Scope & data · Venture Client Directory

First principles

Demand firstCorporates are the demand side of this market, and a platform starts with its demand side. That sets the build order.

Counts, not claimsThe crawl is precise enough for totals per corporation and too imprecise to assert that two named companies worked together.

Understatement convertsA corporation that sees too few collaborations on its profile has a reason to claim it and correct the record.

Static and containedPublic pages are pre-built and every table lives in one schema, so the site survives a database outage and the dataset moves in one command.

Studio: the GlassDollar backend

Everything on the directory is editable by your team, without us. gd.sanerebels.com/studio

Signing inEnter your @glassdollar.com address, we mail you a link, the link signs you in. No password exists. The address is checked on our server before any mail goes out, so nobody outside your domain gets in. The built-in mailer allows two sign-in mails per hour across the whole team, and a session lasts for days, so sign in once and stay in.

OverviewClaimed profiles, open claims, and four work queues: missing description, unresolved industry, missing logo, and the profiles your team has edited. Each queue is a button that opens exactly those profiles.

ClaimsEvery claim with company, contact, role, and whether the person confirmed their email address. Approve marks the profile claimed and adopts the text the claimant wrote about what they scout. Reject leaves the profile untouched. Both are one click and safe to repeat.

CorporatesSearch or filter, then edit name, industry, description, what the unit scouts, the link to your own challenge page, and the claimed and hidden flags. A field a human edits is protected: the next data import will not overwrite it.

VisualsTwo ready-to-post formats: a ranking per industry and a map per region. Square 1200x1200 by default, which is what LinkedIn recommends, landscape available. Every image carries the date it was made, so a repost is visibly newer, and the map frame stays the same from post to post so two months compare.

PublishEdits and approvals are on the public directory within seconds, no publish needed. The Publish button rebuilds the ranking pages and the offline copy of the data, which takes about a minute. Overview tells you when a rebuild is worth doing.

When something is wrongNothing in the Studio is destructive: profiles hide rather than delete, and every edit records who made it and when. Write to Simon and it gets fixed or rolled back.

Deepdive: VCL Data+

Frederic's CSV, dissected: 18 columns, 1,000 collaboration rows from the ~400,000-row crawl. Every column measured, every column given a verdict. Counted 2026-08-14.

695corporations · 92% with description, 84% with a clean industry

992startups · 96% with country, 99% with HQ city

$34.5bnfunding behind the involved startups, where the crawl carries it (39%)

61collaborations with a public case-study link

319collaborations with US startups, the largest origin country

9%of collaborations sit with the top 10 corporations

45corporate HQs, stand-ins so the map and region filter can be shown. Frederic's file replaces them Mon/Tue

5%collaborations with a type: 47 pilot, 3 converted

Every column, with two real values

The full field list of the sample, measured on 1,000 rows. Percentages are how often a column carries a usable value.

ColumnFilledTwo real values

startup_domainidentity key for the startup side100%elimity.comipool.se

corporate_customer_domainidentity key for the corporate side100%sncb.be ica.se

corporate_customer_descriptionone-line company description92%Walmart is a retail corporation that operates several chains of discount departments and warehouse stores. ICA is an online shop that offers ICA's lunch box, healthy food and catering services.

corporate_customer_industriescrawl tags, unnormalised90%{e-commerce,"food and beverage","supply chain management"} {e-commerce,grocery,retail,"retail technology",shopping}

startup_short_descriptionone-line startup description100%Elimity helps their customers to protect their important information assets with superior identity governance solutions. Ipool is a user-friendly application that supports workforce management.

