Instagram ads for roofing companies

Start with the concession, because everything else on this page is built on it. Pew's November 2025 survey puts Instagram at 40% of 50- to 64-year-olds and 19% of over-65s, against Facebook's 74% and 57%. Roof age tracks tenure and tenure tracks the age of the occupant more reliably in this trade than in any other on this site, so on the band that actually owns twenty-year-old roofs this app has roughly half Facebook's reach in one direction and a third of it in the other. That is not a reason to skip it. It is a reason to stop expecting it to do the Facebook page's job and to buy the thing it is good at instead.

The next thing to establish is that the argument the HVAC pair makes for Instagram is unavailable to roofing, because it is the obvious argument and it does not survive contact with this trade. That page's case is sequence: the format selects small photographable jobs, a $340 duct audit is how a household acquires a contractor, and the replacement a decade later goes to whoever did the small job — so what the app is really buying is a customer record. Roofing has no cheap half. The cheapest thing in the offer table below is a $189 attic survey and everything else needs a truck, a ladder and somebody on the roof. Worse, the Facebook guide in this pair has already established that a roof is bought roughly twice in a lifetime, which makes the customer list mostly an exclusion. Both of the usual defences of an Instagram budget — it reaches a different buyer, it builds a list that pays out later — are gone at once.

What is left is the format, and in this trade the format is worth more than it sounds, because roofing's actual problem is verification. A homeowner cannot see their own roof. They cannot evaluate a crew, cannot tell a thorough estimate from an expensive one, and after a storm they cannot tell the local company from the one with the out-of-state plate. Every roofer in the feed makes the identical claim. What this app can carry that a paragraph cannot is the artefact that settles it: a pair of frames shot from the same staked position eleven days apart with the dates printed on them, a membrane rolled out across bare deck before the shingle pallet opens, a card listing the three things this contractor will never say, an itemised quote with a number on it. Facebook buys memory. Instagram buys evidence, and evidence is the only part of a roofing pitch that survives the year most of these decisions actually take.

The batch behind this page is also the proof that the evidence is hard to make rather than automatic, which is why the creative section below is the longest one. The before-and-after was briefed with the camera position, the hour and the light pinned, and it came back genuinely matched: the roof geometry, the gable, the tree and the valley all register, and the timestamps read 2024-05-06 07:42 and 2024-05-17 07:41 — eleven days apart to the day and one minute apart in the hour, agreeing exactly with a headline that says eleven days. That registration is the entire evidentiary content of a before-and-after. And then the thing nobody pinned drifted: only the front plane was re-roofed, the slope across the valley is the same weathered brown in both frames, and the fascia, gutter and rake trim were replaced without being asked for, under copy reading that nothing else changed except the roof. Pin the scope as tightly as you pin the camera.

Now the part that does not invert at all, and it is the whole spine of the Facebook guide carried across unchanged. The two sentences this trade runs more than any others are prohibited advertising for a contractor in a growing list of states. Texas Business and Commerce Code section 27.02(c) makes advertising to pay, waive, absorb, rebate or credit a homeowner's deductible an offence, a Class B misdemeanour under 27.02(d). Texas Insurance Code section 4102.163(a) forbids a contractor to act as a public adjuster or, in the statute's own verb, to advertise to adjust claims on property they are working on. Minnesota Statutes section 325E.66 subdivision 1 bars both and gives the homeowner and the insurer a right to sue. None of those statutes mentions a platform. They attach to what the advertisement says, so a claim ad is exactly as unlawful in a 4:5 Instagram feed image as in a Facebook one, and Meta approves both without comment here as well.

