Instagram ads for home remodeling companies

Start with who is here, because remodeling is the trade with two buyers and this app has one of them. Pew's November 2025 survey puts Instagram at 62% of 30- to 49-year-olds and 40% of 50- to 64-year-olds, against Facebook's 80% and 74%. The older band owns the houses with the equity and buys the whole-house work; the younger band is where the single-room jobs live — the hall bathroom, the kitchen refresh, the one wall that has been bothering somebody for six years — and those are the jobs a single crew can actually schedule. This is not a smaller copy of the Facebook audience. It is the half of your own order book that photographs well, reached at roughly three quarters of Facebook's rate in the band that buys it.

Then the clock, which is this cluster's most useful platform fact and belongs to this app alone. Meta's engagement custom audiences documentation sets a maximum retention of 730 days for an audience built from an Instagram business profile, the same for a Page, 365 for Shopping and augmented reality — and 90 days for Lead Generation Ads. Those two numbers describe the same household at two moments: the one who saved your floor-plan diagram in March, and the one who filled in your form in March. In this trade the drawings are not finished when the second one expires. The first is still addressable when the kitchen is installed.

Now the finding this page exists for. You cannot buy the save. Meta's ad set reference publishes the `optimization_goal` enum, which is the whole vocabulary of things delivery can be told to want, and a save is not in it — not as an ad save, not as a post save, not under engagement. `POST_ENGAGEMENT` is "Optimize for people more likely to engage with your post", which is the cheapest available reaction and in a feed full of kitchens that is a like. A save is not a reaction; it is a decision to come back. The auction has no price for it, which is exactly why it is worth having: the one signal in this account that cannot be bought has to be earned by the creative.

What you can buy is the visit, and it is on the same clock. `VISIT_INSTAGRAM_PROFILE` — "Optimize for visits to the advertiser's Instagram profile" — is in the enum, `INSTAGRAM_PROFILE`, `FACEBOOK_PAGE` and `INSTAGRAM_PROFILE_AND_FACEBOOK_PAGE` are all values of `destination_type`, and a profile visit is `ig_business_profile_visit`, an Instagram business profile engagement event with the same 730-day maximum retention as the save. So the buyable proxy for a two-year audience is a visit to your own grid. That is a real mechanism and it is specific to this app: there is no Facebook equivalent that fills a two-year pool at the price of a tap, and it is the reason the funnel below has a stage that buys an audience rather than a result.

Which of the seventeen events to build, because the menu is longer than the documentation. Meta's list of Instagram business profile engagement events has seventeen entries and five of them carry a description: `ig_business_profile_all` ("the most inclusive engagement type", a union of the others), `ig_business_profile_engaged`, `ig_user_messaged_business`, `ig_business_profile_visit` ("People who visited your Instagram business profile") and `ig_business_profile_ad_saved` ("People who saved either organic content or an ad from your Instagram business profile"). The other twelve — `ig_ad_save`, `ig_organic_save`, `ig_ad_like`, `ig_ad_cta_click`, `ig_ad_carousel_swipe` and the rest — are bare names in the same table with no description at all. Build on the five that are documented. Our own examples page named `ig_ad_save` for the save audience and it is corrected as of today: the description belongs to `ig_business_profile_ad_saved`, and the practical difference is that the documented event explicitly covers organic posts as well as ads, which in this trade is most of what gets saved.

Three mechanics worth knowing before you build one. Multiple rules are a union — "Users that match at least one of the rules will be added to the audience" — so a saved-plus-visited pool is one audience with two rules rather than two audiences you have to reconcile. An ad account can hold a maximum of 500 engagement custom audiences, which is more headroom than any remodeling company will use and is worth knowing before somebody builds one per campaign. And the documentation says an engagement audience is prefilled on creation with people who already engaged inside the retention window — the sentence is written about a Page and the Instagram section does not repeat it, so treat the backfill as a bonus rather than a plan, and build the audience the week you start rather than the week you need it.

Then the parameter that decides whether the pool is a pool. Advantage+ audience is described as a way to "expand the eligible audience pool the delivery system can search", and the constraints it will not expand are enumerated: "location constraints, minimum age, language, and custom audience exclusions". Custom audience *inclusions* are absent from that list, and from v23.0 `advantage_audience` within `targeting_automation` defaults to 1 or requires an explicit 1 or 0 on every new ad set. So on this app, in this trade, the sentence to take away is: your exclusion is a wall and your inclusion is a suggestion. The tier that spends money on the people who remembered you does not exist until somebody sets that value to zero, and the change only binds new ad sets, so an old one carries whatever it was created with.

