The Always-On Analyst · AI data setup for your marketing

Your marketing data. One checked spreadsheet.

Your agency says ROAS is great. Your finance team says profit is down. Connect your marketing platforms to one spreadsheet. Ask Claude or ChatGPT for the numbers, and it builds the table or chart. Every source and calculation stays visible. A secure read-only setup under your control keeps API keys off people's laptops. Missing data is flagged instead of guessed.

Your team can get and check answers without waiting for one analyst. It works with the AI subscriptions you already pay for. Start with marketing, then add finance, HR, operations and sales on the same setup.

  • Free 30-minute call
  • Your hardest question, answered free
  • Open-source version on the call

Best for companies running five or more marketing platforms, with at least two people who need the numbers every week. Built from an internal system used across 8 agencies and 113 workspaces, which together run over $50M a year in ad spend, graded on margin. See the case study.

Two minutes 42 seconds. The presenter and voice are AI-generated; the figures on screen are illustrative.

The offer in four points

  • Outcome: Ask a marketing question. You get a checked Google Sheets report with the source data and calculations visible.
  • Access: A secure read-only connection under your control keeps API keys out of people's hands. Claude or ChatGPT uses it to read only the data you allow.
  • Full setup: €4,900 and two to three weeks. Your marketing sources connected on your infrastructure, one agreed scheduled report, and two people on your team trained. If the agreed report does not run on your data, you get your money back.
  • First call: A free 30-minute call. Bring your hardest question or pick one from our list. You get the open-source version on the call. Connect read-only after it, and the answer arrives free within 3 working days.
Book a free 30-minute call →

No preparation or slides. Bring your CTO if useful. You leave with the open-source version and an agreed question. The answer follows within 3 working days of connecting.

The problem

Your data is ready. Your team still needs a safe way to use it.

Your marketing data sits in six to eight systems. Getting one complete answer usually means sharing a key, inviting another user or pasting files into a chat.

01

Keys go out and never come back

Nobody knows how many there are or where they live. When someone leaves, their access does not: the key lives on in a script and in somebody else's browser.

02

To get AI to count anything, someone pastes the data into a chat

That is exactly the leak you wanted to avoid, done by hand, with no trace in any log.

03

There is no middle step in permissions

A junior or an outside contractor gets full rights, because the alternative is to give nothing and do everything yourself.

Where it lands

Ask a question in Claude. The answer is a sheet, a chart or a Slack message.

The core setup is the same for every company. What lands in the spreadsheet is not. Pick what you run and see what gets connected and what comes out.

Usually connected:Google Ads · Merchant Center · GA4 · Search Console · Meta Ads · Shopify
Also:Amazon Seller Central · Klaviyo · Stripe · ShipStation or your 3PL · Trustpilot · Gorgias · call tracking for phone orders

Product feed health

Disapproved products show no ads and nobody reports it. A daily list of disapprovals with the reason and the SKU, instead of a look at the panel once a quarter.

Wasted ad spend

Queries without conversions, campaigns stuck in learning, exclusion conflicts. The report names the amount and the campaign.

Search and AI visibility

Search Console and Bing Webmaster Tools together. Bing matters because it feeds the AI assistants your customers increasingly ask about products.

Margin after returns and ad cost

Ad cost next to returns and category margin. A campaign with a good ROAS and 40 percent returns stops looking like a win.

How it works

Three stops, one key, no keys in anyone's hands.

Every request follows the same route. We call the secure middle layer the gateway. It holds the keys, checks access and never returns a key in the answer.

Stop 1 · the asker

An agent or a person

Claude, ChatGPT, Codex, a spreadsheet or a script sends an ordinary HTTP request. It authenticates as a member of your organisation, not with a key to Google.

Stop 2 · the secure connection

The secure layer on your account

Everything that matters happens here:

  • checks identity and refuses without it
  • adds the key on the server side
  • lets only the allowed operations through
  • records who asked for what

Stop 3 · the source

Google, Meta, your shop

The source only sees the secure connection. If one account is unavailable, it tries the next one and shows you exactly what happened.

