Lumari

Most procurement software — Coupa, Ariba and the long tail behind them — was built for indirect spend: laptops, SaaS seats, office supplies, where the catalogue is stable and a purchase is a transaction. Lumari's whole positioning is that direct materials (the parts that go into the product you build) are a different problem, and its own FAQ makes the argument plainly: quotes arrive in incompatible formats, lead times move constantly, and one missed supplier update cascades into production (https://lumari.ai/).

What the product claims to do, per Lumari's own structured feature list: distribute RFQs by email with natural-language SOPs; extract and normalise quote data from PDFs, spreadsheets, images and emails; parse bills of materials and map line items across suppliers; track POs from issuance to goods receipt with delivery-risk flagging; capture compliance data (country of origin, HS codes, ROHS, conflict minerals); compute landed cost including freight and tariffs; and produce supplier scorecards and cycle-time analytics (https://lumari.ai/, JSON-LD featureList).

That is a coherent and fairly narrow scope, and the narrowness is the point. This is an execution layer for the chase-and-normalise work, not a source-to-pay suite. Lumari says as much itself in a comparison post positioning against SourceDay, an established PO-collaboration vendor (https://lumari.ai/blog/lumari-vs-sourceday). If you are shopping for contract management, supplier risk scoring or AP automation, this is not that product.

Two case studies are published; two more slots say 'Coming soon' (https://lumari.ai/case-studies). Both published ones are more substantial than the genre usually allows, and this is the strongest thing on the site.

OFS — a furniture manufacturer founded in 1937, based in Huntingburg, Indiana (https://lumari.ai/case-studies/ofs; company confirmed at https://ofs.com/). The write-up states a specific before baseline, which is what makes an after number meaningful: 20–25 hours a week on manual sourcing work, and 1–2 days to prepare a single RFQ. Claimed results: 10× faster RFQ processing, 2× faster landed-cost comparison, two-week implementation, 100% supplier adoption. The quote is attributed to 'Donn M., Sourcing Manager' — first name and initial only: 'Without Lumari, it would have taken three people to do what one person was able to do by himself.'

Carbon Robotics — the agricultural-robotics company behind the LaserWeeder (https://lumari.ai/case-studies/carbon-robotics; https://carbonrobotics.com/). Claimed results: 100% of open POs carrying trusted delivery dates, under two weeks to implement, 24/7 open-order risk monitoring, 100% supplier adoption. The quote is fully attributed to Andy Booth, Director of Procurement. We independently corroborated that a person of that name holds a matching role — Andrew Booth, Director of Strategic Procurement at Carbon Robotics — via a third-party contact database with no relationship to Lumari (https://www.datanyze.com/people/Andrew-Booth/3017394395).

How much weight to put on this: a named executive at a company you can look up is meaningfully better evidence than an anonymous logo wall, and it is fair to credit Lumari for it. But these remain vendor-authored case studies with vendor-computed metrics. '10× faster RFQ processing' has no stated methodology, and neither customer has published anything about Lumari under their own name that we could find. Two references also means two — a sample from which you cannot infer a reliability record.

Separately, the case-studies page carries a rotating wall of 12 customer quotes attributed only by job title ('Purchasing Manager'), with no company. Those are not evidence of anything and should not move your assessment in either direction.

The central bet is that suppliers never have to log into anything. Agents send RFQs and follow-ups from email, and parse whatever comes back — PDF, spreadsheet, image, or prose in the body. Lumari's headline metric is '100% supplier adoption. No portal required' (value confirmed in the page's JavaScript bundle at https://lumari.ai/_astro/HomePage.CWp4bZes.js, not just the marketing copy).

This is a genuinely sensible design choice for the problem. Supplier-portal adoption is the perennial failure mode of procurement software: your leverage over a 40-person machine shop to make it learn your new system is close to zero. Routing around that constraint instead of fighting it is the strongest idea in the product, and it is consistent with the CASA Tier 2 assessment discussed below — that is specifically what Google requires of applications that read Gmail data, so the certification corroborates that deep mailbox access is core rather than incidental.

