BigQuery MCP server: what an AI agent can query, and the guardrails it needs
Google's BigQuery MCP server gives an agent 9 tools, SQL on raw tables and up to 3,000 rows. It has no per-query byte cap and no metric definitions.
9 min read
Precisian Blog
Precisian technical blog — AI-powered data integrity
Ecommerce KPIs: the definition each team needs to agree on
Nine ecommerce KPIs, each with its formula and the choice that moves the number: gross or net, shipping, tax, cancellations and attribution window.
8 min read
Fivetran pricing in 2026: how MAR billing works and what drives the bill
Fivetran bills each connection by monthly active rows. On Standard, the first million MAR costs $500 a month at list price, and the rate falls per connection.
8 min read
What an MCP gateway does (and what it can't fix)
An MCP gateway centralizes auth, call logging and rate limits for AI agents. It can't decide what a metric means. That lives in the semantic layer.
7 min read
Testing an MCP server before it touches production data (MCP Inspector and beyond)
MCP Inspector proves your server speaks the protocol, not that its answers are right. The rest: a scrubbed copy, an answer key, scope and cost checks.
9 min read
MCP vs API: what changes when an AI agent is the client
MCP doesn't replace your API; most MCP servers call one. What changes is who writes the call: a developer at build time, or a model at runtime.
6 min read
Snowflake MCP server: what the agent sees and what still needs a semantic layer
The Snowflake-managed MCP server exposes only the tools you declare. Answer quality still depends on a semantic view your team writes and maintains.
8 min read
UTM parameters: a naming convention that survives five teams and three agencies
GA4 reads nine UTM parameters and treats values as case-sensitive text. Without one written convention, a single campaign splits across report rows.
10 min read
What is gclid? How the Google Ads click ID gets lost before the sale
gclid is the Google Ads click ID. It arrives in the URL, lives in a cookie on your domain, and only reaches Google again if you store it with the order.
7 min read
Windsor.ai alternatives for ecommerce data teams in 2026
Windsor.ai starts at $19 a month but caches data for about three weeks. Nine alternatives scored on seven criteria, with when to stay put.
9 min read
15 best AI tools for data analysis for ecommerce in 2026
A ranking of 15 AI tools for data analysis on criteria C1 to C7. Ten of the 14 competitors need a data warehouse you have already built.
9 min read
14 best CDPs and alternatives for ecommerce in 2026
13 CDPs and Precisian scored on seven data criteria. None of the 13 documents versioned, customer-approved metrics. Activation is where CDPs win.
9 min read
10 best ETL tools for marketing data in 2026
Nine ETL tools and Precisian scored on seven criteria. None of the nine documents versioned, customer-approved metrics; most leave modeling to you.
8 min read
Triple Whale alternatives: when Precisian fits, and when it doesn't
For a US Shopify brand, Triple Whale is usually the right call. Precisian serves e-commerce in Brazil, so it's the alternative only if you operate there.
8 min read
Triple Whale alternatives: 15 marketing attribution and analytics tools for 2026
15 Triple Whale alternatives ranked on seven criteria: data isolation, versioned metrics, MCP access, sources, who runs the data, and data ownership.
9 min read
How to give an AI agent customer context without exposing PII
Cohorts, segments and LTV all come from a pseudonymous ID. An agent almost never needs a name, email, phone number or address to answer them.
7 min read
Which AI agents can reach your data? Building the agent and MCP permission inventory
No console lists every AI agent touching your data. An MCP permission inventory needs seven columns and a monthly and quarterly review.
9 min read
ctwa_clid explained: the click ID behind Click-to-WhatsApp ads
ctwa_clid is the click ID of a Click-to-WhatsApp ad. It arrives once, in a webhook, and only attributes a sale if you store it with the order.
7 min read
Data governance for ecommerce teams: the minimum that actually works
Data governance records who decides what about each dataset. For ecommerce, the useful minimum is five records: owner, definition, access, retention, change.
6 min read
What is data lineage, and how do you trace a marketing metric back to its source?
Data lineage maps a number from source to report. dbt's column-level lineage shows the columns, but not the filter that changed the result.
6 min read
What counts as first-party data for an ecommerce brand, and where should it live?
First-party data is what you record in a direct relationship with your customer. Modeled conversions and platform audiences look like it and aren't.
7 min read
What goes in a marketing data dictionary (and why your AI needs one)
A data dictionary records each metric's name, definition, source, owner, formula and exclusions. A 4 KB definitions file lifted AI accuracy 17 to 23 points.
7 min read
How to log every query an AI agent makes through MCP
Write the agent audit log on the server that runs the query. MCP logging was deprecated in the 2026-07-28 spec and was never built for audit.