Storm work is the one place where this app can do something the HVAC Instagram page explicitly said its trade could not. That page ran no emergency ad at all, on the correct grounds that nobody opens Instagram to find a contractor while the house is at 88 degrees. Roofing's emergency is different in kind: a severe storm is announced hours in advance and arrives across a whole neighbourhood at once, so an ad in this feed on the evening of a warning is not capturing demand — it is being the name already in the phone when the demand arrives two hours later. That is a real job for this app, and it is also the single most dangerous thing on it, because Meta's standard on commercial exploitation of crises prohibits content that exploits crises for commercial purposes and names natural disasters as its example. Our own storm ad is the cautionary case and its note is in the creative section: briefed with no countdown and no scare copy, it came back with a last-slot countdown anyway.

Then the specs, where the received wisdom is backwards. Meta's ads guide recommends about 40 characters of headline for an Instagram Feed image ad against Facebook Feed's 27, so the app everyone treats as the wordy one is the more generous of the two, with primary text about 125. The image is 4:5 at 1440 by 1800, identical to Facebook Feed, so one export serves both — but the aspect-ratio tolerance is 1% here against 3% there and the minimum width is 500 against 600, which means a crop a few pixels out of true passes on one app and fails on the other from the same file. There are four image placements here against Facebook's nine, and none of the missing five is a search result, a right-hand column or a Marketplace listing. Do not buy 1.91:1: nothing on this app is built to want it, which is why the Browse batch contains none.

Measured against that batch rather than assumed, and re-counted for this page rather than copied from the Browse note: 49 of the 50 headlines fit 40 and seven fit Facebook's 27, which is an unusually high rate of short headlines — across all seven Instagram batches published here, only 21 of 350 fit 27. Of the ten lead headlines, nine fit 40 and two fit 27. None of the 50 primary texts fits 125; they run 208 to 559 characters, and across all fourteen batches the count is now 700 of 700 over the recommendation, the shortest ever written being 200 characters. Six of ten descriptions fit 30, in a field this feed does not show and Marketplace does. Seven of the ten exports are 4:5 and three are 9:16 for Stories and Reels, and every one of them was re-framed off a square draft — one came back padded, which is what the export check in the creative rules is for.

Ads Manager offers the same six objectives here as anywhere — Awareness, Traffic, Engagement, Leads, App promotion and Sales — and this page differs from its Facebook twin on two of them. Traffic becomes the default rather than Leads, and Awareness, which the Facebook page argues for harder than any other guide in this cluster, is the objective that transfers worst.

Traffic first. The reason is the same missing surface the HVAC Instagram page found, plus a second reason that page did not have. There are no search results here, no right-hand column and no Marketplace, so nobody arrives with a problem and every ad interrupts. That alone would push toward the read rather than the ask. What roofing adds is the size of the ask: even the cheapest transaction in this trade needs somebody to climb a ladder, and the thing most of these ads are actually arguing is a specification the reader has never heard of — a nailing pattern, an ice-and-water membrane at the eaves, what a layover hides, why a two-frame before-and-after proves nothing. Asking a scrolling stranger for an appointment before any of that is asking too early, and the click is cheaper than the lead. The Browse batch split five Traffic, four Leads and one Awareness, the mirror image of the Facebook batch's five, four and one.

Leads is still a real answer for four ads in ten, and it earns its place on a narrower test than on the Facebook page: the offer has to be priced, bookable and need no explanation. A $189 attic survey passes. A flat $285 night tarp passes. A five-figure replacement does not convert from a feed whatever button you put on it, and pinning Leads to it produces a form fill from somebody who wanted a free thing.

Awareness is where this page and its twin genuinely part company, and it is worth being explicit because the Facebook guide makes an unusually strong case for it. That case is the door-knocker: a competitor will physically be on the porch within seventy-two hours of the hail, no bid wins against a doorbell, and the only defence is having been heard of first. It is a memory buy, and a memory buy is worth exactly what its reach is worth. On the bands that get door-knocked, this app has a third of the other one's reach. So the objective the Facebook page argues hardest for is the one that transfers worst — not because the argument is wrong, but because it is the argument that depends most directly on the number this page opened by conceding. There is one Awareness ad in the batch and it is the strongest ad on the page; it is still the smallest line in the budget here.

