that block.

The conversation stage also matters: before a presentation, after it, or after a completed trial. Confirmed service use after a trial makes the proposal more concrete. Other businesses can configure roles such as technical specialist and procurement manager, but that needs their own rules and materials.

What happens between 1C data and the finished PDF

The manager remains in 1C, rather than moving information to a chat and explaining commercial policy again.

1C sends an agreed structured-data set. CustomOffer validates it and forms the permitted material set for the chosen recipient. The first model call identifies client priorities and selects approved arguments; the second prepares the introduction, transitions, and conclusion while preserving approved examples’ meaning.

The application inserts selected arguments verbatim, adds the calculated rationale, and assembles the result. Data returns to 1C for formatting, final PDF creation, and storage. The service runs in Yandex Cloud and uses Alice AI; the manager presses the familiar 1C button and downloads the document about 30 seconds later.

Any good automation must eventually stop requiring admiration for its architecture.

Economic rationale: where the annual figure comes from

Faster preparation helps the seller. The buyer wants to know the value of the product and service.

For executives, CustomOffer combines several components through a separate mechanism with approved formulas and parameters. The model does not calculate or change amounts.

Service usage. Trial-period consultations and educational events may be free to the client but still have value. With prices and quantities from 1C, the service calculates actual use and, where a completed period exists, an annual projection. It does not invent a value when none is supplied.

User productivity. An approved method links potential productivity growth to the value of working time. This implementation estimates one average user and time freed for other tasks, not headcount reduction.

Tax and other regulatory risks. The agreed method estimates possible reduction in expected losses, using revenue and industry where available. If revenue is unknown, the tax component uses an approved statistical scenario; if industry is unknown, agreed general parameters apply. This assesses potential risk; it does not assert that violations have been found.

Actual service value is separate from annual projection, so it is not counted twice. Each calculation line traces to source data and a rule. Future use, productivity, and risk reduction remain assumptions: the total is potential effect, not guaranteed savings.

Attachments to a commercial proposal can support the argument too

A proposal may include material showing what the client has already received. Here, it is a service report generated with the proposal: services used, quantities, and supplied values. Missing prices remain visible but do not enter the total.

The report is assembled programmatically, without a model. For a list of confirmed services, creative inspiration is a hindrance. Other projects choose attachments according to what helps the buyer decide.

Why AI cannot freely rewrite arguments here

We initially designed the algorithm around my sales expertise and existing rules, recommendations, and manuals. The core product was ready in about five weeks. Then a new responsible director clarified the required result: arguments had to come only from the supplied base, wording had to remain verbatim, and economic rationale had to follow approved guidance. Independent creativity was no longer intended.

That distinction is useful. Finding meaning in client context is an intellectual task; deciding the promises a company can make is business responsibility.

The model therefore does not rephrase approved arguments. It may adapt introductions and conclusions within the specified meaning, but prices, amounts, and facts are outside its discretion. Consultation topics can influence argument choice but must not be quoted or retold. Personalization is not publication of internal request history.

Checks control selected materials, response structure, and other constraints. Failed responses block proposal issuance. The model is not infallible, but its boundaries are concrete: transitions may vary; commercial rules must not depend on inspiration.

What has already changed—and what is still to be measured

The project reached production launch in ten weeks. Preparation fell from 20–45 manual minutes to under one minute, ending with a formatted document in the familiar system.

For each request, the system considers role, selects arguments, adds an executive rationale, and prepares additional material where necessary. The playbook no longer needs to be reproduced from memory.

Conversion and sales impact still must be measured. A quicker proposal and a closed deal are different outcomes; product, price, client need, and seller work remain between them. But sellers now have fewer reasons to spend much of that time assembling the document.

Where else this approach can be useful

The approach can support complex B2B sales of equipment, enterprise software, professional services, and service contracts. The signs are useful client data, argument choice that affects proposal content, and expertise or rules otherwise applied manually every time.

It requires company-specific data, materials, calculations, and integrations. There is no universal button that discovers commercial policy by itself.

The same logic applies beyond proposals: analyze a process, formalize rules, apply AI where matching and text work are needed, check the result, and return it to the working system. This is the kind of implementation f(AI) Studio delivers. CustomOffer is a complete AI function that uses business knowledge in everyday work.

A useful starting question is: how many times have employees today rebuilt something the company as a whole has long known?