Glimpse (YC Summer 2020): teardown vs XMarts

In one line: Glimpse runs AI agents that find and dispute wrong retailer deductions for consumer goods (CPG) brands, and puts the recovered cash back in the brand's books [site] [yc]. Threat: 3, because it is the clearest funded example of our W7 shape: agents do a finance job, are checked by humans, and are measured in dollars, with a named customer list and $52M raised [site] [press]. The catch: it covers one leak (retailer deductions) in one industry. If W7 stays wide ("any margin leak"), treat it as a 2.

1. Fact file

Item	Finding	Source
What they say	"We win back cash for CPG brands." YC line: "the system of action for CPG retail operations".	[site] [yc]
Founded, team, funding	YC S20. 52 people, New York. About $52M raised in total: a $10M round led by 8VC (Apr 2025) and a $35M Series A led by a16z (Mar 2026). Sources disagree on how the 2025 round is labelled (Series A or seed).	[yc] [press]
Founders	Akash Raju (CEO), Anuj Mehta, Kushal Negi. The YC page lists two founders (CEO, CPO); the press names three, with Negi as CTO or similar. They met at Purdue. What YC saw in 2020: a young technical team that could ship fast (inference).	[yc] [press]
Customers	They claim 200+ consumer brands. Named: Suave, Chapstick, Bero, Lemon Perfect, IQBAR, Evermark, Belcam Beauty, Calyan Wax, Table 87, Coaqua, Refresh Gum, Miss Jones Baking, immi, Companion Candles. Integrations: Walmart, UNFI, KeHE, Sam's Club, Amazon, Walgreens, Kroger.	[site] [press]
Price	Not public. YC says pricing is "simple and incentive-aligned" (so tied to money recovered, inference). Small brands ($1M to $30M retail sales): first 3 months free if nothing is recovered, for contracts signed 15 Oct to 31 Dec 2026. Big brands ($100M+): "$10K if no money is found" guarantee.	[yc] [site]
Path	2020: first product was about Airbnb product placements. Hard pivot to retail deductions by 2024 because it "lacked product-market fit" (CEO). Launch Apr 2024. Seed-size round Apr 2025. $35M Mar 2026.	[press]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Finance and FP&A people at CPG brands. Customer stories run from a 2-person team (Table 87) to a "$1B CPG" in the YC text. Most named brands look small to mid-size (inference) [site] [yc].
The pain	"Retailer deductions are draining up to 15% of your bottom line." Retailers short-pay invoices, most go unchecked, and the dispute window closes [site]. A $1B brand's FTE would need about 2 years to review 17,000 deductions (they claim) [yc].
The promise	Find the wrong deductions, win the money back, with little work from the brand (under 1 hour a week for most customers, they claim) [site].
How it works	Connect retailer portals and the ERP once. Agents pull the invoices, proofs of delivery and remittance files, check each deduction against the brand's own records (supply chain, promo calendar), and file a dispute before the window closes. Human experts review edge cases. Recovered cash is applied and synced to the ERP [site] [press].
The output	Recovered cash, a central record of all deductions, and "personalized insights" from invoice data [site].
Proof style	Dollars recovered per customer story ($137K, $145K, $166K, "$2M+") and ROI multiples (3 to 4x on the site, 4 to 5x in press, up to 33x), all self-reported [site] [press].
Who sells it	Own sales ("Book a demo") [site]. A "Deployment Strategist" role at $160K to $220K suggests hands-on onboarding (inference) [jobs].
Price model	Not listed. Results-linked, with a free-trial-if-nothing-recovered guarantee [site] [yc].
The story	Their enemy: the retailer deduction nobody has time to fight. Their belief: every processed deduction makes the system better, "a compounding data advantage" (CEO) [press].

3. Positioning

Category they claim: AI deductions management and revenue recovery for CPG [site]. Also "system of action" [yc].
Category a YC partner would put them in: vertical AI for finance operations, paid on the result (inference).
What they want to be known for: the place a consumer brand sends its deductions and gets money back.
Where they sit on the map: layers L1/L2/L3 = 1 / 2 / 2, buyer = sme (inference), business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (from the system inventory, "Marton's logs show")
L1 Company brain	1. One shared record of deductions, checked against the brand's own supply chain and promo data [site] [press]. The "compounding data advantage" is a claim in press, and the product page describes no learning loop [press] [site].	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning. Learns from calls and checked work.
L2 Agent orchestration	2. Several agent steps (retrieve, validate, dispute, recover, apply cash) with human experts on edge cases. A "Shortage Agent" is named in a customer story [site]. No public detail on customer approvals before disputes go out [site].	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning	2. Does the work and counts the money (dollars recovered, win rates they claim) [site] [press]. Measured results changing the next run is claimed, not shown (so not 3) [press].	Every failure gets a record and an automatic check (about 160). Offer feedback lines fell from 44 to 18 as rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

