Your ads did not break, and Meta did not flip some hidden switch on your account. What happened is almost certainly creative fatigue: the homeowners in your service area have now seen the same ad too many times, and every impression past that point costs more and converts worse. The fix is a refresh cadence, not new targeting. Here is the maths behind the three-week cliff, and the refresh schedule that keeps cost per lead flat through the season.
If you run Meta ads for an HVAC company, or you are the owner watching the dashboard yourself, you have lived this arc. A new ad goes live in early June. Week one it beats your cost per lead target, so you nudge the budget up. Week two the calls keep coming and the install board fills. Week three holds, maybe a shade softer. Then somewhere around day 22 the CPL starts climbing, and by day 30 it has doubled. You changed nothing. The heatwave is still on. Same offer, same landing page, same ad.
Most operators blame targeting at this point and start rebuilding audiences: a fresh lookalike, a wider radius. I understand the instinct, but it rarely helps, because targeting was never the problem. An HVAC company sells to a fixed geography. There is no new audience hiding behind a better interest stack. There are the same 100,000 or so homeowners your trucks can reach, and by week three, most of the ones who matter have already seen your ad.
Why do Facebook ads stop working after a few weeks?
Because ad performance decays with repeat exposure, and that decay is well measured. AdAmigo's frequency benchmarks put the warning line for prospecting campaigns at a frequency of 2.5, with fatigue actively underway above 3. Once people have seen an ad 5 or more times, their data shows costs rising 50% to 80% while CTR falls 40% to 55%. The person who ignored your ad four times is not going to call you on the fifth impression. You are just paying to annoy them.
What makes HVAC different is how fast you get there. A national supplement brand can spend against 40 million people and take a quarter to show anyone the same ad twice. Your addressable market is capped by drive time. One home services guide puts it bluntly: target a city of 100,000 people and everyone will see your ad within a week. Fatigue that takes months to bite in ecommerce bites in three weeks in a bounded metro. Sooner, in peak AC season, when you have pushed budgets up to catch the demand.
Run the maths on your own metro
Here is roughly what the decay looks like in numbers. Say your service area is a 20-mile radius holding about 100,000 Meta-reachable adults who own their home. You spend $150 a day, and summer CPMs for local services sit somewhere near $18. That buys roughly 8,300 impressions a day, call it 58,000 a week. Spread evenly across the full pool, that is a frequency of about 0.6 per week. Sounds harmless.
Meta does not spread delivery evenly. The system concentrates impressions on the slice of the audience it predicts will respond. If 70% of your impressions land on the 20,000 most responsive homeowners, that group absorbs around two exposures a week each. By the end of week three they have seen your ad six times. Meanwhile the blended frequency column in Ads Manager reads something like 2.7, which looks survivable. The specific people who were generating your leads are already deep in the exposure zone where AdAmigo's data says costs jump by half or more.
That is why the cliff feels so sudden. The ad does not decay evenly across the metro. It exhausts your buyers first.
The ad did not get worse. The homeowners most likely to call had all seen it, and Meta had nowhere cheaper left to send it.
Creative fatigue is a label Meta puts on your HVAC ads
None of this needs inferring from third-party dashboards, because Meta names the condition inside Ads Manager. According to Meta's Business Help Centre, the delivery column flags an ad as "Creative limited" once its cost per result runs above its own history, and escalates to "Creative fatigue" once cost per result reaches at least double what it used to be. Meta's recommended action for both is the same: make new ads.
Read that threshold again. By the time Meta prints the words "Creative fatigue" on your ad, your cost per result has already doubled. It is a post-mortem, not a warning. And most HVAC operators never see it anyway, because the label lives in a column that gets checked about as often as the payment settings. If you take one habit from this post, make it a weekly look at the delivery column and the frequency column, side by side, every Monday.
Andromeda finds your buyers faster, then runs out of them
The three-week number is newer than the fatigue problem itself. Meta's Andromeda retrieval engine, rolled out through 2025, changed how ads get matched to people: rather than leaning on your audience settings, it reads the creative itself and hunts for the users most likely to respond to it. We covered the mechanics in our Andromeda explainer. The side effect that matters here is speed. Segwise's 2026 analysis reports that a single ad concept now burns through its addressable audience in 2 to 3 weeks, down from 6 or more before Andromeda. The system finds your responders faster, so it exhausts them faster. The engine is doing exactly what it was built to do, just inside an audience with a hard ceiling.
There is a second Andromeda wrinkle that catches service businesses specifically. The system clusters near-identical ads and treats them as one creative entity. Ten light re-edits of the same testimonial video are, in its eyes, one ad. So the classic cheap refresh (new thumbnail, trimmed intro, different music) resets almost nothing. Only a genuinely different concept, meaning a different angle, face or format, re-enters retrieval as something new.
