Teardown

Ecommerce / Retail — Product-attribute AI · Deep dive

Lily AI

Product-attribute AI for ecommerce — vision-plus-language models that enrich retailer catalogs with hundreds of consumer-facing tags per SKU to lift onsite search, recommendations, SEO and paid-media relevance. Flagship customers include Macy's, Bloomingdale's, ThredUp, Tapestry (Coach, Kate Spade, Stuart Weitzman), J.Crew and — as of July 2026 — Marks & Spencer. ~$57.5M disclosed across a $12.5M Canaan-led Series A (Jan 2020), $25M Canaan-led Series B (Aug 2022) and $20M Fernbrook-led Series B-1 (Mar 2024).

emerging

The question that decides it: Does a dedicated product-attribute AI vendor remain a line item in a tier-1 retailer's budget once (a) Shopify, Google Vertex AI Search Commerce, Algolia, Bloomreach and Constructor each bundle automated attribute extraction into contracts the retailer already pays for, (b) frontier-model API costs fall far enough that a retailer's own data team can stand up a competent enrichment pipeline on GPT, Claude or Gemini for a fraction of a multi-year Lily contract, and (c) the LLM-native entrants (Envive, Velou, Width.ai, Naratix) anchor the market's price expectations below enterprise-SaaS economics? Falsifiable test for winning: by end-2027 Lily discloses ARR crossing $40M, retains Macy's, Bloomingdale's, Tapestry and Marks & Spencer at renewal, lands at least three additional top-25 US or UK retailers, and ships a credible self-serve / mid-market tier that defends against the LLM-era entrants. Falsifiable test for losing: an Algolia, Bloomreach, Shopify or Google-native attribute feature displaces Lily at one top-10 account, Lily raises a down-round or sells for less than disclosed capital invested (~$57.5M), or two of the LLM-native entrants cross Lily's enterprise ARR.

My take

HQ
Mountain View, California
Founded
2015
Ownership
VC-backed private — Canaan Partners, NEA, Conductive Ventures, 3L Capital, Fernbrook Capital Management, Global Brain, Unshackled Ventures
Funding
~$57.5M disclosed across seed (undisclosed), $12.5M Series A (Canaan Partners led, January 2020, with NEA and Conductive Ventures — Venture Capital Journal, 2020), $25M Series B (Canaan led, August 2022, with 3L Capital, NEA and Conductive Ventures — IT Digest / WWD / FashionUnited, August 2022), and $20M Series B-1 (Fernbrook Capital Management led, March 2024, with Global Brain and existing investors Canaan, NEA, Conductive and 3L — Lily AI / GlobeNewswire / CMSWire, March 2024). Note: the launch brief referenced a 'Series B led by Battery Ventures' — Battery is not disclosed as a Lily AI investor in any primary source we could locate; both the Series A and the headline Series B were led by Canaan Partners.
Valuation
Not disclosed. CB Insights, Tracxn and secondary aggregators do not publish a post-money for any Lily AI round (CB Insights; Tracxn, 2026). The $20M Series B-1 structure (an extension rather than a priced Series C) is consistent with a 2022 valuation the company chose not to re-mark in 2024.
Revenue
Not disclosed. GetLatka references a small-single-digit-million ARR estimate (2024), and the company has never confirmed a figure publicly. Lily AI appeared on the Inc. 5000 for 2024, which requires a minimum revenue threshold of ~$2M and three years of growth (Inc. / Lily AI, 2024). Any figure materially above the mid-single-digit millions should be treated as speculative absent company confirmation.
Headcount
~100-200 (LinkedIn band, 2026; Wellfound lists 51-200, 2025). GetLatka tracks the company in a similar range. Headcount is consistent with a post-Series-B-1 scale-up that has not re-expanded materially since the 2022 raise.
Screen
Fast riser — founded 2015, raised >$30M (bucket 3). Also qualifies on bucket 2 if the undisclosed seed is counted ($57.5M disclosed is below $100M so bucket 3 is the better fit).
Published
2026-10-22
Web
www.lily.ai
Elsewhere
LinkedIn · Crunchbase

Founders and leadership

  • Purva Gupta Co-founder & CEO

    Immigrated from India to the US for a Yale School of Management MBA (2013-2015). Prior roles include UNICEF, where she worked on child-protection and public-interest campaigns, and the Clinton Health Access Initiative. In interviews she has described the Lily origin as a street survey in New York asking women why they rejected clothing items they had initially picked up — the gap between the retailer's description (fabric, SKU, size) and the shopper's actual decision criteria (occasion, mood, body-shape fit, aesthetic reference) became the company's founding thesis (Founder Institute profile; CommerceNext; Shoptalk).

