06.07.2026
When ROAS dips, DON’T PANIC, trust the process.
Do you lean into AI for business advice? Let’s be honest, who doesn’t? These days many businesses are using AI as their unpaid business advisor, steering strategy, analysing data and removing the burden of decision-making. With Meta marketing in particular, advice is now much easier to generate but this brings with it hidden risk – the resulting advice is not easier to judge. And this distinction matters, especially for leisure businesses using Meta to drive bookings, local awareness and stronger campaign performance.
When a Meta campaign is performing well, the relationship between the business, the agency and the ad platform often feels straightforward. The ads are running, bookings are coming in and ROAS looks healthy. The strategy is trusted because the numbers are reassuring. Then the market changes.
A hot week keeps families outdoors. A school holiday ends. Household budgets tighten. Local demand reduces or a competitor launches an offer. Footfall drops and the dashboard starts to look uncomfortable. At that point, the campaign data often becomes the focus of anxiety.
It is completely understandable. For an owner-operated leisure business, these are not abstract numbers on a report. They are staff hours, rent, cash flow and quiet sessions that could have been busier. But this is also the moment when campaigns are most at risk of being damaged by over-correction.
The problem with AI
AI can be a useful campaign tool. It can help generate ideas, explore copy routes, summarise reports and challenge thinking. Used well it can help clients ask better questions. Used badly, it creates unrealistic expectations.
The problem is context, or more importantly the lack of context. A client might ask Chat or Claude how to improve a campaign after they see a drop in ROAS, and the AI-generated response recommends more segmentation, new audiences, fresh ad sets, revised budgets, new creative angles or even a different campaign structure.
Some of those recommendations *may* be valid. But they may not be valid now. They may not be valid for that account. They may not even be valid with that budget, that learning status, that booking cycle, that pixel data, that seasonality or that commercial objective. AI often imagines a possibility and turns that into a promise.
A strategist might say, “This could be worth testing, but only if the data supports it.” AI presents the same idea as a confident recommendation. And that is a dangerous shift.
When a campaign is under pressure, a nervous business owner does not need every theoretical lever listed in order of plausibility. They need calm, expert advice about which levers are worth testing, which should be left alone and which could make performance worse.
Meta needs time to learn
Meta advertising is not a simple manual media-buying system. Its delivery is increasingly driven by machine learning, conversion signals and automated optimisation. The platform is continually trying to understand who is most likely to act, which creative is most likely to work, which placements are most useful and where the next conversion may come from. To do that well, it needs signal and stability and that is why constant change can be counterproductive.
Splitting audiences into smaller groups may feel like more control but it can fragment the data the campaign needs to learn. Rebuilding ad sets may look proactive but it interrupts delivery. And moving budgets too frequently makes it harder to distinguish a real trend from short-term noise.
Consequently, good campaign management does not mean leaving everything untouched but it does mean knowing when to act and when to give the system time to gather enough evidence.
In performance marketing, doing something is not automatically better than doing nothing.
Seasonal ROAS needs seasonal expectations
Leisure businesses are of course seasonal by nature. School holidays, bank holidays, weather, payday cycles, local events, Christmas, Easter, summer breaks and family routines all affect demand. Peak periods produce stronger ROAS because people are already looking for things to do. The campaign is capturing demand as well as creating it. Quieter periods work differently.
A lower ROAS outside of peak trading does not automatically mean the campaign is failing. It means the market is softer, intent is lower or the offer needs to work harder. Paid social can support demand but it cannot remove seasonality from the business.
This is why ROAS must be judged in context. A campaign should not just be reviewed against the trading calendar, the equivalent seasonal period or the offer, but also against the weather, local competition, average order value, margin, landing page performance and booking behaviour. It is important to understand that peak-season ROAS is not a permanent benchmark. It is a seasonal high-water mark. For businesses with seasonal demand, campaigns will always have seasonal ROAS, so the skill is to manage that intelligently, not panic every time the curve behaves like a curve.
