Identify
Understand exactly where quote comparison broke down and why manual spreadsheets couldn't keep up
The situation
Mahindra sources raw materials, machined parts, and packaging from a supplier base of several hundred vendors across categories — metals, electronics, plastics, and specialty components — to keep multiple production plants running. Getting a request for quotation from a raw requirement to an awarded vendor is what keeps parts flowing to the line on schedule.
On the procurement floor, the team had already lived with:
- Manual, email-based RFQs with no standardised format
- Vendor quotes re-typed by hand into comparison spreadsheets
- No formal record of why an award was made
- A supplier base ranging from large certified manufacturers to small shops that quote entirely by email
- Sourcing decisions that took days to close, discovered as bottlenecks only after the fact
As RFQ volume grew alongside new product lines, the cost of that process — transcription errors, undocumented awards, no spend visibility — kept accumulating. The vision behind fixing it properly, though, was still strong.
The challenge
The challenge wasn't just comparing quotes faster — it was making every RFQ, from every vendor, in whatever format it arrived, land in one trustworthy comparison with a documented reason behind every award, without forcing every supplier onto a single portal.
Our approach
We were brought in to design a system that would work the way procurement actually sources parts — not the way a typical enterprise workflow assumes every vendor behaves.
Our objective was to:
Design
Build an AI-driven RFQ platform that fits how procurement already sources parts today
Gate
Make it structurally impossible for an award to go through without a documented approval
Scale
Support every vendor format and give leadership full spend and supplier visibility
What we did
1
Built AI-powered RFQ intake and generation
Procurement uploads or forwards a raw sourcing requirement — a spec sheet, drawing, or free-text request — and an LLM pipeline extracts item specifications, quantities, target dates, and required certifications to draft a structured RFQ automatically.
Built to work with how requirements actually arrive:
- Reads spec sheets, drawings, and free-text requests directly
- Suggests a vendor shortlist based on category history and past performance
- Dispatches the RFQ automatically once reviewed
The result: an RFQ goes out in minutes instead of the hours it used to take to assemble by hand.
2
Built two intake paths so no vendor is excluded
Onboarded vendors quote through a structured supplier portal. Everyone else can keep replying by email or PDF exactly as before — an AI parsing pipeline extracts price, currency, lead time, and terms and normalises them into the same schema.
Two ways to quote, one comparable record:
- Supplier portal — structured form for price, lead time, and MOQ
- Email / PDF path — AI extracts and normalises the same fields
- Both paths write to the same structured comparison record
3
Added a mandatory, configurable approval gate
No award can be confirmed without routing through approval based on configurable thresholds — spend amount, category, or deviation from the lowest bid. If it falls outside the rule, it's blocked from auto-approval outright.
What the gate covers:
- Spend, category, and lowest-bid deviation thresholds, fully configurable
- Every approval, rejection, and comment timestamped — no undocumented awards
- Addresses the most common cause of unexplained sourcing decisions
4
Layered in comparison, analytics, and a full audit trail
Every vendor's quote for an RFQ shows up side by side — normalised pricing, lead time, and a rolling quality score — with outlier prices and missing fields flagged automatically before a decision is made.
Supporting details that mattered to procurement:
- Live dashboards for spend by category, RFQ cycle time, and savings realised
- Per-vendor scorecards for pricing trend and on-time delivery
- Every RFQ, quote, approval, and comment logged with timestamp and reviewer identity
Procurement leadership now has a complete, immutable audit trail for every RFQ that has moved through the platform.
How it works
From a raw sourcing requirement to an awarded, fully audited vendor — every stage recorded, one hard gate before any award is confirmed.
1
Requirement submitted & RFQ drafted
An LLM reads the raw requirement and drafts a structured RFQ automatically.
LLM extraction2
Vendor quotes in
Vendors quote through the supplier portal, or keep replying by email or PDF.
Two intake paths3
AI quote parsing & normalisation
Price, currency, lead time, and terms are extracted and mapped into one comparable record.
