AI Data Analyst in the USA
Your team asks in English. The service finds the tables, writes the SQL, runs it against a read-only replica inside your own cloud, and returns the number with the query attached. It cannot touch a table whose schema it has not just read, and that is enforced in code rather than requested in a prompt. Starts at $0: you pay at checkpoints you accept.
The US is where AI budgets are largest and engineering rates are highest, which is exactly why so much of this work is quoted at numbers that have little to do with the effort involved. An agency will price a production MVP in the mid five figures and staff it with three people who have shipped less than the one you actually wanted. The gap between what a build costs and what it is billed at is the whole reason this arrangement works.
What you get
- Deployed in your own cloud account against a read-only replica you create, so we never hold a credential to your database
- A signed schema check the model cannot skip: no SQL runs against a table it has not just inspected, and the tokens expire in fifteen minutes
- A grant list your data owner approves, naming the exact tables and columns the analyst may read, with everything else refused at the database
- Exposed inside Claude or ChatGPT in your organisation's workspace, behind your existing Google sign in, so nobody learns a new interface
- Every question logged with the SQL it ran, the tables it touched and whether it errored, which is what an auditor actually asks for
- Runs on a private LLM on your own hardware where data security demands it, up to fully air-gapped, so nothing at all crosses your perimeter
No payment before a checkpoint is delivered and accepted. Most builds land between $6,000 and $18,000 depending on how many databases are in scope and how much of the schema the first domain covers. Optional upkeep from $1,200 a month, and only if you want it.
Invoiced in USD, payable by Wise or bank transfer.
Request a callback
You speak to the engineer who does the work. No sales rep, no deck.
How many free signups from June converted to paid, by week?
customersplan tier is a column here, not a separate status table
subscriptionsone row per plan change, so take the first paid one
AQIAaPr2xQAAAACqL3d1ZwSELECT DATE_TRUNC('week', s."startedAt")
AS "Week",
COUNT(DISTINCT c.id) AS "Converted"
FROM customers c
JOIN subscriptions s
ON s."customerId" = c.id
WHERE c.plan = 'free'
AND s.plan = 'paid'
AND c."createdAt" >= DATE '2026-06-01'
AND c."createdAt" < DATE '2026-07-01'
GROUP BY 1 ORDER BY 1| Week | Converted |
|---|---|
| 2026-06-01 | 214 |
| 2026-06-08 | 188 |
| 2026-06-15 | 243 |
| 2026-06-22 | 201 |
Stage three refuses to run unless it is handed a valid token from stage two for a table in the query. The model cannot skip the schema read, because the check is a signature in code rather than an instruction in a prompt.
Get it built at $0.
That is not a discount. It is when you pay. The work is split into checkpoints with acceptance criteria written down before anything starts, and each checkpoint is invoiced only after you have seen it and accepted it. No deposit.
- $0 to start
- You hold every dollar until a checkpoint is delivered and you accept it. No approval, no invoice.
- Fixed cost, unlimited features
- Or hire the team outright: one fixed monthly cost, unlimited feature development, any stack.
- The engineer takes your call
- The person on your first call is the one who architects and writes it. No account managers, no bench time.
A US agency quotes $50,000 to $150,000 for the same build and asks for 40 to 50% of it before a line is written. Account managers, project managers, sales commission and bench time. None of it appears in your product.
Three stages, nothing hidden.
Fixed-fee assessment
Five business days from the day read-only access is in place. Schema inventoried, the questions your team actually asks classified as answerable or not, the grant list drafted with whoever owns data protection, and acceptance criteria written as testable statements.
Build in checkpoints
About three weeks. Deployed in your cloud account, connected to your replica and to your assistant workspace, with the semantic layer written for the first domain. Each checkpoint has criteria agreed before work starts and is invoiced only once you accept it.
Handover and closeout
A real question from your team, answered end to end, with the SQL shown and the numbers reconciled against the source rows. Runbook, recorded handoff, code and semantic layer in your repository, and all our access removed with written confirmation.
What actually applies here.
Regulation and data
The pressure almost never comes from a regulator directly. It comes from your own customers. Enterprise buyers now run security reviews that ask where inference happens and who can read the prompt, and healthcare and financial clients increasingly write data-handling terms into contracts that a shared model API cannot satisfy. When that happens, moving inference inside infrastructure you control is usually the shortest path to a signed deal rather than a technical preference.
Contracting and payment
You contract with an individual consultant based in India rather than a US entity. Invoices are issued in USD and paid by Wise or bank transfer, with no payroll and no benefits load on your side. Whatever documentation your finance team or counsel needs from an overseas contractor is provided before work starts.
Working hours
Four or more hours of daily overlap with US Eastern, calls scheduled in your morning, and same-day replies on working days.
Do you work with US companies from India?
Yes, and most engagements are with US companies. You contract with an individual consultant, invoiced in USD, with any documentation your finance team needs provided up front. No US entity, no payroll, no benefits overhead on your side.
How do you handle the time difference with US teams?
Four or more hours of daily overlap with US Eastern, calls in your morning, and same-day responses on working days. Work needing review is handed over at the end of your day and waiting when you start the next one.
How do you handle access to our systems?
Accounts stay yours throughout. Access is named and limited-privilege, work runs on synthetic data where possible, code and configuration are delivered continuously into your repository, and all access is removed at handover with written confirmation.
We already have a BI tool. Does this replace it?
No, and it should not. Your dashboards answer the questions somebody anticipated, and they will keep doing that. This answers the ones nobody built a chart for, which is where the waiting happens. Teams that keep both end up using the dashboards for the standing numbers and the analyst for everything else.
How long before somebody can ask a real question?
Usually inside the first week of the build, on a narrow slice of the schema. The first domain being useful early is the point of doing it in checkpoints: you see it working on your own data before most of the money is committed.
What if our schema is a mess?
Most are, and it changes the shape of the work rather than ruling it out. Undocumented columns, legacy tables nobody dares delete and two spellings of the same idea are normal. The assessment tells you plainly how much of that has to be untangled first, and if the answer is that you need a data cleanup before an analyst is worth building, you will be told that instead of sold this.
AI Data Analyst
An AI data analyst on your own read-only replica. Natural language to SQL that cannot query a table it has not read.