5. Same and different

Topic	Same or different	Them	Us
Belief	Same	Agents should do the finance work and be measured in cash [site].	Same (brief: "do the work and own the number").
Kind of work	Same	Agents finish the work, with humans on edge cases [site].	Marton's logs show agents do and check work, with a different AI as checker.
Owner of the result	Same	The recovered dollars [site].	W7 aims at one number that moves (brief).
Buyer	Different	CPG brands, selling into US retailers [site].	SMEs and scale-ups, UK mid-sized firms (brief). No customer yet.
Scope	Different	One leak: retailer deductions, plus cash application [site].	W7 is any margin leak: pricing or demand, then fix (brief). Not built for a customer.
How it captures	Different	Retailer portals and ERP logins, once [site].	Calls, documents and approved fixes (brief).
Proof	Different	Dozens of named stories with dollar results (they claim) [site].	No customer number yet.
Delivery	Different	Software plus human dispute experts and onboarding staff [site] [jobs].	The system plus the two founders.
Speed and price	Different	About 30 days to first dispute, 45 days to first resolution (they claim); results-linked [site].	Not set yet.

6. Wedge overlap

Wedge	Fit	Why
W7 Find and fix margin leaks for commercial or finance leaders at SMEs and scale-ups	direct	Same pain (money leaking from sales), same kind of buyer (finance lead at a small or mid-size firm), same loop (find, act, measure) [site]. They own one leak in one industry; W7 is wider (brief).
W5 One layer for one type of small firm	adjacent	They are one product for one type of company (CPG brand) with agents on a back-office job [site]. Our W5 is only a guess (brief).

7. Whitespace (where they stop)

Layer gap: the product page shows no learning loop or customer approval step [site]. The loop is a press claim [press]. To own it we need a rule-change log tied to checked runs on a customer's workflow, with rework falling.
Buyer gap: only consumer goods brands selling into US retailers [site]. UK is unknown. To own it we need UK mid-sized firms outside CPG saying they have a margin leak and would run a pilot.
Number gap: not a gap. They publish dollars per customer [site].
Price gap: not a gap. Results-linked with a free trial if nothing is recovered [site].
Speed gap: not a gap. First dispute in about a week of portal access, or 30 days on average (they claim) [site]. We would need to beat that, and we have no customer run to prove it.

8. For the YC application

Proven today: Marton's logs show his system already runs agents that do and check work: about 5,600 jobs, each checked by a different AI, with about 160 failure records that each added an automatic check. Glimpse shows the shape works for one finance job in one industry [site]. We have no customer result yet.
Draft answer to "how are you different from Glimpse?": DRAFT. "Glimpse proves agents can win back cash on one leak for consumer goods brands. Our system already runs agents that do and check work every day on our own business, and we are picking the first leak with real buyers." Founders say it their own way.
Future claim (needs proof): "We find which margin leak is pricing and which is demand at a UK mid-sized firm, run the approved fix and show the before/after number." Needs one customer workflow run for 4+ weeks with a before/after number.
What we understand that they don't: the money in a firm leaks in more than one place, and the cause (pricing or demand) sits upstream of the symptom. needs proof (inference: they do show "personalized insights" [site], so do not say they ignore causes).
Copyability (M3): medium. They already sell revenue recovery and cash application beyond deductions [site], have $35M new money and are hiring engineers [press] [yc], so more finance workflows inside CPG are an easy step. A move to UK mid-sized firms across pricing and demand would pull them off their retailer-portal base (inference). Not hard, so M3 can not score above 3 against Glimpse alone.
Why now (M4): None found in this teardown. Do not use "agents can now do the work": that is general AI progress.
Do not claim: "agents that recover revenue" as new (Glimpse and Rulebase already do it). "Faster than Glimpse" (no customer run). "We learn from checked results on customer money" (only on our own work today). "Nobody does margin leaks" (UNVERIFIED, we only checked one rival).

9. Ideas worth taking

Start with a look-back over past data. One year of old deductions gives a big number in days (they claim 17,000 reviewed in under 24 hours) [yc].
Price on the result with a plain guarantee ("find your money or $10K") [site]. It removes the buyer's risk.
Name each agent by the job it does ("Shortage Agent") and publish one dollar number per customer [site].

10. Unknowns

Real prices and the share of recovered cash they keep. Ask a buyer who has seen a quote.
Whether they sell in the UK or Europe. Check their careers page and retailer list for UK retailers.
Whether the learning loop is real. Look for a technical blog or interview that shows rule changes from dispute outcomes.
How much of the dispute work is done by people. A third-party job listing for a "deduction analyst" suggests a lot [jobs]. Look at the Ashby careers page job texts (the page did not load fully).
Which of the stats is current: the site shows 2.5% topline lift on the home page and 30% on the product page, and ROI of 3 to 4x against 4 to 5x in press. Treat all as UNVERIFIED claims.

Sources

tryglimpse.com home [site]
Product page [site]
Customer stories [site]
Resources and blog, newest post Sep 2026 [blog]
YC company page [yc]
YC jobs page [jobs]
TechCrunch: Glimpse raises $35M led by a16z, 25 Mar 2026 [press]
Glimpse Secures $10M (SalesTechStar), Apr 2025 [press]
Deduction analyst job at Glimpse (third-party listing) [jobs]

Glimpse

What they do

How close to us

Wedges

Who buys

Full teardown