Pilot or an MCP connector: a comparison, with our production figures and their limits

Plugging Claude or ChatGPT into your ad accounts through an MCP connector costs little more than a subscription. Pilot, the tool we build, starts at €225 a month excl. VAT. Our production data only measures Pilot, so the comparison is about how each one works.

Take an ordinary Monday-morning question: "What's my CPA on Meta this month?" By hand, you open Ads Manager, filter the date range, recalculate and paste the result into a spreadsheet. We put that at around twenty minutes. In Pilot, a numbers answer takes 8 seconds, according to the 8 September snapshot published on Pilot's Impact page. For a general-purpose assistant linked to Meta through an MCP connector, we have no measurement to compare.

On a single question, then, nothing allows us to claim Pilot does better. The gap opens elsewhere.

Where the numbers come from

An MCP connector queries the ad platform's API at the moment you ask. Pilot reads its own database, loaded every night, platform by platform, with the same columns for Google Ads, Meta and TikTok.

As long as you work on one platform, the difference is hard to see. It shows when the question spans several. With connectors, the assistant assembles answers in the conversation that each arrive in their own format, with their own definitions of a conversion and of the attribution window. Nothing guarantees that Meta's "cost per purchase" and Google Ads' "CPA" are reconciled the same way from one question to the next. In Pilot, this arrangement into the same columns happens once, at collection time; it does not, however, make comparable conversions that each platform counts under its own attribution rules.

The reverse holds too. To follow the spend pace of a sales day hour by hour, a connector querying the API live has the edge over a database loaded overnight.

Business context works the same way. A connector gives access to the data, not to your goals: if you are aiming for a cost per purchase of €35 on Meta, you have to restate it in every conversation. Pilot keeps it in its Strategy module and uses it to read each result.

When nobody is looking

A connector does nothing on its own. It answers when the assistant calls it, which means when someone asks a question.

Pilot collects the accounts every night and runs scheduled reports and checks at the set time, in read-only mode. By 8 September, 9,014 collection passes had run without anyone involved. Every night, once collection is done, Pilot also runs 106 checks per platform on each account and leaves its proposals for the morning. A budget drift is then flagged in under 24 hours. Without automatic checks, we estimate it takes more like three to seven days to spot, sometimes only once the monthly report arrives.

In our view, this is the most useful gap in practice. A spend drift costs in proportion to how long it lasts before someone opens the right view, and an assistant you only query when you think of it does not shorten that delay.

Who changes the campaign

Pilot reads and analyses freely. For any change (budget, bid, status, audience), it prepares an approval card, and nothing is sent to the platform until someone clicks "Approve". By 8 September, teams had approved 317 of Pilot's 386 proposals, or 82.1%. The other 69 had been rejected, postponed or were still to be decided.

Every rejection is a change that would have gone out unreviewed if Pilot applied its proposals on its own. The reverse needs watching too: at over 80% approvals, the click can become a reflex.

On the connector side, it all depends on the connector and the assistant. Some are read-only, others can write to the account, with or without confirmation. We do not recommend giving write access to a general-purpose assistant before checking exactly what it can change and what it asks you before doing so.

Our numbers have limits

The multipliers shown on Pilot's Impact page (a multi-platform dashboard built 43 times faster, for instance) compare two different kinds of figure. The "with Pilot" durations are timed on real use by the Braike agency and its clients, from 3 March to 8 September. The "by hand" durations are industry orders of magnitude that we estimated. The ratio gives a sense of the gap; it should not be lifted as is into a business case.

The scope is the Braike agency and its clients: connected accounts exclude the 41 demo workspaces, but activity figures cover the whole platform. They say nothing about the number of billed clients.

Finally, none of these figures is about advertising performance. Saving time on a dashboard does not lower a CPA. That time only has value if it goes back into the decisions the dashboard makes possible.

Pricing, account by account

Pilot costs €225 excl. VAT a month with one ad account included, then a per-account price that falls with volume. The pricing page gives the example of six accounts at €422.50 excl. VAT a month. The AI model is included and the price does not depend on the number of users.

An MCP connector costs your assistant subscription, with its usage limits, plus the connector if it is paid. For an advertiser running a single Google Ads account and asking a few questions a week, it is the cheapest option, and we have no hesitation recommending it.

The maths changes for an agency following several clients across three platforms, or an advertiser who wants an alert before the monthly report. There, nightly collection, stored goals and a record of every approval justify a dedicated tool.

The simplest route is to test. During Pilot's seven-day free trial, ask three identical questions to Pilot and to your connected assistant, over the same period, and check both answers against the platform's own figures. Our comparison of AI tool categories lists the other options worth adding to that test.