Sales is a flat no and roofing states it more absolutely than any other page in this cluster. Both roofing batches came back with no Sales ad, and so did both dentist batches, both real-estate ones and gym's Facebook one — seven of the fourteen published here have none, so that is not peculiar to this trade. What is peculiar is how far it reaches: a practice can still sell a whitening kit and an agent a report, and roofing has nothing at all a homeowner buys without somebody physically going up a ladder first.

This is not the Facebook table with different pictures, and the sorting rule is the most useful thing on the page. An offer belongs here if its proof is an artefact you can photograph, and on the other page if the reader has to take it on trust. That is why the warranty transfer, the certified-installer argument and the financed replacement are missing below: they are all claims about a document somebody else holds, and a feed is the worst place in the world to ask a stranger to believe one. What is here instead is the attic photograph, the membrane on bare deck, the fixed-point pair, the never-list card, the quote that foots and the timestamped night tarp.

Each row pins the same four decisions as its twin: who the offer is actually for, which angle family argues it best — every one is a page in the angle library — which objective buys it, and, where the batch has one, a finished ad showing the whole thing assembled. Ten of the twelve have one, which is every ad in the batch used exactly once. Two do not, and an absence is more honest than borrowing a neighbouring ad.

One row needs reading before the rest, because it is the reason the Facebook guide's four-word rule exists and it is the best result this cluster has produced. The claim-work row below is the same offer that failed on the Facebook page. Briefed neutrally there, it attacked deductible waiving by name and then promised to file supplements, stated what the carrier covers, and signed off with a four-word description line handing the whole claim over — the famous violation avoided and the obscure one walked straight into. That page diagnosed it and prescribed a fix: document, meet, invoice, never negotiate. This batch wrote that boundary into the brief, and it held across all eleven artefacts, with the rendered card naming both prohibitions correctly and the fifth primary text stating the line outright. One brief variable flipped the compliance outcome. The disclosure is a property of the brief, not of the model.

Identity and Emotion is on this page as it is on the Facebook one, and for the same closed rule rather than a preference: define the audience by a decision rather than a demographic. The row below is aimed at people who spent the renovation money on the roof, and what came back is second person, frankly emotional, and asserts nothing about anybody — the family voted kitchen, you voted ice-and-water membrane. Meta's personal attributes policy is a list of attributes, and a renovation you chose to do is not on it. An identity ad is dangerous when it tells the reader what they are and workable when it tells them what they did.

Begin by conceding what a reader arriving here usually wants to hear, because it is not true. Choosing this app is not a targeting decision and cannot be used as one. The special ad category is declared on the campaign, and Instagram Feed, Stories, Reels and Explore are placements inside it — so a roof advertised on a fixed monthly payment loses exactly what it loses on the other page, and the loss is more expensive in this trade than in any other. The restricted list removes ZIP codes, neighbourhoods, sub-city, sub-neighbourhood, metro area, small geographic area and electoral district as location types, along with location exclusion, lookalikes, saved audiences, custom age ranges, behaviour and demographic targeting and interest exclusion, and it sets a radius floor of 15 miles or 25 kilometres in the US and Canada. A hail swath is a ZIP list a few miles wide. So declaring the category does not cost a roofer a drive time, it deletes the campaign — and running the financed ad here buys that same deletion against a third of the homeowners.

The declaration itself is compulsory either way and has been since 31 March 2020: every campaign created through the Marketing API specifies a value, with NONE as the answer for the survey, the tarp, the tear-off and the certification. Roofing is not one of Meta's four categories. The category attaches to what an ad offers, not to what a company sells.

The storm plan is the second thing that does not belong to Facebook, and it is worth saying plainly because the Facebook guide's targeting section is built on it. A postcode list is set on the ad set, and placements sit underneath it, so the swath the hail actually crossed targets Instagram delivery exactly as it targets Feed and Marketplace. Storm reports give you that list within a day, a mile outside the stripe there is nothing to sell, and this remains the most valuable lever in a roofing account whichever app the impression lands on. What changes here is only how much of the stripe you can reach.