One nuance worth having straight, stated neutrally: a Lumari Supplier Portal does exist. The privacy policy references it repeatedly and describes 'Information submitted by suppliers in response to procurement requests' flowing through it, and there is a live application at app.lumari.io (https://lumari.ai/privacy). These claims are reconcilable — the marketing claim is that suppliers are not required to adopt a portal, not that none exists — but a buyer should ask in the demo which flows actually run over email versus which quietly route suppliers to a portal, because the distinction is the entire value proposition.

Lumari's structured data names ERP integrations with SAP, NetSuite, Oracle, JDE, Vista and Procore (https://lumari.ai/). The homepage's scrolling logo grid shows a partly different set of 26 marks — including SAP, Oracle, NetSuite, Microsoft Dynamics 365, Infor, Jaggaer, Arena, Windchill, Snowflake, Smartsheet, Gmail, Outlook, Teams, Slack, WhatsApp and AWS. Notably, JDE, Vista and Procore appear in the machine-readable feature list but not among the logos, and several logos (Snowflake, Jaggaer, Arena, Windchill) do not appear in the feature list.

We could not establish what any of these integrations actually do. There is no public product documentation, no API reference, no integration guide and no changelog anywhere on the site: the full sitemap is 57 URLs, of which 42 are blog posts and the remainder are marketing, legal, careers and case-study pages (https://lumari.ai/sitemap.xml). Nothing distinguishes a certified bidirectional ERP connector from a read-only export or a roadmap item.

For a product whose pitch is 'your ERP stays the source of truth, just more accurate' — a direct quote from its FAQ — the depth and direction of those ERP writes is the technical question that decides whether it works for you. Ask for the specifics against your own ERP version, and ask which of the six named systems have live production customers today.

The security page is better than average for a five-person company, and the certifications claimed are specific rather than decorative (https://lumari.ai/security):

• SOC 2 Type I and Type II — 'independently audited annually by a third party firm', report available on request to [email protected]. We have not seen the report; the claim is Lumari's.
• CASA Tier 2 — assessed under Google's Cloud Application Security Assessment by an authorised lab. This one is worth understanding, because it is the most informative badge on the page: CASA is what Google requires of third-party applications requesting restricted Gmail scopes. Its presence is consistent with, and partially corroborative of, the claim that Lumari reads production mailboxes at depth.
• Independent third-party penetration testing, continuous vulnerability scanning, vetted subprocessors (list on request), and role-based access with deletion controls.

The gap a buyer should close before signing: the privacy policy names no AI or LLM subprocessor, and contains no statement about whether customer data is used to train models (https://lumari.ai/privacy). The subprocessor list is available on request rather than published. For a system whose function is to read every supplier email in your procurement inbox — pricing, MOQs, drawings, contract terms — knowing which model provider sees that traffic, under what data-processing terms, and whether it is retained for training is a first-order question. This is an omission to resolve on a call, not evidence of anything wrong, but do not sign without an answer in writing.

One structural note in the product's favour: the privacy policy explicitly distinguishes data Lumari processes on behalf of business customers (governed by the customer agreement) from data it controls directly — the correct construction, and not one every startup at this stage bothers to make.

Lumari, Inc. is a San Francisco company founded in 2025 (https://lumari.ai/). Co-founders are Sam Lamba (CEO) and Eshani Mehta (COO). It went through Y Combinator's Spring 2025 (X25) batch and its YC directory profile lists 5 employees and 3 open roles — a count that matches the three postings on its own careers page (https://www.ycombinator.com/companies/lumari; https://lumari.ai/careers). YC's profile states the founders are 'engineers from Google, Tesla, Amazon, and Stripe'; Lumari's own about page says 'Google, Amazon, and Robinhood'. We did not independently verify any individual's employment history.