  • Sowmiya Chocka Narayanan Co-founder & CTO

    Engineering background spanning BloomReach, Box and SAP — the exact mix (search / content / enterprise SaaS) that shows up in Lily's architecture. Named to Retail TouchPoints 40 Under 40 (2024). Inc. 2024 Female Founders list with Gupta (Inc. / Lily AI, 2024). Represents the company at technical venues including Shoptalk Europe and TechCrunch guest-author columns on retail-AI infrastructure.

Snapshot

Lily AI sells a product-attribute AI engine to large apparel, home, beauty and softlines retailers. The product reads a retailer's catalog imagery and PDP text, runs it through a hybrid computer-vision and language-model pipeline, and emits hundreds of consumer-facing tags per SKU — occasion, style, fit, aesthetic reference, mood, body-shape suitability, pattern, silhouette — that then flow into onsite search, recommendations, personalisation, SEO landing pages and paid-media feeds (Lily AI product pages; Robin Report partner page, 2024). The thesis is that retailers describe products in merchant language ("twill jogger, 7% elastane, model 1348") while shoppers search and buy in consumer language ("comfy weekend pant for petite frames") — and the gap is worth percentage points of conversion. The company has raised ~$57.5M across a $12.5M Canaan-led Series A (January 2020), $25M Canaan-led Series B (August 2022) and $20M Fernbrook Capital Management-led Series B-1 (March 2024), with a customer list that now includes Macy's, Bloomingdale's, ThredUp, Tapestry, J.Crew, ALC, Crate & Barrel and — announced July 2026 — Marks & Spencer (Retail Gazette; Grocery Gazette, July 2026).

Founding story

Gupta came to the US on an MBA to the Yale School of Management in 2013, after a stint at UNICEF and the Clinton Health Access Initiative in India. In retellings on the Danny in the Valley and PCDN podcasts and the Founder Institute profile, she describes running a street survey in New York asking women why they put clothing items back on the rack — the answers were never about SKU, fabric or merchant taxonomy, they were about occasion, body-shape anxiety, aesthetic reference and the item not matching the mental image the shopper had come in with. Narayanan brought the technical story. Her background at BloomReach, Box and SAP covered the three disciplines that Lily now sits between — retail search, enterprise content pipelines, and large-scale SaaS architecture. The two co-founded the company in 2015 and spent 2016-19 iterating on a consumer-facing styling / discovery app before pivoting into the B2B catalog-enrichment product that attracted Canaan's $12.5M Series A in January 2020 (Venture Capital Journal, 2020). The company was named Lily Labs / Lily.ai through the early years; the current Lily AI branding dates to around the 2022 Series B.

How it works

A retailer uploads its catalog — PDP imagery, titles, descriptions, existing taxonomies and merchant-side attributes — into Lily's pipeline, typically via a one-time bulk ingest and then a continuous feed. Lily runs the catalog through a two-stage model stack. First, a vision model trained on retail imagery classifies the garment or product by category, silhouette, print, pattern, closure, neckline, sleeve length and other visual primitives. Second, a language-model layer trained on a proprietary taxonomy of what Lily calls customer-speak maps those visual primitives, plus the merchant text, onto consumer-language tags: "cottagecore", "going-out top", "work-to-weekend", "flattering for pear-shape", "wedding-guest", "coastal grandma". The emitted tags are both the deeper taxonomy Lily writes back to the retailer's PIM and the SEO / paid-media feeds Lily writes to Google Shopping, Meta catalog ads and the retailer's onsite search index. Model updates are a mix of continuous unsupervised fine-tuning on new SKUs and human-in-the-loop review by a labelling team Lily operates internally. The company's technical blog (Infographic: Customer-Speak, Engineered for Scale, 2024-25) frames the architecture as a layered pipeline rather than a single foundation model.

