Human-centered design systems and product experiences
Core services
01UI/UX Design02Web Development03Software Development042D/3D Experiences05App Development06AI Agents & Automation07Website DesignTechnical services
Frontend Engineering
React DevelopmentNext.js DevelopmentWeb ApplicationsEnterprise DashboardsSaaS PlatformsBackend & Infrastructure
API DevelopmentCloud InfrastructureDevOps & CI/CDBackend SystemsCustom SoftwareIndustries we serve
Technology & Software
AI Advertising Analysis Agent Development
Where the budget went, explained
An agent that reads your advertising accounts with read access.
An agent that reads your advertising accounts with read access. It explains what changed and why it probably changed, drafts creative variants for testing, and proposes budget and targeting changes for the person who owns the budget to approve.
Ad accounts generate more data than anyone reads. A weekly report lists spend and results per campaign, and the interesting question, why did this one get worse, takes an analyst an afternoon. Platforms offer to optimise automatically, in their own interest. What a marketing lead wants is an independent reading: here is what changed, here is the likely reason, here is what I would test.
This page describes a workflow we propose to build as a custom project. It does not exist as software yet. It would reuse the evidence-first reporting and the approval steps from our demonstrations. Reading advertising platforms, the analysis itself and creative drafting are not built.
Six jobs it carries through
01Account reading
Reads campaign data with read-only access.
- Spend, results and costs by campaign
- Changes made in the account
- Audiences and placements
- Data from each platform kept separate
02Change explanation
Explains what moved.
- Compares periods fairly
- Separates volume, price and conversion
- Links a shift to an account change where one exists
- Says when the cause is unknown
03Creative review
Looks at how adverts are performing.
- Results per creative over time
- Signs of fatigue
- Messages that work by audience
- Gaps in what has been tried
04Variant drafting
Drafts new creative for testing.
- Copy variants within your brand guide
- One difference per variant
- Claims checked against what you approved
- Held for approval
05Change proposals
Proposes what to do.
- Budget shifts with the expected effect
- Tests worth running, and how long
- Things to pause
- The evidence for each
06Measurement notes
Is honest about what the data can show.
- Where platforms may double count
- Where tracking is incomplete
- What your own sales data says
- What cannot be concluded
How One Request Moves Through
An example of one case, in 6 steps. The same steps play in the picture at the top of this page.
Data read
The agent reads last week's results from two advertising platforms.
Change found
Cost per enquiry on one campaign rose sharply while spend stayed flat.
Cause examined
The click price was stable. The landing page changed on Tuesday, and conversions fell from that day.
Explanation given
The likely cause is the page change, stated with the dates and the data.
Proposal made
The agent proposes restoring the form position and testing two new headlines.
Owner decides
The marketing lead approves the test and declines a suggested budget shift.
Walk One Real Case Through It
Thirty minutes, one case from your own process. We map it against the workflow on this page and tell you plainly what can be reused from our demonstrations and what has to be built for your systems.
Six Things That Slow This Down
Reports without reasons
Cost per result rose, and nobody says why.
Platform advice taken on trust
The account follows recommendations that raise spend.
Changes with no record
A budget is moved and nobody remembers the reason.
Creative fatigue unnoticed
The same advert runs for months while its results decline.
Results counted twice
Two platforms claim the same sale.
Tests that prove nothing
Variants differ in five ways and run for two days.
It Reads and Proposes, the Budget Owner Decides
The agent connects to your advertising accounts with read access and has no ability to change a budget, a bid or an audience. Every recommendation is a proposal with its evidence, for the person who owns the money.
It says what the data cannot show. Results reported by a platform are labelled as the platform's own figures, and where two platforms may be counting the same sale it says so.
- Read-only access to ad accounts
- No spending authority
- Likely causes labelled as likely
- Platform figures labelled as such
- Creative claims checked
- Proposals approved by a person
Where It Connects
The systems named here are the kind this agent works with. Which ones it connects to in your case is settled in the scope.
Ad platforms
- Google Ads
- Meta
- and others you use, with read access
Analytics
- Your web analytics and conversion tracking
Sales data
- CRM or shop results, to check platform claims
Brand guide
- What your adverts may say
Change log
- What was changed in each account, and when
Budget owner
- The person who approves spend
Asked Before the First Call
No. It has read access only. It proposes changes with the reasoning, and a person makes them. If you later want certain safe actions automated, such as pausing an advert that links to a broken page, that is a separate decision with its own limits.
Reporting combines all your channels into one view of what happened. This agent goes deep into paid advertising only: why a campaign changed, what to test, where the platforms' numbers disagree with your own.
It gives you an independent reading alongside it. Platform automation optimises for the goal you set inside that platform. This agent compares across platforms and against your own sales data, and tells you when the two tell different stories.
Not as software yet. This page describes a proposed workflow. It would reuse the evidence-first reporting and the approval steps from our demonstrations. Reading advertising platforms, the analysis itself and creative drafting are not built. A project starts by mapping your process, then proving the first steps on test data before anything touches a live system.
Give It Read Access to One Account
We will show what it would have said about last month, and you can compare that with what your team concluded.