On funding, be careful with what you read elsewhere. Lumari's YC profile describes it as 'a well-funded startup', with no figure. We could not verify any amount from a primary source: there is no SEC filing of any kind under the name Lumari — EDGAR full-text search returns zero hits across all form types, and EDGAR's company search returns no matching registrant. Crunchbase and its funding-round page both return HTTP 403 to automated fetches. Third-party aggregators surfaced a $500K figure, but per this site's standing policy we do not publish aggregator funding numbers, which routinely disagree by an order of magnitude. Treat the total raised as not established. YC backing itself is confirmed by YC's own directory.

What this means practically: five people, one year old, two public references, and an unverifiable balance sheet. That is a normal and unremarkable profile for a YC company at this stage — it is not a criticism. But if Lumari becomes the system routing your supplier communications, continuity risk is real and should be priced in: ask about data export, what happens to in-flight RFQs if the service stops, and whether your ERP retains a complete record independent of Lumari.

There is no pricing page — https://lumari.ai/pricing returns 404, and no price, tier, seat rate or usage unit appears anywhere on the site. Every call to action is 'Book a Demo'. Nothing indicates a free trial, a self-serve tier or a published starting price.

This is normal for enterprise procurement software and not a mark against the company, but it has a practical consequence: you cannot size this against alternatives without entering a sales process, and you have no anchor going in. Since the product's own framing is agent-based automation over high email and PO volume, the questions to ask early are what the unit of pricing actually is — seats, POs, RFQs, suppliers, or agent actions — and how the bill behaves at 3× your current volume. A per-action model and a per-seat model produce very different numbers for a small team processing a lot of orders, which is precisely the profile Lumari says it serves.

Three numbers appear prominently and carry no source or methodology. They are the vendor's, and are reproduced here only so a buyer recognises them as claims rather than findings:

• '80% reduction in manual touchpoints' — a homepage headline figure with no case study, sample or definition behind it. (Note: this value is rendered by JavaScript as an animated count-up; the static page shows '0%'. Confirmed in https://lumari.ai/_astro/HomePage.CWp4bZes.js.)
• 'the world's first AI-powered procurement automation platform purpose-built for direct materials' and the site's 'first AI supply chain platform for direct procurement' — a primacy claim we made no attempt to adjudicate and would not repeat. The category has several well-funded entrants.
• 'deploys 100s of always-on AI agents' (YC profile) — 'agents' here is a marketing unit, not a measurable one.

The case-study figures in the customer section above are also vendor-computed, but they are attached to named companies and stated baselines, which puts them in a different and considerably more credible class than these.

Worth a demo if: you manufacture or assemble a physical product; your buyers genuinely lose 15+ hours a week to RFQ assembly, quote normalisation and PO chasing; your suppliers are small enough or numerous enough that a portal mandate has already failed or is obviously hopeless; you run one of the named ERPs and want a layer over it rather than a migration; and you can tolerate a five-person vendor. The OFS profile — a mid-size manufacturer with a small sourcing team drowning in spreadsheets — is the shape Lumari has demonstrated on, and it is a real shape.

Probably not yet if: your spend is mostly indirect (this is the wrong category of tool, by the vendor's own argument); you need published pricing to get budget approval; procurement policy requires a documented API, a public trust centre or a named-subprocessor list before an integration review; or your organisation requires multiple referenceable customers at your scale, since there are currently two.

Independent signals available today: essentially none. We found no G2 or Capterra listing, no user reviews, and no independent editorial coverage — the one launch write-up we located sits on a startup-accounting vendor's blog and reads as syndicated announcement copy rather than reporting (https://fondo.com/blog/lumari-launches). For a company founded in 2025 this is completely normal and is not a negative signal; it does mean the only evidence available is Lumari's own, so weight the reference calls accordingly and ask to speak to Carbon Robotics or OFS directly. The site itself is actively maintained — 42 blog posts with sitemap entries updated as recently as 6 August 2026 (https://lumari.ai/sitemap.xml).