Agentic AI System for Construction Industry Tender Management and Quote Generation

tendos_ai · Other · 2026 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.2 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Construction tender management

Tendos AI needed to automate manufacturer tender workflows involving emails, attachments, product matching, and quote generation.

Constraints:

Unstructured customer emails
Large attachments
Complex product catalogs
Human review before quotes

Architecture

Tendos AI automates construction manufacturer back-office workflows from inbox to human-reviewed offer. The system classifies incoming emails as support, offer, or order requests; extracts project, product, revenue, planner, partner, submission-date, and attachment details; narrows product search space; semantically matches descriptions against product portfolios; uses a planning pattern that can revise the workflow when categorization or matching is wrong; creates an offer/proposal for human approval; and maintains customer-specific evaluation sets to monitor performance as products and use cases expand.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  customer-email-inbox[Customer email inbox] --> intent-classifier[Intent classifier]
  intent-classifier --> entity-extraction[Entity extraction]
  entity-extraction --> product-search-planner[Product search planner]
  product-search-planner --> product-portfolio-matcher[Product portfolio matcher]
  product-portfolio-matcher --> offer-generator[Offer generator]
  offer-generator --> human-reviewer[Human reviewer]
  evaluation-sets[Evaluation sets] --> product-search-planner

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Customer email inbox`	Source of support, offer, order, and tender requests with attachments.	`Intent classifier`
`Intent classifier`	Determines whether the email is support, offer, order, manual, replacement, or renovation-related.	`Customer email inbox`, `Entity extraction`
`Entity extraction`	Extracts products, revenue estimate, due date, planner, partners, and project context.	`Intent classifier`, `Product search planner`
`Product search planner`	Plans and revises product-search workflow based on context and rules.	`Entity extraction`, `Product portfolio matcher`, `Evaluation sets`
`Product portfolio matcher`	Semantically matches customer requirements to manufacturer products.	`Product search planner`, `Offer generator`
`Offer generator`	Creates proposal, support, or quote drafts for review.	`Product portfolio matcher`, `Human reviewer`
`Human reviewer`	Approves or rejects the generated response before sending.	`Offer generator`
`Evaluation sets`	Customer-specific tests for product matching and workflow performance.	`Product search planner`

Data flow

An email with attachments enters the workflow. Tendos classifies intent, extracts key tender/request data, decides which products or categories matter, searches and semantically matches the manufacturer's product portfolio, updates the plan when findings contradict the initial route, generates an offer or support response, and presents it to a human for approval before anything is sent.

3 supporting quotes

“Like we are categorizing it, we're prioritizing it. We are already putting it into the right projects and assigning it to the CRM.”transcript, stated

“This is all data that we extract. So we can present it to our customers because normally they would do that by hand and then put it into their CRM just to keep to have some sort of monitoring and to be accountable for AI actually did that and we know what we did”transcript, stated

“So we use basically a planning pattern where we look at the context that we provide and then based on the rules that we've discovered try to come up with a plan on how to proceed with this and based on the findings that that are in the workflow.”transcript, stated

Why this architecture

Tendos uses a planning and review architecture because construction tender inputs are semantic, inconsistent, and often buried in PDFs and product catalogs built for humans. Starting narrow let the team learn the workflow, while evaluation sets and human approval manage the risk of wrong product matches as the system expands across categories.

3 supporting quotes

“your input is semantics and it's not it's not structured and the product data as well has some structure to it but most of the critical information because it should be readable for humans is semantics it's very hard for computers to understand before the age of LLMs basically.”transcript, stated

“position quite well for small documents. And all of a sudden our users started to experiment and the documents grew larger and larger to the point that we said okay now either now we need to invest more into entity extraction because all of a sudden the problem is not on the side of do we understand”transcript, stated

“the user has to say yeah submit that's the right answer please send it out so that's the current situation that we can support this workflow what is something we're looking into currently is more the technical planning part where it get really about calculations”transcript, stated

Models used: LLMs

Scale

The product expanded beyond radiators into multiple construction segments, while infrastructure use cases remained out of scope.
“So we are not only supporting like the you said radiators but we are supporting everything when you look around”transcript, stated

Alternatives considered (roads not taken)

Start with all construction product categories

Rejected because: Tendos began with a narrow radiator/design-partner workflow and expanded as it learned.

“We looked at a very specific subp part of a portfolio of one of the med of the design partner that Daniel mentioned and it was radiators but a special kind of radiators and we could say okay this is they have a specific department for for these.”transcript, stated

Fully automate outbound offers

Rejected because: Users still approve generated proposals before they are sent.

“as a user, you have to approve the proposal that Tendos has created for you in any case. So let it be an offer. But also for support, we don't send out an email. the user has to say yeah submit that's the right answer please send it out so that's the current situation”transcript, stated

Patterns

`multi-agent-orchestration` (tender-workflow)

The tender pipeline has multiple specialized steps across categorization, extraction, matching, and quoting.

“enterprises are open to rethink their whole workflows before in the SAS world it was always the case give me the”transcript, implied

`human-in-the-loop-gate` (quote-review)

Generated quotes are reviewed by a human before delivery.

“sounds like it is generating an offer, but then it's reviewed by a human before it goes back to the customer. And also”transcript, stated

Tradeoffs

How to expand product coverage safely.

Chose start narrow with radiators and a design partner over broad initial product coverage.

Rationale: The summary says they started with a narrow focus and iteratively expanded.

“from complex catalogs using semantic understanding, and generates detailed quotes for human review. Starting with a narrow focus on radiators with a single design partner, they iteratively expanded to support full workflows across multiple product categories, employing sophisticated agentic architec”summary, stated

How to expand workflow coverage.

Chose narrow category learning plus evaluation sets over broad construction automation from day one.

Rationale: The team could iterate quickly on a focused workflow and measure changes before adding more products and use cases.

“change small parts of it, run small evaluations on it, see if it performs on par or better”transcript, stated

Gotchas (what bit them)

Tender automation became harder as users moved from small scoped documents to large construction-site documents.

Root cause: The problem shifted from extracting entities from short requests to understanding large bodies of PDF and project data.

Resolution: Tendos staged expansion by product group and invested in entity extraction, product matching, and evaluation before broadening coverage.

“document or entity extraction from a document is quite hard but it's manageable if you have smaller documents and it really gets hard if you go into the hundreds of pages.”transcript, stated

“And all of a sudden our users started to experiment and the documents grew larger and larger to the point that we said okay now either now we need to invest more into entity extraction”transcript, stated

Evaluation

The company uses extensive evaluation frameworks at each pipeline step, according to the summary.

Metrics: pipeline-step evaluations

“ategories, employing sophisticated agentic architectures with planning patterns, review agents, and extensive evaluation frameworks at each pipeline step.”summary, stated

Outcomes

The product expanded from one design partner/radiator focus to broader construction workflows.

No hard production metric is provided in the summary.

“otes for human review. Starting with a narrow focus on radiators with a single design partner, they iteratively expanded to support full workflows across multiple product categories, employing sophisticated agentic architectures with planning patterns, review agents, and extensive evaluation framewo”summary, stated