Product and business overview

Lily for Product Attributes is the core enrichment product — the catalog-wide tagging engine. Lily for Search feeds the enriched taxonomy into the retailer's onsite search platform (Algolia, Bloomreach, Constructor, Searchspring, Coveo) and typically prints as recall and precision lift on head and long-tail queries. Lily for Recommendations powers upsell / cross-sell / "complete-the-look" surfaces by matching enriched attributes rather than co-view behaviour alone. Lily for Paid Media and Shopping writes the attribute-rich feed to Google Shopping, Meta catalog ads and Pinterest shopping, where feed quality drives Quality Score and CPC efficiency. Lily Max (introduced 2024-25) packages the above into a bundled SKU priced per catalog volume. The go-to-market is reference-heavy enterprise sales — a six-to-twelve-month pilot into one category, graduated to a full-catalog, multi-year contract.

Business model and pricing

Lily publishes a pricing page but not pricing. The published FAQ ("How much does Lily Max cost?" — lily.ai, 2025) says deals are scoped by catalog size, modules selected (search, recommendations, paid media, SEO), and integration depth. Retail-SaaS comps and GetLatka's historical tracking imply enterprise contracts cluster in the low-six-figure to low-seven-figure annual range per retailer, with mid-market tiers running materially lower. Implementation is high-touch: 60-120 days to backfill a tier-1 retailer's catalog, with ongoing feed updates. The company does not disclose gross retention, net retention or logo churn, which is itself informative in a 2026 enterprise-SaaS market that has normalised disclosing them.

Traction over time

Date Milestone
2015 Founded by Gupta and Narayanan
2016-2019 Consumer app phase; seed and seed-extension rounds (Unshackled Ventures, Canaan participation, Global Founders Capital reportedly)
Jan 2020 $12.5M Series A, Canaan Partners led; NEA, Conductive Ventures (Venture Capital Journal, 2020)
2021-22 Pivot complete to B2B enterprise retail; early wins at Macy's, ThredUp, Bloomingdale's
Aug 2022 $25M Series B, Canaan led; 3L Capital, NEA, Conductive Ventures (FashionUnited; WWD; IT Digest, August 2022)
2023 Tapestry, J.Crew, ALC case studies; PYMNTS "Early Traction as Retailers Seek AI Help" piece (PYMNTS, 2023); NVIDIA Inception program
Mar 2024 $20M Series B-1 extension, Fernbrook Capital Management led; Global Brain + existing investors (GlobeNewswire; Axios Pro Retail Deals, March 2024)
2024 Inc. 5000 debut (ranked by three-year revenue growth); Inc. Female Founders list for Gupta and Narayanan
Jul 2026 Marks & Spencer rollout announced for UK ecommerce product discovery (Retail Gazette; Grocery Gazette; Retail-Systems, July 2026)

The B-1 structure of the March 2024 raise — $20M as an extension rather than a priced new round — is the most informative data point on the series. Companies choose B-1s when the step-up to a priced Series C is harder than existing investors want to re-mark.

Market analysis

Retail ecommerce product data is a vastly underinvested asset. Industry attribute coverage per SKU has historically run 5-15 attributes against what the shopper would want to filter by (50-200+). Width.ai (2026) and Channable (2024-25) frame the global product-data-enrichment opportunity as a function of global ecommerce retail (~$6.5T in 2025, eMarketer / Statista) times a share of the technology-services budget spent on catalog operations; a credible bottom-up TAM is a mid-single-digit-billion-dollar annual market across platform fees, services, and in-house tooling. The structural driver is that generative-AI shopping surfaces (ChatGPT shopping, Perplexity shopping, Google AI Overviews shopping) retrieve on structured attributes, and the retailer that can supply the richer schema wins the citation. Lily's bet is that this makes it strategic infrastructure. The counter is that the same forces make the capability look native to whichever platform the retailer already pays — Shopify, Google, Algolia, Bloomreach — and the catalog-enrichment vendor becomes a redundant seat.