Five signals it is fatigue and not the economy
HVAC owners tell me a version of the same story whenever this happens: leads dried up, must be the market. Sometimes it is. Usually it is not, and you can tell the difference in about four minutes inside Ads Manager. Fatigue looks like this:
- Frequency above 2.5 on prospecting in a metro under 200,000 people (above 3.5 the ad is cooked)
- CTR down 20% or more against the same ad's own 7-day average
- CPC or CPM up 25% or more with no seasonal auction change to explain it
- Cost per result heading towards 2x its best stable week, or the "Creative limited" label already showing
- Results decaying while reach flatlines, which means Meta is re-serving the same people
Two or more of those together and the diagnosis is done. Zentric's creative fatigue playbook uses near-identical thresholds for accounts at any spend level. A genuine demand slump looks different: CTR holds, frequency stays flat, and lead quality drops across the whole account rather than one ad at a time.
The fix is a cadence, not a better ad
Here is what I keep seeing first-hand. Nearly every home services account that lands with us arrives in the same shape: one or two ads carrying the whole budget, both past their peak, nothing in production behind them. Not because the owner is lazy. Because for a service business, a new ad means a project. Book a videographer, pull a tech off a job, find a customer willing to be on camera, wait two weeks for the edit. When production hurts that much, the winning ad runs until it dies, and then everyone stands around the corpse arguing about targeting.
So the durable fix is production design. Three parts.
Set the refresh calendar before you launch
In season, a genuinely new concept should enter rotation every 2 to 3 weeks. In a small metro under 200,000 people, closer to two, with hooks re-cut every 10 to 14 days. Off season you can stretch towards monthly. The point is that refresh dates go in the calendar the day the campaign launches, because deciding to refresh only after fatigue shows up means two or three weeks of doubled CPL while you scramble a shoot together. Remember the arithmetic of Meta's own label: by the time the flag appears, a $45 lead has become a $90 lead. Our HVAC cost per lead breakdown covers what those inflated weeks do to a monthly budget.
Shoot once, cut many
One half-day shoot with a tech, a real customer and a phone produces three distinct concepts. Each concept cuts into two or three hooks and a couple of lengths, so 12 to 18 assets from one afternoon. That is roughly six weeks of rotation from a single production day, which is what makes the calendar above survivable for a company whose actual job is fixing air conditioners. The Andromeda clustering point earns its keep here: those cuts need genuinely different openings and claims, not the same video wearing different music.
Rotate the angle, not just the edit
HVAC hands you more angles than almost any category. The 2am emergency. The comfort story, a family sleeping downstairs because the bedroom will not drop below 85 degrees. Replacement plus financing. The seasonal tune-up offer. The owner or a tech talking straight to camera about what a $99 service visit actually includes. Each angle lands with a different slice of the metro, so rotating angles resets fatigue in a way that re-editing one angle never will. And if nobody at the company wants to be the face, that is a solved problem now.
One more thing, because it costs accounts real money: do not pause a fatigued winner to "rest" it. Meta's guidance is that ad sets paused for 7 days or more re-enter the learning phase when they come back, so you get a reset, not your old ad. Iterate on the winner instead with a new hook on the proven body, and when it is genuinely done, kill it on a rule, not a feeling.
Key takeaway
In a bounded service area, creative fatigue is a scheduling problem before it is a creative problem. Put refresh dates in the calendar at launch: a new concept every 2 to 3 weeks in season, hook re-cuts every 10 to 14 days, and never let one ad carry more than half the budget.
Where this breaks down in practice
Everything above is doable in-house if someone owns it. The place it falls apart is volume. Most HVAC companies can manage one shoot. The grind is turning that shoot into 15 assets, every month, indefinitely, while also running the actual business. Somewhere around month two the edits slip, the winner stays live past day 30, and the account is back where it started.
That production treadmill is the thing Spark exists to remove. We build rolling libraries of creator-style ads for service businesses, so a fresh concept is always ready before the current one fatigues. You can see how the engine works and what it costs, and the rest of our operator playbooks live in the resources library.
If your best ad is on day 19 right now, the replacement should already be in production. Start a project and we will have new concepts in rotation before it dies.
Frequently asked questions
Why did my HVAC Facebook ads suddenly stop working?
The most common cause is creative fatigue: the homeowners in your service area have seen the same ad too many times. Check the frequency column (above 2.5 to 3 on prospecting is the danger zone in a local market), compare CTR against the ad's own 7-day average, and look for Meta's Creative limited or Creative fatigue labels in the delivery column. If frequency is climbing while CTR falls, you need new creative, not new targeting.
What frequency is too high for local service ads?
Treat 2.5 as the warning line on prospecting campaigns and act well before 3.5. Frequency benchmark data shows costs rising 50% to 80% and CTR falling 40% to 55% once people have seen an ad 5 or more times. Local audiences hit those numbers much faster than national ones, and the blended frequency figure understates how often your most responsive homeowners have actually seen the ad.
How often should HVAC companies refresh ad creative?
In peak season, put a genuinely new concept into rotation every 2 to 3 weeks and re-cut hooks every 10 to 14 days in metros under 200,000 people. Off season, monthly is usually enough. Plan production ahead of the calendar: one half-day shoot can produce 12 to 18 assets across three concepts, which covers roughly six weeks of rotation.
Should I pause a fatigued ad and run it again later?
No. Ad sets paused for 7 days or more re-enter Meta's learning phase when they resume, so you get a reset rather than your old winner back. Iterate instead: keep the proven body and test new hooks against it, or rebuild the same offer with a different angle and a different face. Retire the fatigued version once a replacement is beating it.