See a real permission error
gateway response · a request for ad data·example
{
  "error": "Ad account owner has NOT granted ads_read permission",
  "_tried": [
    { "source": "ads-main",    "status": 403, "ms": 215 },
    { "source": "ads-eu",      "status": 403, "ms": 237 },
    { "source": "ads-partner", "status": 403, "ms": 332 }
  ]
}

Why this matters to your team. The gateway reports every attempt. A chain that shows only the last error points at the least likely culprit, and someone spends a week fixing an account that works. Here you see it at once: the problem is a permission on the ad account owner's side, in all three sources, in a quarter of a second.

What it looks like

Three questions, and what came back.

There is no dashboard to learn and no new login for your team. You ask in the chat window they already have open, the agent reads only what you allowed it to read, and the answer lands where the work happens. Two of the three questions below need a source that no reporting tool connects on its own, and one of them nobody asks twice because the agent keeps doing it.

Claude · your workspaceexample, not a client account

Head of growth asks

Google Ads says these three campaigns lose money. Check the phone calls before I switch them off.

What the agent was allowed to read, and nothing else

googleads.cost_by_campaign       last 90 days     0.6 s
calltracking.calls_by_campaign   last 90 days     0.4 s
crm.deals_won_by_source          last 90 days     0.5 s
What the agent answered, as an example
CampaignSpendFormsCallsWon
Service · exact€4,1206419
Service · broad€3,3801172
Brand€910142212

The campaign that looked worst brings most of the work. Calls never reach the ad statistics, so a campaign that pays for itself reads as a loss. Switching it off is the expensive move, and no advertising platform would have warned you.

Where it landed: Google Sheets, tab “Paid · weekly” · Slack #marketing

The figures inside the previews are invented, so nothing here claims a result. What is real is the shape: which sources a question touches, how long the gateway takes to answer it, and the fact that the answer arrives in a sheet or a channel where your team already works.

Work that shipped

Mega-Context Sandbox: project knowledge that loads before the first question.

The same infrastructure now keeps client context, permissions, skills and live data organised across an agency group. A new person or AI session starts from what the project already knows, while each client remains in a separate repository with separate access.

AI & Automation · Internal rollout · System read 25 Aug 2026

The Mega-Context Sandbox

One repository per client. The same layout, rules and safeguards everywhere. Context loads automatically, connected data stays current, and credentials remain on the gateway.

113
sandboxes with the same structure
8
agencies in the rollout
27
data sources through two gateways
42
people working in the sandboxes
Human setup time for one new project
  • Before · rebuilt for each project3.5 to 7 h
  • Now · repository from template5 to 10 min

The before-and-after range comes from enumerated setup steps and team experience. It is an estimate, because no stopwatch baseline was captured before the rollout.

What one session starts with

  • About 28,800 characters of project context load automatically.
  • The model sees which sources are connected, stale or missing before it answers.
  • One centrally maintained skill or security fix can reach every sandbox.

Read the full case study →

This is an internal MTA Group rollout built by the team behind this offer. Zero Fluff Digital is part of MTA Group. Every scale figure comes from the system read dated 25 August 2026 in the case study linked above, except the ad spend, which is the combined yearly spend run by all agencies in our group; the setup-time comparison is explicitly an estimate.

The offer

Start with one working report. Expand after you have seen it run.

Before work starts, you agree the sources, questions and acceptance test in writing. You decide whether to connect more of the company only after the report passes.

Step 1 · your marketing

Full marketing setup

€4,900

Two to three weeks, paid once

Money back if the agreed report does not run

The setup passes when the scheduled report runs on your data and returns the agreed outputs. If it does not, you get your money back.

We install for three companies a month, one engineer per install, start to finish. The free call has no limit.

Your marketing data in one spreadsheet, without handing API keys to employees or external partners. Your team can use and check the setup without us.

  • A secure read-only connection under your control
  • The marketing sources agreed for the setup, connected read-only: ads, analytics, shop, feed, CRM
  • One scheduled report in Google Sheets or Excel, with the questions and acceptance test agreed before work starts
  • Two people trained to use, check and manage the setup
  • A kickoff call to agree sources, questions and the acceptance test in writing
  • A one-hour review of what the numbers say, with whoever owns the budget
See what can happen after the setup

Step 2 · the rest of your company

Implementation

Four to eight weeks, quoted after the setup

Finance, HR, operations and sales on the same setup. Your company works on it every day, without a consultant inside every task.