Competitive intel

Syte is the closest direct peer — visual-AI product discovery, >$70M raised through 2021, strong on image search but quiet since 2023. ViSenze attacks from APAC retail and visual-shopping widgets rather than deep catalog enrichment. Clarifai, with >$100M raised and a horizontal vision stack, is the dangerous "build-your-own" baseline a motivated retailer would use. Google Vertex AI Search for Commerce is the strategically most threatening — Google wants retailer product data richer because its own Shopping monetises on relevance, and Lily's own case studies describe using Lily data to feed Google surfaces. Algolia, Bloomreach, Constructor and Lucidworks are the onsite-search layer Lily feeds, each of which has been quietly adding automated attribute extraction to its own pipeline, which risks reducing Lily to a feature inside the search platform. Shopify Magic and Amazon's seller-side listing-generation tools cover the SMB and marketplace tails. Envive, Velou, Width.ai, Naratix and Channable are the LLM-era entrants attacking from below — cheaper and faster to procure, each anchoring the market's price expectations downward.

History and evolution

The 2015-19 consumer app phase, the 2020 Series A pivot, the 2022 Series B at scale, and the 2024 B-1 extension are the four inflections. Beyond that: the 2023 pivot in product narrative from "AI stylist" to "the catalog layer under generative discovery" — a repositioning that reads partly as a response to ChatGPT shopping and Google's AI Overviews, and partly as an attempt to recast point-SaaS attribute tagging as strategic infrastructure. The July 2026 Marks & Spencer announcement is the first disclosed tier-1 UK logo and reads as a successful European expansion toehold. We have found no disclosed layoff, down-round, acquisition or executive departure through October 2026.

What people say

The case for. Public case studies are the strongest part of Lily's story: ThredUP reports a 15% sell-through lift attributed to enriched product data (Lily AI case study; ThredUP, 2023); Tapestry documents incremental Google Search traffic and sales from enriched feeds (Lily AI case study; Tapestry, 2024); Marks & Spencer named Lily as a core AI partner in a July 2026 product-discovery launch (Retail Gazette; Grocery Gazette; Retail-Systems, July 2026). Trade press coverage (WWD, FashionUnited, PYMNTS, Robin Report, CMSWire) has been consistently positive on the specific wedge. Inc. and Retail TouchPoints have profiled Gupta and Narayanan on multiple founder lists (Inc.; Retail TouchPoints, 2024). Unshackled Ventures' involvement is a credible immigrant-founder bet from a seed fund that specialises in them.

The complaints. G2 and Capterra listings exist but have essentially no review volume on the Lily AI product (G2; Capterra, 2026) — which, for an eleven-year-old enterprise-SaaS company, is unusual and informative. Glassdoor reviews cluster in the typical range for a 100-200-person startup, with recurring themes around sales-cycle length and the mismatch between the AI narrative and the human-in-the-loop labelling work the pipeline actually requires. The B-1 extension structure in March 2024 (rather than a priced Series C) is read in the Axios Pro Retail Deals coverage (March 2024) as the usual signal: existing investors willing to put more in, not willing to re-mark. And the ambient skepticism — that LLM-era entrants and native platform tooling will compress a point enrichment vendor into a feature — is widely held among the Shopify-adjacent builder community and visible in Width.ai's 2026 competitive write-ups.

Outlook: the open question

Lily wins if, by end-2027, it discloses ARR crossing roughly $40M, retains Macy's, Bloomingdale's, Tapestry and Marks & Spencer at renewal, lands at least three additional top-25 US or UK retailers, and ships a defensible self-serve mid-market tier. It loses if an Algolia, Bloomreach, Shopify or Google-native attribute feature displaces Lily at one top-10 account, Lily raises a down-round or sells for less than disclosed capital invested (~$57.5M), or two of the LLM-native entrants cross Lily's enterprise ARR. The underlying question is whether a specialist product-data-enrichment vendor is a durable company or a feature. The 2015-23 answer was clearly a company — Lily built the taxonomy IP, the labelled datasets, and the enterprise references no horizontal platform had. The 2024-26 answer is contested — Shopify, Google, Algolia, Bloomreach and Shopify's SMB cohort are each adding native attribute extraction, LLM-era entrants are compressing price expectations, and frontier-model APIs are cheap enough that a motivated retailer data team can stand up a competent in-house replica. Lily's bet is that its labelled taxonomy, retailer-specific tuning and enterprise GTM compound into a moat before the bundle arrives at the same answer from the other side.