Now the two levers that genuinely invert, and both of them invert against the advice on the HVAC Instagram page, which is the closest neighbour to this one in the matrix. The first is age, and it is doing more work here than on any other page in this cluster. That page bands toward 30 to 49 because this app's strongest band, 18 to 29 at 80%, is substantially renters. The renter problem is identical in roofing, and there is a second one on top of it: this trade's buyer is defined by tenure, so even the owners in the app's strongest band are the least likely people in the country to be living under a twenty-year-old roof. You are banding against the platform's strength. That is survivable, and the way it survives is the offer table — the ads above are aimed at the same households the Facebook page sells to, met earlier: two winters after finishing the attic, one ice dam in, one deferred winter. Target a band and never name it in the copy, because age is on the personal attributes list.

The second inversion is the radius, and here the HVAC page's advice is actively wrong for this trade. That page says to hold the radius tighter than feels comfortable, because a $349 thermostat install cannot absorb a forty-minute drive. Roofing's tickets are five figures and its cheapest job is a tarp with a flat price built to cover the call-out, so the drive is not the constraint — crew logistics and licensing are. Set the radius to the area you can actually put a crew in, which for most single-depot contractors is wider than the HVAC rule and tighter than the county. The ordinary floor is 0.63 miles or 1 kilometre, far tighter than anybody sets it.

Two things survive everything and both are under-used. Custom Audiences keep inclusion, exclusion and expansion even under the financial products and services category, along with Advantage+ Audience and Detailed Targeting Expansion, which makes your own data the only precision instrument left inside a declared campaign — and you can check a list with is_eligible_for_sac_campaigns before launch rather than discovering the problem at it. In roofing the valuable list is not the customer list, because a roof is bought twice in a lifetime. It is everyone you inspected and did not sign: a roof you have personally been on, a written finding, and a decision already in progress that frequently takes a year to close. This app is one of the places that list gets filled, and the funnel section is built on that.

One hard-error footnote, because it fails at launch rather than in delivery. A business admin has to accept Meta's non-discrimination policy in Business Settings before a housing, employment or financial campaign will deliver at all: the API returns error 2859024, Certification Required. Its neighbour 2909035 comes back for a custom age range, a saved audience, a lookalike, or a radius under the floor. If a financing campaign will not start, read those two codes before touching the creative.

The default here is a landing page, and it follows from Traffic being the default rather than being a separate decision. Almost everything above is arguing a specification the reader cannot name — a membrane at the eaves, what a bare deck shows, why two frames of a finished roof prove nothing — and that needs a page with the price on it. Never the homepage: somebody who tapped an ad about a stain on their own attic sheathing and lands on a rotating drone shot has been asked to start over.

The Instant Form is right for the two offers that are priced and bookable, and the type matters. Meta gives you two, and More volume is built to be filled and submitted in one thumb movement from pre-filled profile data. That is what makes it cheap and what makes it useless in this trade, because you will be ringing back hours later into a queue of people who have been called by six roofers. Use Higher intent, add two or three qualifying questions — age of roof, own or rent, storm date if there was one — and accept the volume drop.

The call ad is the interesting one, because the HVAC Instagram page removes it entirely and this page keeps it. That page is right about its own trade: nobody opens Instagram to be rescued from a 96-degree house, so buying call ads there is buying the format without the moment. Roofing's emergency arrives differently. A severe storm is announced hours in advance and lands on a whole neighbourhood at once, so the evening buy is real, and call ads do deliver on Instagram Feed and Stories. Judge it on 60-second calls rather than calls placed, daypart it to the hours somebody answers, and understand what you are buying: not a rescue, a phone number in the right place two hours early.

