Stop Asking Google Ads to Guess Your Business Model

Why every under-fed ad account is running a theory of your business that nobody wrote, nobody approved and nobody checked

If you’ve ever tried to escape an escape room, you’ll have a decent handle on this blog’s central argument. There are plenty of red herrings, and you can spend a long time trying to solve a clue that ultimately serves no purpose.

Paid media accounts can end up in a similar situation. Give an ad platform a conversion goal and a target, and it will use the signals available to work out what your business values. It builds a working model of what a valuable outcome looks like, then optimises towards it with considerable efficiency.

The problem is that it is optimising towards its best guess of your business model.

That guess can be surprisingly far from reality. The divergence is usually quiet and gradual, and often doesn’t become obvious until the numbers that really matter – margin, repeat rate and actual profit – start telling a different story from the dashboard.

What the Platform Sees When You Tell It Nothing

An ad platform only has access to the signals you give it. It sees a conversion, a form submission or a transaction and its associated value. From there, it has to infer everything else: which customers to pursue, which audiences resemble your best buyers and how much a particular click is really worth.

If all conversions look the same, the platform has no way of knowing that one form fill represents a £40,000 client while another is unlikely to become a customer at all. It treats them as equivalent signals and optimises towards whichever is easiest to generate at scale.

That can create some very efficient-looking wrong answers.

Consider an eCommerce business selling both high-margin own-brand products and lower-margin third-party stock. If the platform receives a single purchase conversion with revenue value, it can optimise effectively for that revenue. Over time, it may learn that the third-party products generate more conversions at a lower CPA, and naturally shift budget towards them.

Revenue can look healthy. ROAS can remain strong. But if those sales carry significantly less margin, the business can end up generating more of the wrong revenue.

The platform hasn’t made a mistake. It has done exactly what it was given the information to do.

The Platform’s Theory Hardens Over Time

Once the platform has built a model around your business, every optimisation decision reinforces it.

It finds more of what it thinks you want. That generates more data confirming that this is what you want. The additional data then makes the model more confident, and the cycle continues.

A flawed signal therefore doesn’t necessarily self-correct. It can become increasingly entrenched.

This is one of the less obvious problems with automated optimisation. The platform can become extremely good at pursuing an objective that was never quite the objective you intended to give it.

And the longer that runs, the harder it can be to untangle.

You Already Have What the Platform Is Guessing At

The information the platform is missing usually isn’t exotic. It already exists somewhere inside the business.

You know which customers are profitable. You know which products carry the strongest margin. Your sales team knows which leads actually turn into customers. Your CRM knows what those customers are worth over time, rather than simply at the point of conversion.

The platform is trying to infer information that your business already knows.

The opportunity is to turn that knowledge into usable signals. Value-based conversion data, first-party data connections, CRM integrations, offline conversion tracking and other forms of enriched data make it increasingly possible to give ad platforms a much more accurate picture of what success actually means.

The challenge is less about whether the technology exists and more about whether the business is prepared to use it.

What Changes When You Stop Letting It Guess

When an ad platform receives real business value rather than proxy signals, its model of the business can start to align much more closely with reality.

Instead of simply looking for cheap conversions, it can optimise towards the customers, products or outcomes that actually contribute to growth.

This is also where agentic systems can add another layer. An agent with access to commercial context can help maintain the connection between business priorities, first-party data and the signals being fed back into the platforms. Rather than treating measurement as a one-off implementation, it can become an ongoing process of checking whether the optimisation inputs still reflect what the business values.

The change is often quieter than expected. The platform doesn’t suddenly discover a completely new way of working. It simply has better information about what it is supposed to work towards.

Budget that was previously being absorbed by low-value conversions can begin moving towards higher-value outcomes. Over time, that can mean the account is not just becoming more efficient at generating conversions, but more effective at generating the outcomes the business actually needs.

The Theory You Never Chose

Somewhere in every under-fed account, there is a theory of the business that nobody wrote, nobody approved and nobody checked.

It has been assembled by a machine from whatever signals were available, then pursued with total conviction.

That is the real risk of leaving an ad platform to infer your business model. It isn’t necessarily going to make obviously bad decisions. It can make perfectly rational decisions based on an incomplete understanding of what you value.

Much like the red herring in an escape room, the problem is that you may not realise you’re solving the wrong puzzle until a lot of time and money has already been spent.

The better approach is to stop asking the platform to guess.

If your business knows what a valuable customer, product or outcome looks like, your paid media should have access to that same information.