7 min read
Meta Conversions API: what it recovers and what it doesn't fix
The Conversions API sends Meta the events your server recorded. Without a Pixel-matching event_id inside 48 hours, one purchase counts twice.
9 min read
One AI agent, many client accounts: how to keep tenant data apart
When one agent serves many tenants, data rarely leaks through an attack. It leaks via leftover context, a cache keyed without the tenant, or one shared token.
8 min read
AI data governance: what the policy has to cover once an agent reads customer data
AI data governance decides what a model may read, not just who may use it. Five questions cover it: scope, isolation, retention, audit trail, metrics.
8 min read
ChatGPT and Claude for marketing data analysis: what they get right and what they make up
ChatGPT data analysis is safe for exploring a spreadsheet and unsafe for deciding with it: the math is right, the revenue definition is a guess.
8 min read
Composable CDP vs data lake: which one an ecommerce brand needs first
A composable CDP activates customer profiles that live in your warehouse. A data lake keeps every source for analysis. One sends, the other explains.
7 min read
Data lake vs data warehouse vs lakehouse for ecommerce and marketing data
A lake keeps raw data, a warehouse keeps modeled data, a lakehouse puts transactions on open files. None of them defines net revenue.
8 min read
Fivetran alternatives for ecommerce data: what the pricing page leaves out
Fivetran bills per monthly active row and Airbyte by credit, but in Airbyte Cloud only 67 of 573 source connectors are maintained by Airbyte itself.
7 min read
The official Google Ads MCP server: what it answers and what it can't see
Google's Ads MCP server reads your account through GAQL, read-only. It reports what Google logged as a conversion, not shipped orders, Amazon sales or returns.
7 min read
Google Analytics MCP: why the numbers it returns are already modeled
The Google Analytics MCP server reads GA4 through the Data API, which follows your reporting identity. On Blended, the numbers arrive already modeled.
8 min read
Why text-to-SQL fails on marketing data, even when the query runs
The query that runs is the worst case of text-to-SQL: the wrong revenue column, the wrong window or a fan-out join returns a clean number and no error.
9 min read
LTV:CAC for ecommerce brands that also sell on Amazon
LTV:CAC divides the margin a customer returns by what it cost to win them. Selling on Amazon hides repeat buyers and platform data shrinks CAC.
7 min read
MCP security for marketing data: read-only, scopes, and the permission you shouldn't grant
Google's Ads MCP server is read-only, yet its OAuth scope can edit and delete your account. For ad data, MCP risk lives in the token's scope.
9 min read
MER vs ROAS: measuring ad efficiency on revenue you actually booked
MER divides total booked revenue by total marketing spend; ROAS divides platform-attributed revenue by one platform's spend. Formulas and limits.
6 min read
Meta Ads MCP: the official connector, its permissions, and whose numbers it returns
Meta's official Ads MCP server entered open beta on April 29, 2026 with 91 tools. The conversions it reports are the ones Meta attributes to itself.
7 min read
Shopify's MCP servers are built for shopping, not for revenue questions
Shopify's official MCP servers act for buyers, customers and developers. None answers a merchant's margin-by-channel question; community servers fill that gap.
8 min read
Supermetrics alternatives when the consumer is an AI agent, not a spreadsheet
Windsor.ai, Funnel and Improvado all ship MCP servers now. Choose a Supermetrics alternative by who defines the metric, not by connector count.
7 min read
Connect Google Ads, Meta and GA4 to one AI and you get three revenue numbers
Google Ads counts purchases 30 days after a click; Meta, 7 days after a click or 1 after a view. Plug all three MCPs into Claude and one order counts twice.
8 min read
Build or buy a data lake: only one side publishes prices
S3 storage is US$ 0.023 per GB-month from a published table. The managed side documents the model and not the rate.
7 min read
Redirect payments and your missing GA4 purchase event
Mollie says it plainly: the customer may not return. Your purchase event lives on the page they never reached.
8 min read
Are third-party cookies actually gone? Read the dates
Chrome still enables them by default. Safari and Firefox blocked years ago. The replacement programme was wound down.
7 min read
Attributing sales that close inside WhatsApp
The webhook hands you the ad ID, not the campaign. And the person identifier has no equivalent on the web side.
8 min read
Answer lineage: auditing what an AI agent actually read
The chat transcript records what was said, not what was true. Lineage links an answer to the definition and data behind it.
6 min read
Why GA4 and your BigQuery export will never agree
Google lists BigQuery export among features that do not support behavioural modelling. The interface estimates; the export observes.
8 min read
Marketplaces return traffic, never its source
Amazon's own schema returns 28 traffic fields per product and zero origin fields. The absence is in the schema, not your integration.
8 min read
What breaks in your analytics during peak season
GA4's BigQuery export caps at 1 million events a day, and an exceeded export pauses without reprocessing previous days.
