AI in practice

AI at SANE/REBELS · 03

Our AIs speak Google Ads

Paid media in a checked loop: data-driven, every live change approved by a human.

On this page

The first principle

Why this setup

Three truths that everything else follows from.

01

We don't beat Google's algorithm. We feed it. Today Google's Smart Bidding prices the auction better than we could by hand, so we feed it real signal rather than fight it: your real margins, your real new customers, your goals. We set the limits, and if a smarter way shows up, we take it.

02

Where there's a right answer, code does the math, not AI. Numbers, checks, and carrying out a change are handled by tested code. We use AI only where there's no clear right answer: which idea, which copy, what comes first.

03

AI has become cheap. Good judgment hasn't. AI produces analyses and copy in minutes. The decision is what stays valuable, and a human makes it. In a controlled trial, experienced developers with AI even came out 19 percent slower while feeling faster, with most of the time going to reviewing and correcting the output.

Here's what that looks like in practice.

How we use AI

What actually optimizes your account

You don't optimize an account in one go. Four things move the numbers, and we're careful about which does what. Anything with a right answer runs as code. Anything that takes judgement stays with us, the same experienced hands that have always run accounts. The AI sits in between, reading and drafting so we move faster. It's the smallest part of this.

Platform

Runs the auction. Smart Bidding and Advantage+ price every impression in real time, faster than anyone could by hand. We leave the bids to them and work on the signals and limits they optimize against.

Code

Does the work that has a right answer. The math, the checks, the budget steps, the push to the account. It runs the same way every time.

AI

Reads and drafts. It works through the search terms, writes first versions of copy and structure, and explains what it finds. Everything it produces is a suggestion, and one of us decides what to do with it.

SANE/REBELS

Make the calls. We run your account the way we always have, on judgement from a lot of years and a lot of accounts. The AI hands us reads and drafts; what actually changes is ours to decide.

A few things keep the whole thing honest. Bids move on rules you can read in plain language. Budgets step up slowly, and nothing scales until your tracking has been clean for two weeks. Every change gets written down and can be undone, so you see exactly what happened and can reverse it in one step.

Cleaning up negatives and splitting brand from non-brand is housekeeping, the kind any decent operator does. The calls that actually move a business, how the account is built and where the next euro goes, come from people who have run a lot of accounts. The AI speeds those people up. It doesn't replace their judgement.

The model

Two things we don't control: how Google's and Meta's algorithms optimize, and how your business actually runs, your margins, your goals. Our tool sits in between. It gives the algorithm the right signals and limits, and turns what comes back into a clear decision brief.

UnderstandRead · DiagnoseActPrepare · ApplyVerifyApprovalhuman-in-the-loop

Between understand and act sits the approval. Nothing goes live without a human approving it, and a change counts as done only once a follow-up check confirms it.

Onboarding

Context, context, context.

An AI is only as good as its context. You give us that once, as your Brand Hub profile: the things no platform knows, your margins, your goals, your brand.

You give us once

Your Brand Hub profile

Mandate. How success is measured.

Economics. Margins, contribution margin, targets.

Brand. Voice, colors, banned words, protected terms.

Conversion truth. What counts as a conversion.

Approval cadence. How you want to decide.

The tool pulls on its own

Everything measurable

Campaigns, budgets, bidding strategy, status

Impression share, quality score, cost, conversions

Keywords, asset groups, feed

History: spend, ROAS, conversions

What we can read, we never ask you for.

The Brand Hub profile is your fixed context with us: reviewed and versioned. Every recommendation traces back to it. Change the profile and what the tool proposes changes. That's point 1 from above: you give the AI the right signals.

What the tool does

Four phases, one loop. This is what happens in each phase, not how it computes in detail.

01

Understand

read & diagnose

Audit across both platforms. One run reads Google and Meta together, traceable to the source.

Check the measurement first. If conversion tracking is off, the tool proposes nothing. Optimizing on broken measurement wastes the most budget.

Intent mining. Search terms become negatives and new keywords, brand and generic kept separate.

Portfolio against mandate. Spend and return per channel against the agreed target mix.

02

Prepare

draft & check

Brand-bound creatives. Copy and image in brand colors and voice, grounded in what already wins in the account.

Campaign drafts. Search and PMax as checked proposals.

Pre-launch check. Character limits, forbidden characters, AI-typical filler, and your banned words stop a text before it reaches the proposal list.

One list. Every proposal lands in one place, each with a rationale, nothing scattered.

03

Apply

approve & execute

Execution with approval. Google changes go live only once a human approves them, and a follow-up check confirms they actually landed.

Meta creative kit. Approved creatives as a validated bulk import; the final upload stays deliberately manual.

Platform limits respected. Bid values belong to Smart Bidding and Advantage+, we don't write them. Budget steps stay under the learning-phase thresholds.

Value bidding where the data supports it. When margin per segment is available, we steer toward contribution margin instead of first purchase.

04

Optimize

measure & learn

Attribute the impact. A tagged change gets a verdict against the account trend after 7, 14, and 30 days.

Quarterly report. The KPIs you defined, with the change and a short memo on it. Then the loop closes.

Three controls keep the loop safe: check the measurement first (nothing on broken tracking), then human approval (no live change without a human confirming the exact change), then the follow-up check (done only once it's confirmed).

In detail

How we use AI, area by area

The same loop, four areas. For each, we say honestly what works and what doesn't.

Foundation · Data analysis

Analytics, shop, first-party

We connect, read-only, your analytics and your shop: orders, cost and margin, new versus returning customers, customer value. That builds a picture of real profit, across every channel, not the ROAS number each platform credits to itself. That's the signal the algorithm then bids on.

Live · Google Ads

The full loop

Audit, search terms into negatives and new keywords, structure, budgets, choice of bidding strategy, RSA copy, PMax fed. Changes go live only once a human approves them, a follow-up check confirms them. We don't set the bids ourselves, Smart Bidding does that, we set the target.

Live · Meta Ads

Read, create, export

Read the numbers, generate creatives in your brand (copy and image) and check them against brand and platform rules. Delivery runs through a checked import file, the final upload stays manual until Meta opens the direct interface. Image and video never go live automatically.

Roadmap · Microsoft Ads

Coming soon

Microsoft now runs search ads through Bing and Copilot with AI Max. We'll bring Microsoft into the same loop when it's time. Today Google and Meta run in production.

How we build this

The system is built so the safe thing happens by default, even on a busy day.

Architecture

Deterministic core, AI as the shell

The AI reads the account and proposes, with the reasoning shown. It never decides a live change. Every calculation, and the decision about what actually changes in the account, is tested code.

Governance

Every output is a proposal or an explanation

What the AI produces is either display-only or a proposal that has to pass the check and a human approval before it touches an account. Creatives included.

Engineering

Agentic engineering, not vibe coding

Our own developers review the systems. The building is the quick part. Most of the work is hardening it against the edge cases, and that's what we test for.

See it on a real example?

30 minutes, the whole loop live. You see what gets proposed, what a human approves, and what lands in the end.

Request a call

Related content

Last updated June 2026

Citation rule: if you use or build on these ideas, credit SANE/REBELS (KNUS GmbH) and link to sanerebels.com/about/ai.