Anything touching an insurance claim gets a landing page, and this is a compliance decision rather than a conversion one. The statutes that make handling a claim prohibited do not stop at the ad unit — a page saying the same thing is making the same representation with more words, and it is advertising too. Write it to the same standard as the ad: document, meet, invoice, never negotiate, never interpret.

The fourth option is the one this app adds and the one to be most careful with, and roofing has a better use for it than most trades. Lead ads can open a conversation on Messenger, Instagram or WhatsApp, and the instant forms and Messenger conversion location runs one campaign and lets delivery pick per person. In this trade the thread is where a photograph arrives without anybody asking for it: shingles in the garden, a stain on a bedroom ceiling, a phone held up at the roofline. That is genuine triage and it is worth having. It is also where a household messages at eleven at night during a storm, gets no reply until Tuesday, and believes the whole time that a crew is coming. Use it where a person actually answers and the conversation ends with a date, and never as the destination for the storm ad.

Start with the law, because it is the only section on this site where the binding rule is not Meta's and because nothing about it changes on this app. Texas Business and Commerce Code section 27.02(c) makes it an offence for a seller expecting payment from insurance proceeds to advertise or promise to pay, waive, absorb, rebate or credit the insured's deductible, a Class B misdemeanour under 27.02(d), with a 12-point boldface notice required by 27.02(b) on such contracts of $1,000 or more. Texas Insurance Code section 4102.163(a) says a contractor may not act as a public adjuster or advertise to adjust claims for property they are providing or may provide contracting services for, regardless of a licence or a power of attorney. Minnesota Statutes section 325E.66 subdivision 1(a)(1) bars advertising or promising to pay any part of a deductible as an inducement, subdivision 1(a)(3) bars interpreting policy provisions or advising on coverages without a public adjuster licence, subdivision 1(c) includes residential roofers by name, and subdivision 2 gives both the insured and the insurer an action. The operative verb in both statutes is advertise. Meta will approve either sentence without comment.

The compliant version of the service is four words long and this batch proved it can be briefed. Document, meet, invoice: you photograph the damage, you are on the roof when the adjuster is, you invoice the carrier directly. That describes exactly the same work, asserts nothing about what a policy pays and claims to act for nobody. The Facebook batch produced the failure — a claim ad that attacked deductible waiving by name and then promised to file supplements, stated what the carrier covers, and handed the whole thing over in a four-word description line. This batch wrote the boundary into the brief instead, and it held across all eleven artefacts: the rendered card lists both prohibitions correctly, all five primary texts stay inside the line, and the fifth states it outright — never waive your deductible, never promise a free roof, never negotiate your claim, and then what the contractor will do, which is inspect, document and build to scope. One brief variable flipped the outcome, which is the general finding: the disclosure is a property of the brief rather than of the model.

That result is one half of the batch's central finding and the other half is the expensive half. Five briefs stated a positive requirement and all five came back exact — the statutory boundary, the advertiser's role, arithmetic that foots to the cent, the signature, the vantage. Three stated a prohibition and all three lost, every one of them to something else the same brief had asked for. The storm ad pinned a scarcity sub-angle and then said no countdown and no scare copy; it came back with the briefed weather alert deleted, a glowing last-open-crew-slot notification, five scarcity headlines out of five, and an invitation to watch a ceiling stain spread. The quiz forbade pre-marking an answer by enumerating the markers it must not use — marked, circled, ticked, highlighted or otherwise indicated — and came back with an orange accent bar down one option that appears on none of the others. The crew photograph asked for roofers harnessed and tied off to a ridge anchor, and the harness is in the frame, lying loose on the deck, while a roofer works on his hands beside it. Brief what you want, not what you fear: a prohibition survives only when nothing else in the brief pulls against it, and the sub-angle, the visual direction and the format all pull.

