Depthdata
Depthdata is an enterprise AI spend intelligence platform. It connects read-only to the admin consoles of ChatGPT, Claude, Copilot, Gemini and other vendors, merging adoption, spend and cost per outcome into a single governed view for CEOs, CFOs and CIOs.
What is Depthdata?
Depthdata positions itself as the system of record for a company's AI investment. The problem it addresses is fragmentation: ChatGPT, Claude, Copilot and Gemini each ship their own admin console, their own metrics and their own definition of usage, so companies spending six figures a year on AI cannot confidently answer the board's biggest question — is any of it working?
The mechanism is deliberately narrow. One read-only OAuth grant per vendor, then a daily read-only sync, feeds a five-step pipeline that is never run in reverse: INGEST, NORMALIZE, BASELINE, ESTIMATE, SCORE. During normalisation, every signal is attached to a canonical person through SCIM directory identity; records that cannot be matched are quarantined rather than guessed. Baselines are set against the company's own first four weeks, department by department, never against an industry benchmark. Every figure then carries one of three confidence labels: Measured (returned as-is by an API), Derived (computed from measurements, with the formula shown) or Modelled (an estimate with a plus-or-minus band).
Four product surfaces sit on top. Overview gives a single workspace health score; Coaching detects patterns among advanced users and routes playbooks and nudges to lagging teams; Analytics breaks results down by department, tool or seniority and handles exports; Leaderboard ranks on a depth score built from chained workflows, tool breadth and rework patterns rather than raw message counts. Cost-side connectors cover ChatGPT Enterprise, Claude, Copilot, Gemini and Cursor among others; production-side connectors cover Jira, GitHub, Linear, HubSpot, Salesforce, Zendesk and Intercom, so spend can be divided by resolved issues, merged pull requests, closed deals or solved tickets, weighted for complexity.
The published limits are as explicit as the claims. A connector transparency matrix names the vendors that expose nothing: Notion AI and Replit do not expose spend, Lovable exposes seats only through SCIM, Copilot audit events require GitHub Enterprise Cloud. Prompt and output content is never ingested, and where a vendor feed carries it the content is discarded — there is, the site says, no setting to turn that off. The workspace figures on display (287 licensed seats, 218 active, $48.2k of monthly spend, a 76/100 health score, $4.04k recoverable) come from a modelled demonstration workspace: the vendor states it has no customers yet.
What it does
- Connect the admin consoles of the AI tools you already pay for, through one read-only OAuth grant per vendor
- Normalise tokens, credits, actions and premium requests into a single schema, with one definition of an active user
- Compute a workspace health score from a published formula with published weightings
- Quantify waste line by line: dormant seats, inactive licences, duplicate coverage
- Break adoption, usage depth and spend down by department, tool or seniority
- Route nudges and playbooks to the teams leaving the most value on the table
- Export board-ready reports with the methodology attached, or push the data to your BI stack
When to use Depthdata / When not to
A quick filter to help you decide if Depthdata is the right fit.
When to use Depthdata
- Finance leaders who have to defend the AI budget line and answer the question the site puts on its own home page: “What did we get for the $48.2k?”
- CIOs and IT teams administering seats and licences across several AI vendors at once
- Procurement and vendor management, who need a ranked waste list they can hand over and defend row by row
- HR and enablement leads driving adoption, with nudges, playbooks and department-level breakdowns
- Executives and board members who want a single workspace health score and a board-ready PDF export
When not to use Depthdata
- Individuals and freelancers: the product assumes a multi-vendor estate and admin consoles you control
- Companies on consumer plans — most vendor admin APIs require that vendor's Enterprise or Business tier, and coverage reflects the plan you hold with each one
- Anyone wanting the quality of AI answers scored: the vendor rules content quality scoring out permanently
- Anyone expecting an industry benchmark or a forecast: there is no benchmark chart, and benchmarking is named as a future direction with no date
- Buyers who need reference customers or a dated roadmap: Depthdata states it has no customers yet and puts no timelines on the page
How to use Depthdata
A typical end-to-end flow, from setup to results.
