Data and connections
Product data and selection rules form a shared model. PIM and ERP connections depend on the implementation stage.
From knowledge scattered across many spreadsheets to one selection tool
An MVP of a tool meant to replace more than 30 selection spreadsheets: product formulas and dependencies are stored in a database, and a designer describes the need in a plain sentence and gets matching variants.
Pilot target
- up to 1 h → <10 min
- pilot target: product selection time
- approx. 99%
- of selection formulas from the ERP imported automatically
- 12 weeks
- from MVP start to a working application
How it works
Spreadsheet and ERP formulas in one database
imported without retyping
Need described in a sentence
filters you can correct
Live calculations
rules from the database, not code
PDF selection sheet
ready to send
- selection_category_A.xlsx
- selection_category_B.xlsx
- selection_category_C.xlsx
- ERP formulas
Product database
99% of formulas imported automatically
I need a variant for outdoor installation, capacity of at least 800.
Filters to check:
installation: outdoorcapacity ≥ 800category B
Configuration
CategoryB
Capacity820
Versionoutdoor
Order codeB2-820-Z
Rule validation: passed
Selection sheet · B2-820-Z
Variant parameters
Dimensions and weight
Order code: B2-820-Z
Installation notes
PDF readyPilot target:under 10 mininstead of up to an hour
01
Challenge
Product selection relied on more than 30 Excel spreadsheets and several separate selection programs. A designer collected PDF data sheets from different tools and retyped order codes by hand, and errors only surfaced at pricing. The selection logic lived in spreadsheets and in the ERP. Adding a product to the program required an external contractor.
A shared model organises the agreed selection scope. Experts verify the rules and additional product variants.
02
Solution
Together with the client's team we built a data model in which product parameters, formulas and dependencies are stored in a database, so the form, validation and calculations are generated from records, not from code. The import moves formulas from spreadsheets and the ERP library without retyping, and the catalogue syncs with the PIM. On this foundation run a configurator with live calculations, PDF selection sheets and a search that understands plain sentences. The target to measure in the pilot: selection in under 10 minutes instead of up to an hour.
03
How we built it
Three applications, the API, the admin panel and the app for designers, share one database, and the API is the only source of rules and validation. Formulas in ERP notation are checked on save, and their results are stored so the search can also filter by calculated values. A language model turns the user's sentence into configuration filters that can be corrected. It does not run database queries, and results come only from product data. AI is available only after login, with a daily cost limit and a query log, and every change goes through a separate test environment before release.
TypeScriptNext.jsSupabaseRailwayOpenRouterClient's PIMClient's ERP
Workflow
How the solution works
- 01
Describe requirements
The user specifies the parameters and conditions for product selection.
- 02
Check the rules
The configurator checks the variant against structured rules and identifies constraints.
- 03
Expert verification
Product experts review the supported scope. The result becomes an input to further quoting.
The human role
Experts approve the selection model, rules and supported product range. Each extension is verified against agreed examples.
Lessons
What made the difference.
- 01
A sentence instead of a form
The model turns a description of the need into configuration filters the designer can see and correct, and results always come from the product database.
- 02
ERP formulas without retyping
The import reads formulas from spreadsheets and the ERP library, checks variables and dependencies on save, and shows errors in the panel.
- 03
Data from catalogue sheets
In the PoC phase an AI model read performance tables from PDF data sheets, analysing the text together with the page image, and saved them to the database.
Product dataAI searchPDF extraction
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
Product parameters, formulas and dependencies from more than 30 spreadsheets and the ERP in one database, synchronised with the PIM.
- Skills
Selection logic stored as records: the form, validation and calculations come from data, not from code.
- MCP
An API as the single source of rules and validation for three applications.
Value
Where value is created
Time saved
Less manual comparison of spreadsheets and parameters.
New possibilities
Consistent rules make product expertise accessible to more people.
What comes next
Further steps expand the verified rules and product set, with each addition checked against 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