How to attack it

Four wedges are credible. LLM-native self-hostable catalog enrichment: a startup shipping a thin, API-first attribute-enrichment service that runs on the retailer's own cloud, calls GPT, Claude or Gemini for the heavy lifting, and prices by catalog volume at a fraction of Lily's enterprise contract. Envive, Velou, Width.ai and Naratix are each prototyping this; none has the brand yet. The retailer's data team is the buyer, not the merchant team, which inverts Lily's GTM. Agent-native product data: an attribute pipeline specifically engineered for retrieval by ChatGPT shopping, Perplexity shopping, Google AI Overviews shopping and Amazon Rufus — a startup that treats generative-engine discoverability as its first-class use case rather than as a bolt-on onto onsite-search optimization. Vertical catalog AI: Lily's wedge is widest in apparel, home and beauty; health and wellness (iHerb, supplements), industrial MRO (Grainger, Fastenal), auto parts (Advance Auto, O'Reilly) and specialty hardware are large catalog verticals Lily does not cover deeply, with their own taxonomy complexity and higher willingness to pay. Catalog-feed-as-a-service bundled with paid media: a startup that gives away the enrichment and monetises on the Google Shopping / Meta catalog ads performance lift, pricing on a share of attributable revenue rather than SaaS seat — a model that is structurally painful for Lily to copy. Documented weaknesses that compound: no publicly disclosed ARR, no public G2 / Capterra review volume despite eleven years in market (G2; Capterra, 2026), a B-1 extension rather than a priced Series C in March 2024 (Axios Pro Retail Deals, March 2024), and a pipeline that still requires non-trivial human-in-the-loop labelling — a cost structure the LLM-era entrants can plausibly undercut.

Adjacent-segment play

The core capability — vision-plus-language taxonomy generation over a catalog — generalises along several axes. Industrial / MRO (Grainger, Fastenal, HD Supply) is the biggest unexploited vertical: deep taxonomy, high search-abandonment pain, buyer willing to pay enterprise price, far fewer competitors than apparel. Automotive parts (Advance Auto, O'Reilly, RockAuto) carries similar characteristics with the additional complication of fit-and-compatibility logic. Grocery / CPG product data (where Syndigo, Salsify and Akeneo are entrenched) is a different buyer and arguably a crowded adjacency, but the LLM-era rewrite is open. Second-hand and resale (ThredUp is already a customer; Vinted, Depop, Poshmark, StockX) is the natural adjacency where consumer-speak attribute enrichment is operationally critical because sellers write their own listings — Lily has the technology to standardise them and the ThredUp reference to sell on. Mid-market Shopify / BigCommerce DTC is an obvious down-market expansion but it is defended by Shopify Magic and the LLM-era entrants at a price point Lily's cost structure cannot easily match. European tier-1 retail, now opened by the Marks & Spencer win in July 2026, is the geographic adjacency Lily is best positioned to lead on today — Mytheresa, Zalando, John Lewis, Harrods, LVMH's digital properties are the obvious next targets. Comparable adjacent-segment companies today: Syndigo and Salsify in MRO and CPG master data (incumbents, not AI-native); Pattern and Flywheel for marketplace enrichment; Spring and Chapsvision for European retail-data vendors. The adjacency Lily can most credibly lead on near-term is European enterprise retail; MRO is the biggest-dollar prize but a different buyer.

