When AI Optimises the Wrong Thing: The Hidden Cost of Automated Growth
A founder told me he did not trust the black-box bidding on Meta and Google. “It only works for companies with big budgets,” he said. “I’ve never seen good performance.”
I asked one question. Did the algorithm learn to hit the business outcome you set?
He paused. “Business outcome? We set the objective to traffic. We want as many users as possible.”
There it was. He blamed the machine. The machine did exactly what he asked. It bought traffic: cheap, abundant, and commercially useless.
This is the quiet failure inside automated growth. AI optimises the objective you set, not the outcome you want. The platform numbers look strong while the commercial results rot underneath.
AI does not misfire. It aims with terrifying precision at the wrong target.
This article argues that automated growth fails in three predictable ways: proxy metric optimisation, short-window attribution, and creative homogenisation. Each one produces a better dashboard and a worse business. Spotting all three is the difference between commanding the machine and being fooled by it.
1. The Algorithm Optimises Your Objective, Not Your Intent
The model has no idea what you actually want. It only knows the number you handed it.
Set traffic, and it finds the cheapest possible clicks. Set leads, and it finds form-fillers who never buy. The objective becomes a proxy for the outcome, and the proxy is always a poor stand-in.
Amazon learned this at scale. It built an experimental hiring AI trained on past hires. The model learned to penalise resumes that mentioned women, then Amazon scrapped the tool after it showed bias against women.
The system optimised for resemblance to historical hires. Not for future talent. It did its job perfectly. The job was wrong.
💡 Key Takeaway: The objective you type into the platform is a guess about your real goal. AI treats that guess as gospel.
2. The Proxy Improves While the Business Decays
The most dangerous misalignment looks like success. Costs fall. Efficiency climbs. The dashboard glows green.
Klarna is the cautionary case. The company promoted its AI assistant as doing the work of 700 agents. Then it reversed course and rehired human support after service quality dropped.
Klarna optimised hard for cost reduction. The customer experience outcome deteriorated underneath the savings.
Think about a bodybuilder who optimises only for scale weight. He gains 10kg fast. The number on the scale improves. He also adds fat, loses conditioning, and hurts his performance. The metric got better. The outcome got worse.
3. Short Attribution Windows Reward the Wrong Channels
Most ad platforms optimise inside a short conversion window. They credit the last cheap click before purchase.
This rewards channels that harvest demand you already created. It starves the channels that build it. Retargeting and brand search look brilliant. The top-of-funnel work that filled the pipeline looks weak and gets cut.
The algorithm is not lying. It reports exactly what it measured. It simply cannot see beyond the window you gave it. So it defunds your future to flatter your present.
If your customer takes weeks to decide, a seven-day window is not measuring growth. It is measuring the finish line and ignoring the race.
4. Creative Homogenisation Erases Your Positioning
AI-generated creative does not just save time. It quietly flattens difference.
A 2025 meta-analysis of 28 studies found that people using generative AI produced more creative outputs alone, with a positive effect on quality. The same analysis found a sharp negative effect on idea diversity.
Two more studies confirm the pattern:
- A longitudinal study found that AI-assisted users produced increasingly homogenised ideas, and the sameness persisted even after AI was removed.
- Researchers found that LLM-assisted ideas became less semantically distinct from each other, even as volume rose.
For a brand competing on positioning, this is the worst possible trade. AI lifts average quality and erases your edge. You get more ads that look like everyone else’s.
5. Scale Turns a Small Error Into a Structural One
A misaligned objective run by one marketer is a mistake. The same objective run by AI across millions of decisions is a system failure.
Reuters reports that more than 80% of US employers use AI hiring tools, with lawsuits now emerging over algorithmic bias. Workday faces a sprawling case over whether its screening tools discriminated.
Automation multiplies whatever objective you feed it. A small misalignment becomes a structural one at scale.
Final Thoughts: AI Is a Mirror, So Check What You Point It At
The sharp objection here is fair. AI is not the problem. Leadership is. The model only optimises the KPI a human chose.
I agree completely. That is the entire point.
AI is a mirror. It reflects your measurement architecture back at you, faster and harder than any human ever could. If you point it at a proxy metric, it will chase that proxy off a cliff and call it performance.
The fix is not less automation. The fix is better objectives. Define the commercial outcome before you define the platform metric. Set the window to match your real sales cycle. Protect creative difference on purpose.
If your dashboards are green but revenue is flat, you have an objective problem, not an AI problem. That is a growth system problem, and it is exactly what I fix.
Book a discovery call, or connect with me on LinkedIn and tell me the one objective your platform is optimising right now.
A note before you close this tab. The fact that you read this far tells me something. You already sense that the way you’ve been thinking about growth might be incomplete. That instinct is worth following.
Mervyn Chua is a growth-transformation consultant helping founders and CEOs build the strategic clarity and systems to grow in an AI-first world. If this raises questions worth exploring for your brand, let’s talk.