Paid social cannot replace a business plan
A Meta campaign can do a great deal – increase visibility, drive traffic, support offers, retarget warm audiences, promote events, stimulate bookings and help a venue stay front of mind. But it cannot solve everything.
It cannot guarantee footfall every day, it cannot make every quiet period behave like school holidays. It cannot fix weak offers, poor customer experience, unclear pricing, limited availability, slow websites or booking friction. That is why leisure operators need a business plan around every campaign.
Planning ensures seasonal dips are anticipated before they arrive. Offers, creative, email, landing pages, event programming, local partnerships, retargeting and budget phasing can then all work together. Paid social should support that plan, not be expected to replace it. And, when a quiet period arrives without that wider plan in place, the campaign quickly becomes the scapegoat for broader commercial issues.
More intervention isn’t always better
It’s completely understandable to want answers when performance changes. Asking questions, reviewing strategy and exploring new ideas are all part of a healthy client-agency relationship.
However, it’s important to recognise that every significant change carries a cost. Introducing new audience segments, restructuring campaigns, rewriting ads, changing bidding strategies or repeatedly acting on AI-generated recommendations interrupts the learning process, makes performance harder to interpret and ultimately reduces the effectiveness of the campaign.
The most valuable approach is to distinguish between meaningful optimisation and reactive intervention. Good agencies are already monitoring campaigns, testing creative, refining audiences, adjusting budgets and reporting on performance as part of their day-to-day management. These ongoing optimisations are designed to improve results while maintaining strategic consistency.
That doesn’t mean clients should simply accept everything without question. Constructive challenge, regular reviews and open discussion are essential. The strongest partnerships are built on transparency, expertise and trust, not blind acceptance.
If an agency has consistently delivered strong results during periods of peak demand, it is rarely sensible to abandon a proven strategy simply because trading conditions have become more challenging. Quieter periods usually require measured optimisation rather than wholesale tactical resets.
It’s also worth recognising that repeated requests for campaign restructures, bespoke reporting, AI strategy reviews, urgent meetings or alternative tactical approaches often involve work beyond the agreed scope of campaign management. While these requests are sometimes entirely justified, they require additional consultancy time and can divert attention away from the activities that are most likely to improve performance.
Knowing what not to touch
The skill in digital marketing is not simply knowing every possible lever. It is knowing which lever to pull, when to pull it and when to leave the campaign alone. A better response to softening ROAS could be a creative refresh, a stronger offer or a landing page change, not a root-and-branch transformation. Other times it could mean fixing tracking, an email push, a local promotion or a seasonal campaign. Most often though, the best answer is to give the campaign enough time to work. The best results come when the client, agency, Meta and AI each do the job they are good at.
The client brings commercial knowledge of the business. The agency brings strategy, experience and account context. Meta’s system optimises delivery when given the right signals and AI can support thinking and generate useful questions. The problems start when AI-generated second opinions are treated as strategy and every short-term dip becomes a reason to rebuild.
A calm campaign is not an ignored campaign, it is a managed campaign. And in quieter periods, that calm may be the difference between considered optimisation and expensive over-correction.
Plan beyond the panic
At Oyster, we help leisure brands build campaigns around commercial reality, not dashboard anxiety.
That means setting up the right tracking, agreeing the right campaign objectives, understanding seasonal trading patterns and knowing when to optimise, when to test and when to give the platform enough room to learn.
We work with clients across paid social, Google Ads, creative, web, SEO, campaign strategy and wider marketing planning, helping them connect digital activity to real business outcomes rather than short-term noise.
If your campaigns are being judged week by week, rebuilt too often or expected to solve every seasonal dip, it may be time to take a calmer look at the bigger picture.
Talk to us about building a digital marketing strategy that gives your campaigns the structure, evidence and stability they need to perform.
Meta Business Help Centre: About the learning phase
Meta Business Help Centre: About ad delivery
Meta Business Help Centre: Reduce audience fragmentation
Meta Engineering: Meta Andromeda and Advantage+ automation
OpenAI Help Centre: Does ChatGPT tell the truth?
Google Cloud: What are AI hallucinations?
