By Artem Korsakov ·
Comparing supplier quotes gets complicated when one supplier quotes a price per metre, a second per coil, a third changes the material grade and leaves out delivery. Simply sorting by the total amount picks options that are not comparable. Construction procurement benefits from an assistant that brings quotes into a single format, shows gaps and keeps a verifiable link to the source documents.
In short: AI extracts line items and terms from emails, PDFs and spreadsheets, suggests how they map to the request and flags differences. The buyer checks comparability, an engineer confirms technical deviations. Totals are calculated by approved rules, and the purchasing decision is made by a person.
Comparison starts with the site requirement
Before reading quotes, you need the original request: the site, the section, the designation of the material or equipment, quantity, unit, completeness requirements and the required date. A link to the current specification helps check what exactly the supplier priced.
For industrial equipment, the comparison may include components, installation supervision, documentation and acceptance terms. Their composition is defined by the requirements of the specific project. If one supplier offers a complete set and another only the main unit, the same short name does not make the prices comparable.
Whether alternatives are allowed is recorded separately: an exact match is required, or a technical specialist is willing to consider an offer with a deviation. AI must not turn a similar trade name into an approved equivalent.
What data to extract from each quote
For the first version, two levels are useful: the details of the whole quote and the data for each line. The details include the supplier, number, date, currency, validity period, payment terms and delivery terms. The lines contain the original name, designation, quantity, unit, price, amount and explicitly stated characteristics.
Next to each extracted value, a link to the page, cell or email fragment is stored. Unstated delivery stays unknown; it is not treated as free. If a lead time applies to the whole order, the assistant does not distribute it across lines as confirmed individual dates.
Documents may contradict each other: the email promises stock availability, the attached quote says made to order. The service shows both values and drafts a question to the supplier. Picking a convenient answer without checking means hiding real uncertainty.
How to map items to the specification
First, exact designations and an agreed reference list are used. When the match is incomplete, AI suggests a pair “request line, quote line” and explains the differences found. The buyer confirms the match or sends a question to a technical specialist.
For a hypothetical cable, the characteristics set by the design matter; similar letters in the name are not enough. For equipment, the version, power supply and scope of supply may differ. Which parameters are critical is determined by the engineer; the assistant only checks whether explicitly specified values are present and match.
One request line is sometimes covered by several quote lines, and one package offer by several requirements. In the first version, it is better to show such a mapping explicitly to a person. There is no need to promise error-free automatic assembly of a complex specification from arbitrary documents.
Units, packaging and taxes
A price per package is comparable to a price per unit only when the package contents are known. The conversion is done by an ordinary formula using a confirmed factor. If coils come in different lengths, the model does not substitute a typical value from general knowledge.
The tax regime, the VAT indication and the currency are kept from the source. The comparison basis is chosen by a finance specialist taking the company's situation into account. You cannot automatically assume that a quote without VAT is cheaper by a fixed percentage or infer the applicable rate out of habit: you compare the confirmed terms of specific quotes.
For prices in foreign currency, the chosen exchange rate, its date and the conversion rule are stated. If the total depends on the rate on the day of a future payment, the comparison comes with a scenario. It must not be presented as the final cost in the reporting currency. The external rate source and its updates are designed separately.
A hypothetical example: a low line price does not mean a low total
The site needs 1,000 metres of material. All the following amounts are illustrative, in Russian roubles (RUB), and given on the same agreed tax basis. Supplier A offers RUB 100 per metre and delivery for RUB 12,000. The comparable total is RUB 112,000.
Supplier B offers ten coils of 100 metres at RUB 10,800 each, delivery included. After checking the coil length and technical compliance, the total is RUB 108,000. The difference is RUB 4,000 in favour of B under this set of terms.
But if B confirmed a delivery date later than the required date, its offer does not automatically win. The buyer shows the date and its source, and the manager checks the impact on the work. The unknown cost of downtime cannot be invented and added for the sake of a ready-made ranking. The final decision may take into account another verified supply option.
AI extracts the composition of the offer and the exclusions; the buyer confirms comparability. The numbers in the example are illustrative and are not market prices.
Illustrative example on the same tax basis comparing the full cost of two quotes. A: 1,000 m at RUB 100 plus delivery of RUB 12,000. B: 10 coils of 100 m at RUB 10,800, delivery included. The delivery date is checked separately.
A lead time must be read together with its condition
“Six weeks” is not a calendar date until the starting point is known. It may depend on an advance payment, order confirmation or approval of documentation. The service stores the original wording, the selected event and the calculated date after an employee has checked it.
For long manufacturing lead times, it is useful to highlight intermediate milestones if they have been agreed: approval of documentation, readiness, shipment, delivery. A promised shipment is not equated with arrival at the industrial site. Unloading and access conditions may also need clarification.
An expired quote remains in the history but is not shown as a current price. A new version of the quote replaces it only after confirmation, with the differences kept. AI can prepare a list of changes: an item has become more expensive, the validity period has shortened, the prepayment amount has changed.
What to keep in 1C and the existing procurement process
1C:Integrated Automation (1C is a widely used Russian business accounting platform) already provides for storing purchase terms, orders and supplier prices. 1C:ERP for Construction Companies (USO) links the needs of construction work with purchases and deliveries. These functions must be checked in the specific installed configuration.
Procurement automation in construction can start with handling an inconvenient input: quotes arrive in different formats, and an employee transfers the data by hand. The add-on prepares a verified comparison table, while purchase approval and the order stay in the existing system. An agreed export is possible first; automatic exchange requires a separate check.
Do not create a second supplier list without matching rules. The counterparty identifier, the quote version and the link to the request must be preserved when the result is transferred. Changing a line after review means it requires re-confirmation.
How to set up an AI assistant for a buyer
AI in procurement is useful at four steps: extract data, suggest mappings, highlight non-comparable terms and prepare questions. Processing fields and tables from documents is described, for example, in Microsoft AI Builder. This is an example of the technology; it does not guarantee quality on your suppliers' scans.
The interface shows the source document on the left and the extracted values on the right. The employee corrects disputed lines and confirms them. Price, quantity, currency, designation and lead-time conditions require especially careful checking. Hidden spreadsheet rows, poor scans and illegible characters should be sent for manual review.
After review, formulas calculate the comparable total. The assistant writes an explanation of the confirmed differences and a draft request for missing data. Sending the email, choosing the supplier and placing the order are done by an authorised employee.
How to check the result on your own documents
Take one typical procurement category and a set of anonymised quotes, including different units, partial coverage of the request, a substituted item and a new version of a quote. Prepare the correct table by hand in advance. Then compare processing time and errors in critical fields.
Comparing commercial quotes is considered useful if the employee gets a verified picture faster and sees where data is missing. The percentage of filled cells is not enough: a wrongly filled price is worse than an honest note that the information is missing. Correction time is included in the pilot evaluation.
A prototype can be built on one real anonymised request and several quotes. It will show comparable items, terms and questions, and provide an export for your established procurement process.
Related reading: Material requests in construction: from site foreman to delivery; Supplier invoice reconciliation: order, receipt and help from AI.
Частые вопросы ### Can AI choose a technical equivalent by itself?
It can suggest a candidate and show the differences against the specified parameters. Whether the substitution is acceptable is confirmed by an authorised technical specialist. A similar name is not enough for that.
Can quotes be compared if some terms are unknown?
Yes, but the unknown fields stay visible. The total is marked as incomplete, and the employee gets a list of clarifications. Missing delivery, components or lead times are not turned into zero.