LLM parsing4
Approval gate
The award routes for approval based on configurable spend and category thresholds.
Hard block until approved5
Audit trail & dashboard
Every RFQ, quote, and approval is logged and visible on live procurement dashboards.
Full spend visibility“
We used to spend most of a week just getting quotes into a format we could actually compare. Now that happens before lunch, and I trust the numbers because nobody re-typed them by hand.
Mahindra
“
The dashboard flags outlier pricing automatically now. Before, we would only find out a quote was inflated after we had already awarded it.
Mahindra
~80%
Faster RFQ turnaround
Days down to hours
22%
Avg. cost savings
Vs. prior quote cycles
100%
Audit trail
Every RFQ, quote, and approval recorded
0
Vendors excluded
Portal or email — every format accepted
The outcome
What changed after rollout:
- AI drafts and dispatches RFQs from raw requirements, with a mandatory approval gate before any award
- Every vendor quote — portal or email — normalised into one comparable, auditable record
- Live dashboards for spend by category, RFQ cycle time, and savings realised
- Per-vendor scorecards surfaced pricing and delivery trends leadership couldn't see before
- Two long-standing supplier contracts renegotiated after performance data came to light
- Procurement decisions made same-day on average, down from days per RFQ
Frequently asked questions (FAQs)
1. What is AI-powered RFQ automation?
It's a platform that uses an LLM to draft, dispatch, and parse request-for-quotation workflows automatically — turning a raw sourcing requirement into a structured RFQ, collecting vendor quotes in any format, and normalising them into one comparable record.
2. Why was manual, email-based RFQ processing unreliable at scale?
Every vendor replied in a different format — a PDF, a spreadsheet, or prices typed into an email — so quotes had to be re-typed by hand into a comparison sheet. That manual step introduced transcription errors and meant sourcing decisions took days to close.
3. How does the platform draft an RFQ automatically?
An LLM reads the raw requirement — a spec sheet, drawing, or free-text request — and extracts item specifications, quantities, target dates, and required certifications to draft a structured RFQ, then suggests a vendor shortlist based on category history.
4. What happens if a vendor doesn't use the supplier portal?
They can keep replying by email or PDF as usual. An AI parsing pipeline extracts price, currency, lead time, and terms from the document and maps them into the same schema as portal submissions, so no vendor is excluded from the comparison.
5. How does AI quote parsing handle messy formats like PDFs and emails?
The LLM-based pipeline is trained on real historical RFQs and vendor quotes across categories, so it can extract pricing, MOQ, and terms reliably even from unstructured, inconsistently formatted documents rather than clean, templated ones.
6. What is the mandatory approval workflow and why does it matter?
Sourcing decisions route for approval automatically based on configurable thresholds — spend amount, category, or deviation from the lowest bid — and every approval, rejection, and comment is timestamped, giving procurement a defensible audit trail for every award.
7. How is spend and supplier performance visibility achieved?
Every RFQ, quote, and award rolls up into live dashboards covering spend by category, RFQ cycle time, and per-vendor scorecards for pricing trend and on-time delivery — visibility that didn't exist when data lived across spreadsheets and inboxes.
8. Can vendors submit quotes without any special software?
Yes. Vendors can keep quoting by email or PDF exactly as before — the supplier portal is an option for a faster, structured submission, not a requirement to participate in an RFQ.
9. How does the system handle a quote with a missing field, like lead time?
Incomplete quotes are flagged automatically and routed back for clarification, so a vendor with a missing MOQ or lead time doesn't silently skew the comparison or get lost in the queue.
10. How does Atomic Loops build AI systems for procurement teams?
By designing around how procurement actually works — two intake paths so no vendor is excluded, configurable approval rules, and a full audit trail — rather than forcing every vendor onto a single portal or generic workflow.
Industry Services
Education
Focus areas - Education
© 2026 London School of DIGITAL BUSINESS.
All Rights Reserved