- Attach accounts: grant one read-only OAuth connection per vendor, from the admin console you already pay for — no agent to deploy, so no security review to schedule
- Let the data sync: daily pulls rewrite each vendor's units into a common schema
- Map identity: signals are attached to people through your SCIM directory, and unmatched records are quarantined rather than guessed
- Anchor your baselines: define what good means per department, measured against your own first month rather than an industry average
- Review the insights: four labelled headline figures, then the ranked waste list
- Route the insights: health scores to executives, nudges to team leads, spend alerts to finance
- Take action: reclaim seats, resize tiers and steer enablement — every item names the seats it refers to
- Measure the improvement: the same pipeline re-scores whether the action worked, against the same baseline
- Export anywhere: push to your BI stack, or export board-ready reports with the methodology attached
Pros & Cons
Pros
- Every figure carries its confidence label — Measured, Derived or Modelled — and estimates arrive with a band
- The methodology is public and versioned (Measurement spec v3.23), shipped inside the product and attached to every export
- Agentless deployment: read-only OAuth, nothing installed on endpoints, and no security review to schedule
- Prompt and output content is never read, and anything that arrives anyway is discarded at ingestion
- The connector matrix names, vendor by vendor, exactly what is not exposed
- Progress is measured against the company's own history rather than an industry average
- The vendor publishes its product trade-offs, down to the larger saving it refuses to recommend (31% against the 8.4% it stands behind)
Cons
- No public pricing: no rate card, no pricing page and no announced free plan — the only way in is a demo request
- No terms of service, privacy policy or security page: the footer links are dead anchors and /privacy, /terms and /security return 404
- SOC 2 Type II and GDPR appear as badges on a product card, with no compliance page behind them
- The product has no customers yet, by the vendor's own statement, and every figure on display comes from a modelled demonstration workspace
- Coverage depends on the tier you hold with each vendor — most admin APIs require Enterprise or Business — and two of the eight vendors tracked expose seats only
- No scoring of AI answer quality, no industry benchmark, no forecast and no dated roadmap
- No DPA, no sub-processor list and no published retention period; the case-study page still shows three unfilled “[ STAT NEEDED ]” blocks
Pricing & Plans
No pricing is published. The website has no pricing page — /pricing returns 404 — and states no amount, no currency and no tier anywhere. Neither a free plan nor a free trial is announced. The only commercial route is a demonstration request, through the “GET DEMO” call to action and a contact form asking for a name, a work email, a company and a free-text message. The pricing model is therefore contact-sales, established by deduction rather than by any statement from the vendor.
Data, GDPR & hosting
A consolidated view of how Depthdata handles your data.
GDPR overview
A “GDPR” badge appears on the SECURITY card of the platform section, alongside SOC 2 TYPE II, SSO / SAML, SCIM, DATA RESIDENCY and AUDIT LOG. These are graphic elements on a product card: no compliance, security or privacy page documents any of them, and the word GDPR is never developed into a written compliance claim. There is no DPA, no sub-processor list, no Article 27 representative, no DPO and no GDPR contact address, and no retention period is published. What is genuinely documented is a posture rather than a compliance programme: read-only access, metadata only, and conversation content discarded at ingestion — “Metadata-scoped API access. No writes, no device agents, no scraping, no content.” Anyone with a GDPR obligation should treat the badge as a claim to verify.
Who owns the data?
No terms of service, privacy policy or security page is published: the footer's Privacy, Terms and Security links are dead anchors, and /privacy, /terms and /security all return 404. Ownership of customer data is therefore governed by no contract a reader can consult. The only statements available are marketing claims on the home page — “Your data stays yours” next to the export step, “your data never trains anyone's models” on the platform security card, and “your employees' conversations are not our business”. Access is described as read-only OAuth with read-only API scopes, no agents, no browser extensions and no scraping. At the date of collection, all of it should be treated as an unverified claim rather than a commitment.
Reuse rights
Reuse by the customer is designed into the product rather than granted contractually. Depthdata is built to push data to your BI stack or to export board-ready reports, and the calculation method travels with the data: the formula ships with every export. Branded PDF reports are produced in one click. However, no terms of service exist at the date of collection, so nothing legally frames what you may do with those exports, who may receive them, or what the vendor may do with the underlying metadata. In practice reuse looks unrestricted; contractually, it is simply undocumented.