The three failures each have a mechanical form worth carrying into any brief in any trade. A prohibition cannot defuse the sub-angle the same brief pinned, because the taxonomy is what the concept stage is built to execute and the prohibition is a sentence — if you do not want urgency mechanics, do not buy a scarcity angle. Prohibiting by enumeration invites the item you did not enumerate; specify the positive invariant instead, which here would have been that all four rows are identical in styling. And naming a noun does not name its state: the brief got its harness and never said it had to be worn, which in a trade governed by fall-protection rules is not a detail, and the copy inherited the omission because the copy stage reads the render and describes what is actually in the frame.

The same mechanism runs through people and figures, and both replicate cleanly here. This batch contains two author-implying directions with one variable different between them: the founder's card was told to sign as the owner rather than by a name and signed exactly that, with no name and no face, while the storm ad named a role — the dispatcher on duty — without forbidding a name, and the pipeline invented two people and then sold the practice as a feature. So the corpus rule sharpens from name the speaker to this: the pipeline invents a proper name unless the brief forbids one, because a role is a slot and slots get filled. The proof-slot rule is the same mechanism applied to numbers and it replicated for the fourth batch running — zero of the ten images here renders a fabricated statistic, because a raw phone photograph, a quiz, a lock screen, a quote card and a before-and-after split have nowhere to put a tally. Every invented figure on that page is in the copy, where it is easier to catch.

Then two number rules that pull in opposite directions and are both replicated in this batch. A count set as typography is reliable: four numbered callouts were briefed on the membrane ad and exactly four render, and the four figures on the cost-per-day poster are all correct. Arithmetic is reliable when it is written into the brief: $27,375 across a 25-year system is 9,125 days at $3.00 exactly, restated correctly in the copy as a division. What is not reliable is a count of drawn objects, which is why the Facebook batch's nailing diagram drew five nails under a caption saying four and seven under a caption saying six. If the number is the argument, keep it in type and let the picture carry the mechanism.

Regulation Z gets a paragraph even though this batch has no financing ad, because the result is the useful kind. The HVAC Instagram batch found a credit line in an ad that was never briefed for one, so this batch deliberately contained no financing brief at all and its price ad invited the payment frame while forbidding credit. Nothing came back: no monthly payment, no APR, no instalments, no deferred terms anywhere in fifty headlines and fifty primary texts, and the ad says instead that the job is paid in full when the crew finishes. Worth stating precisely, because the HVAC finding is easy to misread. 12 CFR 1026.24(d) makes four things triggering terms — the amount or percentage of any downpayment, the number of payments or period of repayment, the amount of any payment, and the amount of any finance charge — and they attach to a consumer credit advertisement. This ad offers no credit, so no deposits required is not a triggering term here. The moment you do quote a monthly figure, the ad owes the reader the downpayment, the repayment terms over the full term and the annual percentage rate using that phrase, and your finance partner has approved language for exactly that.

Finally the shapes and the counts, and they are not the Facebook page's. Render 4:5 at 1440 by 1800, the same export Facebook Feed wants, but hold the crop to 1% rather than 3% and mind a minimum width of 500 rather than 600. Do not buy 1.91:1 — the surfaces built to want it are Facebook's right-hand column and search results, and this app has neither, which is why the batch contains none. Keep 9:16 for Stories and Reels, where three of these ten sit, and check every re-framed export by eye: one of those three came back with a flat cream band across the top where the square draft ran full-bleed. On copy, write the headline to 40 characters and check it against 27 as well if the same ad runs on Facebook Feed, because both numbers are live on one ad with Advantage+ placements on. Seven of this batch's fifty headlines fit 27, which is a high rate — only 21 of 350 across seven Instagram batches do — but half of those seven are claims rather than nouns, so the rule the apparel page derived, that 27 characters is room for a noun and not for a claim, does not hold cleanly here. None of the fifty primary texts fits 125, and none of the 700 in this whole corpus does. Longer is not forbidden, it is truncated, so the argument has to survive inside the first sentence.

