MHT Experience
Client
AI TRANSFORMATION
MHT Experience
AI proposal system that cut electrical bidding time by 87%.
Construction
Industry
AI Product Matching
Services
Romania
Region
2025-present
Year
The bottleneck was mechanical.
MHT Experience builds electrical installations for commercial and industrial construction, with contracts from tens of thousands to millions of euros. Every euro in a proposal has to be defensible against an engineering peer on the buyer side.
A single commercial proposal took 3 to 4 weeks and up to 26 hours of senior-engineer time, with up to 4,000 manual copy-paste operations per bid. Price lists arrived in inconsistent formats nobody had time to standardise. The business could only produce as many proposals as the engineering team could type out, so senior engineers who should have been doing technical design were reconciling spreadsheets instead.
3 to 4 weeks
Per commercial proposal
26 hours
Senior-engineer time per bid
4,000 ops
Manual copy-paste per bid
Mechanical
The ceiling was typing speed
Four specialised AI agents, one orchestrator, three tiers of human oversight.
A guided digital flow replaces a 20-step manual workflow. An engineer creates a project, uploads the client's requirements in whatever format arrived (Excel, scanned PDF, Word), uploads supplier offers, and reviews only the decisions worth making. The system adapts to the documents, not the other way around.
Four specialised agents work in sequence. An ingestion agent turns any document into structured data via OCR and parsing. A matching agent decides whether a client line and a supplier line describe the same product, using hard technical checks on cable type, voltage class, and cross-section before any soft scoring, then returns a confidence tier. A historical pricing agent queries past proposals so every bid compounds the accuracy of the next. A pricing and margin agent layers material, labour, and indirect costs into a final price per line.
From upload to export, the orchestrator runs four agents in sequence.
01.
Ingestion agent
OCR plus a document parser extract product name, quantity, specs, unit price, currency, and category per line, from Excel, Word, or scanned PDF.
02.
Matching agent
Gemini with a structured-output schema returns text and semantic similarity per pair, weighted 40/60, with hard technical compatibility checks first so incompatible products never collapse together.
03.
Historical pricing and margin
Queries MongoDB for the same product across past proposals, then layers MHT's margin rules at project, category, and line level.
04.
Proposal export
Returns the client's original document filled with MHT's prices, sheet order and column positions preserved, with a branded summary sheet as the first page.
The numbers speak.
87%
Proposal turnaround reduction
<30 min
Manual price entry, from 8 to 16 hours
80%+
Matching automated, under 2% error
€13.2M
In bids handled through the system
Proposal turnaround dropped from 3 to 4 weeks to 2 to 5 working days. Senior engineers were redirected from spreadsheet work back to technical design and client relationships, where their time compounds.
Win more bids in a fraction of the time
We built MHT an AI system that cut their bidding time by 87%. If quoting and proposals eat your team's week, we can find where AI takes the load off.
Starts with a free 30-minute discovery call.
Next case study
A WooCommerce platform for selling training courses by date: a custom block library, a course-date commerce flow through cart and checkout, and a.