  • Three to five more sources per area, each area seeing its own slice
  • A separate list of allowed write actions
  • Permissions assigned by area across the whole team
  • Written procedures and training
  • Later: ClickUp, Jira and meeting transcripts, so each week you see if your team and your agency still work toward your goal
  • A data warehouse, BigQuery or Snowflake, if you want one. The vendor connectors then read from it

Step 3

Care

Monthly, quoted after a call

Platforms change their APIs and access rules. Care keeps the connections working.

  • Maintenance and response to vendor changes
  • New sources as you need them
  • A review of permissions and costs
  • Supervision of background processes
What the price does not cover
AI model usage on your subscription, third-party tools, or work done by another agency. Anything outside the scope above is quoted before it starts, never added afterwards.
What access looks like
Reader roles on your platforms are granted to one secure connection. Individual people do not receive the API keys. Before you connect, we sign a mutual NDA and a data processing agreement. Both can use your template or ours.

Next step

Book a free 30-minute call

Bring your hardest question, or pick one from the list below. After the call you get a read-only access link. Within 3 working days of connecting, you get the answer for free, in a spreadsheet you can check formula by formula. On the call you also get the open-source version, with a guide to run it in your Claude or ChatGPT.

Questions you can pick from

  1. Which campaigns still make money after returns, discounts and product costs?
  2. Which products do your ads sell at a loss once ad spend is counted?
  3. Where do Meta, Google and your shop disagree on sales, and by how much?
  4. What does a new customer cost you in each channel, after returns?
  5. Which customers come back to buy again, and which channel first brought them in?
  6. How much of last month's spend went to search terms or audiences that never sold?
  7. In which weeks did spend go up while profit went down, and what changed?
  8. Which best-selling products get little or no ad budget?

Or open the calendar in a new tab.

Prefer to ask in writing?

Ask in writing

Send us your question and what you run. Within one working day you hear back with a time for the free 30-minute call.

You can also book the free call directly. Thirty minutes, no slides.

Optional

By sending this you agree to our privacy policy and terms of service. We use your details to answer you, nothing else.

The first 30-minute call is free and comes before any quote. After the call, connect read-only and your hardest question is answered free within 3 working days.

Review the open-source version before you decide

Open source

Get access to the open-source version.

Want to look first? Get repository access now. On the free call we walk you through running it in your Claude or ChatGPT. The setup runs on your account, in your repository, with your variables. Handover means removing our access. The four modules are open code, so you can review them before you sign anything.

Fill in the code access form. Your repository invitation will arrive by email within one working day.

Free setup guide: connect Meta Ads, Google Ads and Merchant Center to Claude or ChatGPT, read only. Open the guide. The full Gateway, with ready routes for your ad platforms, comes free after the call.

megacontext-gateway
read from many sources, control writes
megacontext-guard
rules an agent cannot skip
megacontext-workspace
a project skeleton with durable context
megacontext-fleet
one change across many projects

Before you sign anything

Questions to answer before you start.

Short answers, so you can pass them on without us in the room.

What stays when we end the engagement?

Everything: the setup, code, documentation and reports. They run on your account, so your team can maintain them or give them to another partner.

How does this handle GDPR and AI training?

The setup uses business AI plans that do not train on your data and providers covered by a data processing agreement. Each person reads only the data allowed for their role, and every request is recorded.

We already have an agency. Does this replace them?

Your agency keeps running the work. Their report sits next to your shop, CRM, finance and other sources, so you can check what is driving the result.

Can AI change campaigns, landing pages or creative?

The setup starts read-only. Advanced teams can later allow a clear list of actions. Every action is recorded, and a person approves high-risk changes.

What happens when a platform changes its API?

Clear errors show what broke. Two people learn to use and check the setup, and one becomes its owner for common maintenance. You can also ask us to maintain the connections.

Will it work with our own systems?

If a system has an API, database view or regular export, it can usually be connected. You hear what is possible during the free call, before you pay for any work.