Sources and further reading

Capital history

DateRoundAmountValuationLead(s)
2016-2019 Seed / seed extensions Undisclosed (reported ~$2-4M cumulative) Undisclosed Unshackled Ventures, Canaan Partners (seed participation), Global Founders Capital (reportedly)
2020-01 Series A $12.5M Undisclosed Canaan Partners (lead); with NEA and Conductive Ventures (Venture Capital Journal; TechCrunch, January 2020)
2022-08 Series B $25M Undisclosed Canaan Partners (lead); with 3L Capital, NEA, Conductive Ventures; coverage in WWD, FashionUnited, TechCrunch, IT Digest (August 2022)
2024-03 Series B-1 extension $20M Undisclosed Fernbrook Capital Management (lead); with Global Brain and existing investors Canaan, NEA, Conductive Ventures, 3L Capital (GlobeNewswire; CMSWire; Pulse 2.0; Axios Pro Retail Deals, March 2024)

Investors / owners: Canaan Partners (seed, Series A lead 2020, Series B lead 2022, Series B-1 follow-on 2024), NEA (Series A 2020, Series B 2022, Series B-1 2024), Conductive Ventures (Series A 2020, Series B 2022, Series B-1 2024), 3L Capital (Series B 2022, Series B-1 2024), Fernbrook Capital Management (Series B-1 lead, March 2024), Global Brain (Series B-1, March 2024), Unshackled Ventures (seed) — immigrant-founder-focused seed fund

Competitive set

  • Syte — Tel Aviv-based visual-AI product-discovery platform; raised >$70M across rounds led by Viola Ventures, Magma Venture Partners and Naver Corp through 2021; roughly 100-200 employees (LinkedIn, 2025). Syte competes head-to-head on visual search and image-driven recommendations but has been strategically quieter since 2023 and did not announce a 2024-25 round, suggesting Lily has outrun Syte on US tier-1 retail penetration.
  • ViSenze — Singapore-based visual-commerce / product-discovery AI platform; raised >$30M including a 2024 round led by Lenovo's venture arm (ViSenze press, 2024). Attacks Lily's wedge from APAC retail footprint — Myer, Rakuten, Zalora — and increasingly from visual-shopping widgets rather than deep catalog enrichment. Different buyer motion (search team vs merchandising/data team).
  • Clarifai — US-based horizontal computer-vision / multimodal-AI platform; raised >$100M across rounds led by NEA, Menlo Ventures, Union Square Ventures and others. Clarifai sells into retail among many other verticals and does not package the retail-taxonomy workflow Lily productises — but the underlying vision models are strong and a motivated retailer could stand up a competent attribute pipeline on Clarifai + an LLM wrapper for materially less than a Lily contract.
  • Google Vertex AI Search for Commerce (GCP) — Google Cloud's hosted retail-search product, including Attribute Config and generative discovery features (Google Cloud docs, 2025). Already integrated into Macy's Google Shopping surfaces (one of Lily's own public case studies describes using Lily enrichment to feed Google Shopping better data). The quietly most strategic competitor: Google wants retailers' product data richer because its own Shopping and Vertex surfaces monetise on relevance, and it has every incentive to bundle attribute enrichment into GCP.
  • Algolia / Bloomreach / Constructor / Lucidworks — The onsite-search layer that Lily feeds. Algolia (~$320M raised, Accel/Salesforce Ventures-backed), Bloomreach (~$295M raised, Bain Capital Tech / Sixth Street-backed) and Constructor (~$80M raised, Silversmith) have been quietly adding automated attribute extraction into their own pipelines, which risks turning Lily into a feature inside the search platform rather than a parallel SaaS the retailer buys.
  • Shopify / Amazon catalog AI — Shopify (NYSE: SHOP) has been shipping native product-data generation and tagging into Shopify Admin since 2023 (Shopify Magic / Shopify AI, 2023-25). Amazon's seller-facing listing-generation tools do the same. The two-sided risk: Shopify SMB customers no longer need Lily; Amazon-adjacent merchants simply never shop a third-party vendor; and the DTC segment that was Lily's natural wedge in 2015-19 is now smaller than the enterprise opportunity but increasingly covered by platform-native tooling.
  • Envive, Velou, Width.ai, Naratix, Channable (LLM-era entrants) — A 2023-25 wave of smaller, LLM-native product-data-enrichment startups attacking from below — Envive (data enrichment, 2024), Velou (AI product-data enrichment, 2024), Width.ai (bespoke LLM pipelines for product attribute enrichment, 2024-26 content), Channable (feed management, longer-standing), Naratix (automated attribute enrichment, 2025). None are at Lily's enterprise-logo level but each is cheaper to procure and easier to replace, which anchors the market's price expectations downward.