7 min read
Semantic layer vs data warehouse: they don't compete
One stores and computes; the other defines what is being computed. There is no NOT NULL for semantics.
7 min read
Teaching a data agent to say I don't know, and why that is hard
On a benchmark that prices wrong SQL, abstaining from everything scores 50% and the best real systems score 29.8% to 54.5%.
8 min read
Giving an agency data access without giving everything
GA4 has 5 roles and 2 data restrictions, one hiding revenue. The market default is to grant Administrator anyway.
8 min read
How to tell if your data pipeline broke, or just lied
Airflow documents it in one line: if the run never finishes, the SLA is never checked. That is the easy version.
8 min read
Why your product isn't showing on Google Shopping
Preemptive item disapproval disapproves what Google suspects, not only what it checked. And GTIN is not required; brand is.
8 min read
How long does GA4 keep data, and how long should you keep marketing data?
GA4 standard allows 2 or 14 months, and large properties get 2. The aggregate report survives; the exploration does not.
8 min read
MCP server vs BI tool: which question is each one for?
A dashboard answers the question already asked; an MCP server answers the one nobody anticipated. Both read the same definition.
7 min read
Why Shopify and GA4 revenue never match
Adobe says it plainly: the two should not match in almost every case. GA4 excludes tax and shipping; Shopify includes both.
8 min read
What the ERP returns as sales data (and what it does not)
The Omie order returns neither CPF nor CNPJ, and in Olist the field called origin is not the marketplace: it means point of sale.
9 min read
Why you can't compute LTV across marketplace and own store
Amazon documents its buyer email as anonymized, and no order carries a traffic source. Cross-channel identity is absent by design.
8 min read
MCP for e-commerce analytics: expose questions, not tables
The spec makes tools model-controlled and tool annotations untrusted. Your tool surface is the security boundary.
7 min read
Prompt injection through marketing data: the fields nobody validates
A Google Ads campaign name takes 256 characters with no documented content validation. Meta documents no maximum at all.
8 min read
What is a trusted data layer for AI agents?
Access, definition, freshness and provenance, decided before the question. Connection is solved; trust is what nobody shipped.
7 min read
Agents Schema and MCP: not competing standards, and the difference that matters
Fivetran's own FAQ answers 'Do I need MCP?' with 'No. MCP is one option.' They are layers, not rivals. The real difference is governance.
6 min read
Connecting an AI agent to production data without exposing it
MCP makes authorization optional, and 40.55% of live remote servers expose tools with none. Safety is what you add on top.
8 min read
Is server-side tagging worth it? Read both documents first
Google promises full durability on a subdomain. WebKit caps CNAME-resolved subdomain cookies at seven days. Both are current.
8 min read
Marketing and finance never close the same revenue number
Both numbers are right. The accounting standard recognises revenue on transfer of control; marketing measures at checkout.
8 min read
Why your reported ROAS exceeds actual revenue, mechanically
Google documents that other platforms take full credit for a conversion with other touchpoints. Nobody deduplicates across vendors.
8 min read
What is a data contract, and what does it enforce?
A data contract declares shape, guarantees and meaning. On Snowflake, BigQuery and Redshift, a declared primary key is enforced by nobody.
6 min read
Per-tenant isolation for MCP: giving access without giving away the warehouse
Row-level security fails at the role the agent connects with. Per-tenant isolation moves the boundary out of application code and into architecture.
8 min read
LGPD and AI: where your customer's data can live if you use a foreign LLM
To send customer data to an LLM outside Brazil there are two live legal paths, not three. Global corporate rules are not one of them.
6 min read
Query cost guardrails: what an AI agent can do to your warehouse bill
An AI agent can scan your whole warehouse without breaking a rule. Adding a LIMIT does not reduce what you are billed on non-clustered tables.
8 min read
What is a semantic layer? A field-by-field definition
A semantic layer is where your metric definitions live in a form a machine reads before it computes. Not a tool: a set of written decisions.
7 min read
Metric hallucination: why your AI reports a different number than your dashboard
Metric hallucination is AI computing correctly on a definition nobody wrote down. Supplying the definition raised accuracy from 45.5% to 67.7%.
7 min read
Hidden cost calculator: find out if your ROI is real or fantasy
Dashboard shows 5x ROI but the cash doesn't add up? Find the hidden costs and non-incremental sales that turn profit into an illusion.
10 min read
Regional Holdout: The Only Real MMM Validation Test
Regional holdout is the only test that separates an MMM that found real signal from one that memorized noise. Learn to design and run it.
8 min read
Welcome to the Precisian Blog
Our first article explains what to expect from the Precisian blog: technical insights on GA4, data quality, attribution, and analytics engineering.
3 min read