Data and connections
Enquiries and attachments, product information, ERP and ticket handling support the path from customer requirements to a prepared quote.
A quote ready for the client, days ahead of the competition
Salespeople were overloaded with manual work, so customers waited an average of 7–8 days for a quote. A quoting agent with skills and connections to ERP, CRM and the helpdesk prepares the quote, and a salesperson checks and approves it. When the agent handles the enquiry, the quote reaches the customer within a day, while salespeople focus on closing sales rather than producing quotes.
Results
- 7 days → 1 day
- response time for enquiries handled by the agent
How it works
What the agent does
Reading the request and 2 attachments
specification.xlsx · drawing-A12.pdf
Identifying the items0 / 38
Matching them to the catalog
36 matched · 2 for the sales rep to decide
Prices and framework agreement discount
7% discount · valid until Dec 31
Client’s order history in the CRM
last 3 orders
Building the quote in the company template
PDF + item sheet
Quote QT/2026/1042
Ready for reviewFor
Procurement · B2B client
Delivery date
by Nov 28, 2026
Aluminum profile 40×40120 mPLN 18,600
Corner bracket480 pcsPLN 6,240
Ball valve DN2564 pcsPLN 9,920
Pneumatic actuator24 pcsPLN 31,440
+ 34 items
Net totalPLN 0
Review and sendEdit
Sent to client · 1 day instead of 7
01
Challenge
The sales support team receives several thousand requests a month: email text, PDF and XLSX lists, scans and photos, often with competitors' product names. A salesperson read the attachments, translated each line item into a product and configuration from the catalogue, and then retyped it into the ERP. This knowledge sat in the heads of a few experienced people. The customer waited an average of 7–8 days for a quote and often asks several suppliers at once.
At higher enquiry volumes, manual preparation delays responses. Salespersons are needed for exceptions and customer conversations, yet spend much of their time preparing data.
02
Solution
The Quoting agent picks up the request from the helpdesk, reads the text and attachments, extracts line items with quantities and parameters, matches products and prepares the quote. The salesperson approves it before it goes to the ERP and replies to the customer personally. The ERP always calculates the price. Non-standard and off-catalogue items go to a person for a decision.
03
How we built it
The state of each request is stored in a database, so after a failure the process resumes from the interrupted stage: reading attachments (OCR for scans and photos), extracting line items, matching to the catalogue, configuration and the quote in the ERP. Every line item the model extracts must have a verbatim citation in the source. Recurring items are recognised by rules without the model, which is cheaper and carries no risk of hallucination. The product configuration is confirmed by the ERP configurator, not by the model's self-assessment. The models have no access to systems, and every call is logged with its cost, time and prompt version. Any change of model or prompt is tested on a set of requests that the client confirmed as correctly processed.
ClaudeOpenAIMistral OCRLaravelPostgreSQLClient's ERPClient's helpdesk
Workflow
How the solution works
- 01
Read the enquiry
The agent extracts the enquiry and line items from attachments, preparing inputs for product selection.
- 02
Select and price
It matches products to requirements. Prices follow ERP rules rather than a language model’s answer.
- 03
Review and respond
A salesperson checks the quote, resolves unusual line items and approves the result before further customer handling.
The human role
The salesperson retains the decision on the quote. The agent prepares the result and reduces manual reading and data transfer.
Lessons
What made the difference.
- 01
Reads any attachment
The agent turns email text, PDFs, spreadsheets, scans and photos into a list of line items with quantities and parameters, even when the customer uses competitors' product names.
- 02
Every line item with evidence
The model can only return a line item that can be pointed to in the source, so the salesperson sees where each line of the quote came from.
- 03
The ERP sets the price, not the model
The model does not guess parameters or prices: the ERP confirms the configuration, and the salesperson approves the quote before it is sent.
AI agentDocument extractionERP integrationHuman oversight
Company of the Future
How this builds the Company of the Future
A project delivers results on its own, but it also leaves organized context behind in the company. Every next AI solution builds on it. This project added to the AI Engine:
- Context Lakehouse
Customer requests with attachments and their matches to the catalogue are stored in a database, each item with a citation from the source.
- Skills
What a few experienced salespeople knew about turning an item into a product is written down: repeatable items are recognised by rules, without the model.
- MCP
A controlled connection to the helpdesk and the ERP: the ERP always calculates the price, the ERP configurator confirms the configuration, and models have no access to the systems.
Value
Where value is created
Time saved
Less manual reading and retyping of line items.
Sales and revenue
Faster responses let the team return to customers with a quote sooner.
What comes next
Further development expands supported variants and improves rules using reviewed examples.
Most of our projects are covered by NDAs, so we only share part of them, without client names or details, as inspiration. We are happy to tell you more about our experience in a conversation.
Let’s talk about the work you want to improve.
Tell us how a process works today. Together, we will explore where AI Engine can help and which module to start with.
Book a callOr write to: hello@wearefuture.ai