Three stages, and a fourth line that is deliberately not on this app at all. The shape rhymes with the Facebook page's and the weights do not, because the middle stage there is a burst and the third stage here is the payout.

Keep it this thin on purpose. A single-depot roofer's budget is small relative to the audience arithmetic Meta's delivery system wants, and it is smaller again on an app with a third of the reach, so every extra ad set divides the same evidence into thinner piles. Variation belongs at the ad level inside one ad set, and the thing that varies is the offer rather than the audience. This is also the reason not to build Instagram-only ad sets, which the budget section works through.

The third stage is the one to read carefully, because it is what this app is actually buying. The Facebook guide's finding is that in roofing the customer list is mostly an exclusion — a roof is bought twice in a lifetime — and that the list which behaves like a customer list is the one nobody uploads: every inspection that never became a job. Those households have a roof you have been on, a written finding and a decision already in progress, and the interval between the inspection and the signature is frequently a year. The ads at the top of this page are how strangers enter that list, and the documentation arguments are what keep them warm through the year it takes. An Instagram budget in this trade is a bet on that interval, which is why the measurement advice below is about twelve months rather than thirty days.

Do the arithmetic before you copy a number from anywhere, including here. Meta's delivery system needs roughly 50 optimisation events per ad set per week to leave the learning phase; below that the ad set sits in learning limited, which is not a penalty so much as an admission that the system cannot optimise with what you are giving it. Take your own cost per lead, multiply by 50, and that is the weekly spend at which one ad set delivers with full information.

Now the concession, and it is the exact thing the HVAC Instagram page got to celebrate. That page's good news is that its offers are small, so its events are cheap, so 50 a week is reachable in a way it is not for a campaign selling five-figure replacements — the app with the worse reach has the better chance of getting an ad set out of the learning phase. Roofing gets the opposite of that on both terms. The offers are not small, so the events are expensive, and the audience is a third to a half the size before you have narrowed anything. An Instagram-only ad set in this trade is the most likely structure in this entire guide to sit in learning limited indefinitely.

Which produces the most practical instruction on this page, and it is a structural one rather than a creative one: do not buy this app as its own ad set. Leave placements on Advantage+, let one ad set pool its evidence across all thirteen surfaces, and treat Instagram as a briefing decision — which offers you write, and how you photograph them — rather than as a media buy with its own budget line. Restricting placements to Instagram is a real cost and needs a real reason, and the only good one in this trade is a 9:16 asset built for Stories, which is a reason to add a placement rather than to subtract eight.

The storm collision from the Facebook page carries across and gets one wrinkle worse. The learning phase runs about a week, a hailstorm's advertising window is a few days, and the door-knockers are on the street within seventy-two hours — so an account switched on the morning after the hail spends the whole window learning while somebody is ringing doorbells. Keep a standing storm ad set alive at a low budget through the quiet months so that the storm week is a budget increase and a creative swap rather than a cold start, and accept the honest trade-off: adding the swath's postcodes is an audience change, which is a significant edit, so there is no version of this advice that avoids some learning cost. The Instagram wrinkle is that you cannot spin up an app-specific ad set for a single evening and expect it to deliver, which is another way of saying the same thing as the paragraph above.

Finally, what to judge it on, because the wrong report kills this budget faster than the wrong creative. Do not compare an Instagram cost per lead to the Facebook one — and on a properly built account you cannot anyway, because the split is a delivery outcome inside a shared ad set rather than two campaigns. Judge the whole account on booked inspections inside the postcodes you targeted, and judge this app's contribution the way its own funnel works: on how many households entered the inspection list, and on how many of them signed something between six and eighteen months later. An account judged on a thirty-day window in a trade whose decision takes a year will switch off the campaign that was working, and it will do it right before the campaign pays.

The first three are not about advertising performance at all, and they are first because Meta's approval tells you nothing about any of them. The rest are the ways a workable roofing account quietly becomes a worse one, and most of them come out of the ten ads on the Browse page rather than out of anybody's opinion.