Data retention & training
Hosting summary
No security or privacy page is published: /security returns 404 and the footer link is a dead anchor. The only hosting-related statement on the site is a “DATA RESIDENCY” badge on the SECURITY card, with no country and no region named. No jurisdiction, no sub-processor and no data centre is disclosed for the product platform itself. What can be observed concerns the marketing site only, not the platform that would hold customer data: depthdata.vercel.app resolves to 64.29.17.131, an anycast node on AS16509 (Amazon.com Inc.) geolocated to Walnut, United States, served from the Vercel hosting platform, and the vendor's own domain depthdata.app points at the same deployment. Nothing licenses the assumption that customer metadata would be stored in the same place. The scope of what would be hosted is at least described: metadata only — seats, sessions, usage events, spend and metadata about completed work — with conversation content discarded at ingestion.
Things to keep in mind
Risks and trade-offs to weigh before adopting Depthdata.
- Individual usage measurement: per-employee profiles, depth scores and named leaderboards are read as surveillance even when no prompt is ever opened
- Gamification: a leaderboard, even one built on depth, pushes people to use AI for the sake of using it — the vendor is aware of this and answers that the only way to game it is to actually get better at AI
- Consent controls are presented as employee-facing, but the settings themselves remain under the employer's control
- Modelled figures such as hours saved or return on AI can end up in a board deck after the confidence band has been lost to a copy-paste
- Seat-cutting decisions taken on activity thresholds risk removing genuinely useful access from atypical profiles
- No terms, no DPA and no published retention period: contractual and GDPR exposure is undocumented at the date of collection, and the SOC 2 Type II and GDPR badges must be verified before any security review
- Partial vendor coverage produces an incomplete picture presented as a unified one, on a product that has no customers yet and shows demonstration figures from a modelled workspace
Setup & Integrations
Technical difficulty
Low for an administrator, out of reach for anyone else. The site advertises one hour to go live, admin consoles only, zero installs, and about twenty minutes of read-only grants: one OAuth connection per vendor with read-only API scopes, nothing installed on endpoints and, in the vendor's words, no agent to deploy and so no security review to schedule. The real prerequisites are organisational: admin rights on every vendor console, the Enterprise or Business tier each admin API demands, a SCIM directory to attach signals to people, and a configuration step where you set per-department baselines.
Deployment
Integrations
Behind Depthdata
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Does Depthdata read our conversations?
What has to be installed to deploy it?
Which tools does it connect to?
How can it report a return without an ROI figure?
How much does it cost?
How long does it take to go live?
How do I know whether a number is measured or estimated?
What happens when a vendor exposes nothing?
How is the work produced actually measured?
Are there reference customers?
Should you pick Depthdata?
Depthdata addresses a real and still young category: a normalisation layer above the admin consoles of AI vendors, so adoption, spend and cost per outcome can be read from one ledger instead of six dashboards. Its distinctive move is epistemic honesty. Every figure carries a confidence label, estimates arrive with a band, the measurement specification is versioned and published, and the connector matrix names the vendors that expose nothing rather than hiding the gaps. The site even publishes the larger saving it refuses to recommend.
Maturity is the other half of the picture, and the vendor says so itself: Depthdata has no customers yet. Every number on display — 287 licensed seats, $48.2k of monthly spend, a 76/100 health score, $4.04k recoverable — comes from a modelled demonstration workspace, not a deployment. The site carries the marks of that stage: partly inert navigation, three unfilled “[ STAT NEEDED ]” blocks, no legal pages of any kind, no pricing, and SOC 2 Type II and GDPR shown as badges with nothing behind them. The vendor's own domain was registered on 18 August 2026, and the product launched on Product Hunt at the end of July 2026.
There is also a hard entry condition: coverage reflects the tier you hold with each vendor, and most admin APIs require an Enterprise or Business plan. Without those tiers, a partial view still presents itself as a unified one.
The bet suits organisations running several hundred AI seats that need to defend a budget line to a board; for them, the hour it takes to connect is worth spending on an evaluation. Any buyer whose process requires a DPA, a retention policy or a reference customer should revisit this one later.
- Choosing a selection results in a full page refresh.
- Opens in a new window.