startup_long_descriptionparagraph version, adds nothing over the short one73%Elimity eliminates identity-based risks and implements mature identity governance… Eoliann harnesses satellite data and proprietary ML algorithms to predict climate risk…

startup_founded_yearfounding year of the startup100%2017 2010

startup_hqcity and country of the startup99%Mechelen, BEL Stockholm, SWE

startup_countryISO country code96%BEL SWE

startup_employee_countheadcount as a range98%1-10 11-50

startup_employees_countheadcount as an exact number, a second column45%10 19

startup_funding_usdtotal funding raised39%893076 1596572

startup_investorsnamed investors40%{Tenity,"Gemma Frisius Fund"} {"Primo Capital","Exor Ventures","Compagnia di San Paolo"}

startup_categorycrawl tags on the startup side95%{"cyber security","information technology","risk management"} {"facilities support services",software}

startup_category_groupcoarser grouping of the same tags95%{"information technology","privacy and security"} {"administrative services",software}

startup_referenced_customersother customers of the startup, with a partnership type18%[{"type": "pilot", "domain": "workday.com"}] [{"type": "pilot", "domain": "rabobank.com"}]

partnership_case_study_urlpublic write-up of the collaboration6%thoropass.com/customers/rimidi/ workato.com/customers/nutanix

partnership_created_atwhen the crawler saw the row, not when the collaboration happened100%2023-11-30 06:24:40+01 2025-07-28 08:02:44+02

BUILD

Identity, descriptions, startup geography

Both domain columns are 100% filled, and the domain is the identity key everywhere. Corporate descriptions 92%, startup descriptions 100%, founded year 100%, startup HQ 99%, country 96%. Plus 61 rows with a public case-study link: the provenance we link instead of asserting collaborations ourselves.

CARE

Usable only through rules

Industries and categories (90% / 95%) are messy crawl tags and reach the UI only through the deterministic taxonomy. Funding (39%) sorts the works-with examples and is never published as a company figure. Employee ranges (98%) beat the numeric counts (45%). Investors (40%) and referenced customers (18%) enrich where they match and stay silent where they do not.

SKIP

Flaky crawled datapoints we do not build on

partnership_created_at looks like a date and is a crawl artifact: 40% of rows sit on two days in July. No feature touches it. Partnership types resolve on 5% of rows, too thin to filter or badge on. Long descriptions (73%) add nothing over the short ones.

TODAY

What the solid columns answer

Who buys most: Amazon 18, Google 17, Microsoft 14. Which industries buy: Retail & Consumer 180, Technology 154, Automotive & Mobility 133, Financial Services 125, Telecom & Media 84. Where the startups sit: USA 319, UK 94, Germany 60, France 58, India 39, Canada 36. How concentrated: barely, the top 10 hold 9% and 528 of 695 corporations show exactly one collaboration.

MONDAY

What the locations unlock

Built and running on stand-in coordinates already: HQ pins as a second map layer, the countries stat, and the region filter. Frederic's file swaps the coordinates and widens the coverage; still to come are region x industry slice pages ("Top Financial Services, Germany"), buyer country against startup country, and the real cutoff (3 or 5) from the full distribution.

NEVER

What this dataset cannot answer

Trends over time: the dates are crawl timestamps. Deal sizes: not collected. Who to call: company and claimed state only. Completeness: the crawl finds what companies published, so every count is a floor.

Decisions logged

Settled in the call on 2026-08-12 and in the days after. Each one is live in the product.

Primary userCorporates. The startup side is deprioritized.

Conversion eventClaim my profile, answered with an offer of a 15-minute call in which GlassDollar fills the profile in. The demo path stays on glassdollar.com.

RankingIdentified collaborations, with counts below the cutoff shown as "fewer than 5" so unreliable rows cannot look precise.

Named peopleNames of innovation managers stay out of the product entirely. Profiles show the company and its claimed state.

Looking forAI-drafted for the top 100, visibly labeled, corrected by claiming.

ChallengesEach profile can link out to the corporation's own challenge page. GlassDollar hosts no challenge module.

ProposalsOut of scope, so that GlassDollar never becomes the pre-qualification bottleneck.

Claim verificationWork email on the corporation's own domain, plus a confirmation link to that address.

Content engineIndustry and region slices as pages and LinkedIn visuals. Recurring top-10 formats.

DomainA glassdollar.com subdomain. App and database run on KNUS infrastructure, transferable on request.

LaunchSeptember, with the LinkedIn push.

Decided on 14 August

The small calls of one working day, so nothing has to be reconstructed from the build.

TaxonomyProfessional Services added as the eleventh category. Audit and staffing firms sat under Financial Services.

EntitiesWWF and Sebrae removed from the ranking. An NGO and a public agency are not corporations buying from startups.

DescriptionsRewritten into one neutral sentence each, median 72 characters against the crawl's 413.

Names36 display names repaired where the crawl had concatenated the domain.

Profile orderRank first, then what the unit scouts, then partners, then the description.

Multi-rankEvery ranking a corporation places in, narrowest first.

SearchDrops the ranking cutoff, so corporations below it stay findable. Covers city and industry.

MapPins cluster, and the map opens on the visitor's region, read from the browser time zone.

Claim formTwelve scouting categories to tick alongside the free text.

Claim alertA new claim posts to a webhook, so it does not wait in a queue nobody opens.

Bot defenceTurnstile on the claim form and the Studio sign-in, inert until the keys are set.

FindabilityOwn robots.txt, generated sitemap, structured data, llms.txt, Open Graph image.

VisualsSquare 1200x1200 by default, every image dated, map regions on a fixed cutout.

Growth chartBuilt and removed the same day. The curve drew the crawl window, not corporate behaviour.

Rank historyOne snapshot per corporation per day, with industry and country, starting today.

Call length15 minutes instead of 30.

Claimed tileAppears from 25 claimed profiles. A public zero says the opposite of what it is for.

Trial cardGlassDollar's own product card sits in the hero and links to the trial.

Phone layoutFigures instead of filled cards, ten per page, pager on one line.

Copy ruleNo sentence that explains the reader's own motivation back at them.

Requirements

The constraints the build has to hold, independent of features.

Not bulk-extractableNo export endpoints, paginated pages, and rate limits on every write path, because the dataset is GlassDollar's asset.

Listing is not opt-inEvery profile carries a visible correction and removal route, with one owner who answers.

Provenance where it existsWhere a collaboration has a public case study, the profile links it as the source; everything else stays a count.

Human edits winOnce a field is edited in the Studio, the data loader never overwrites it.

Germany basics before launchImpressum, privacy policy, terms. Cookieless analytics, so no consent banner.

Noindex until launchMeta tag and header, removed together on the final domain.

Access denied by defaultRow-level security on every table, service-role reads at build time, no database credentials in the browser.

One schemaEverything lives in the gd_directory schema, so the dataset can be dropped or handed over in a single command.

Still open

Calls that are not ours to make, or that need the full dataset first.

Cutoff: 3 or 5Set from the real collaborations-per-corporation distribution once the full CSV lands.

Taxonomy sign-offTen industries, keyword rules. Needs one review pass from Frederic.

Logos at scaleTrademark posture for showing corporate and startup logos. Initials are the safe fallback.

Removal flowAgreed in principle, not designed. Who decides, how fast, what the profile shows afterwards.

Per-profile SEOGoogle currently finds the industry ranking pages, not the individual corporations. Giving every corporation its own page would multiply the search surface, and nobody has decided yet whether that is worth building.

RecurrenceCollaboration counts move slowly and dates are unusable, so what brings visitors back after launch week is unsolved.

Concept drafts

Explorations from the scoping week. Kept for reference, not maintained.

The original exportGlassDollar's design-tool prototype, restyled. The starting point.

The DirectoryProfile-first browsing. Superseded by the landing page.

The ListSearch across every unit, with no profile pages behind it.

MatchPick what you build, see who buys it.

The EvidenceThe category in numbers, every figure with its source.

The BenchmarkFive questions score a unit against the published benchmarks.

The MapVenture client units as pins. Live.