Measurement, honestly. There is an action type for saves — `onsite_conversion.post_save`, labelled "Post Saves" — and the reference gives no platform for it and lists no Instagram-specific action types at all, so before you make saves a reported KPI, break the column by publisher platform in your own account and confirm it moves. For the visit tier the result is the thing you bought, which is the cleanest measurement in this account by a distance. And for the money, nothing has changed from the Facebook page: attribution is 1-day click, 7-day click and 1-day view, and a kitchen signed in the spring is invisible to all three.

Finally the specs and the counts, measured on this batch rather than assumed. Meta's ads guide recommends 4:5 at 1440 × 1800 for Instagram Feed — the same export Facebook Feed wants — with about 125 characters of primary text and a 40-character headline, an aspect-ratio tolerance of 1% against Facebook's 3% and a minimum width of 500 against 600. There are four image placements here against Facebook's nine, and none of the missing five wants a 1.91:1, which is why this batch contains none. This run wrote fifty headlines and every one of them fits 40, the longest at 39 characters — the first Instagram batch on this site with no exceptions, against 362 of 400 across the eight published here. Five fit Facebook's 27 and none of the ten lead headlines does. Nine of the ten description lines fit Marketplace's 30, in a field this app does not show. Six exports are 4:5, three are 9:16 for Stories and Reels, one is a square.

And the copy-length result, which is the most transferable thing the batch produced. Across the eight Instagram batches published here, exactly five of 400 primary texts fit the 125-character recommendation — and all five belong to one ad on this page, the floor-plan diagram, where they run 63 to 90 characters. The other 45 in this batch run 247 to 395. The Facebook twin got the identical result from its live-in kitchen photograph, five out of five at 74 to 99 characters, in the same trade from a completely different direction. So copy length is not a property of the platform or the trade: it tracks how much of the argument the image is already making. A dimensioned plan argues completely and leaves the copy nothing to explain. If you want short copy, do not ask for short copy — brief a picture that finishes the case.

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 three of them. Traffic replaces Leads as the default, Awareness drops from a real recommendation to rarely, and Engagement is the trap that looks most like the right answer.

Traffic is the default, and on this app it has two destinations rather than one. The ordinary version sends a reader to a page that shows the document the ad argued from — the allowance sheet, the permit card, the day-by-day schedule. The version that is only available here sends them to your profile, bought against the profile-visit goal, and what you are purchasing is not a session, it is a two-year membership: the visit is an addressable engagement for up to 730 days. The Browse batch split five Traffic, four Leads and one Awareness, the near-mirror of the Facebook batch's six Leads and three Traffic, and that split is roughly the right shape for a remodeling account here.

Leads is the second line and it is aimed at a different audience than on the other page. On Facebook the lead ad is how a stranger becomes a consultation. Here it is mostly how somebody who has been carrying you around for eleven months finally books, which means it belongs pointed at the 730-day pool rather than at cold geography — and it inherits the Facebook guide's whole apparatus: Instant Forms on "Higher intent", the held consultation as the performance goal wherever volume allows, conversion leads optimisation if you can clear its gate. Remember what the form audience itself is worth, though: a Lead Generation Ads engagement audience retains for 90 days, which is the shortest memory in the account and the one most likely to be mistaken for the long one.

Awareness is where this page parts company with its twin, which argues for it harder than almost any guide in this cluster — the decision takes a year, it happens in a conversation you are not in, and you are buying being named in it. That argument is right and this app has a better instrument for it. A reach buy leaves nothing behind; a profile-visit buy reaches somebody and leaves an addressable trace that lasts two years. When the same money can produce an impression or a membership, buy the membership. Keep an Awareness line only for the thing a visit cannot carry, which is a 9:16 story about the shop, and expect it to be the smallest number in the plan.

Engagement is the trap, and it is worth being explicit because it is the objective every article about saves recommends. It optimises for the cheapest available reaction — Meta's own description of the post-engagement goal is "Optimize for people more likely to engage with your post" — and in a feed full of kitchens the cheapest reaction is a like from somebody who likes kitchens. A save is not a cheap reaction, it is a decision to come back, which is precisely why there is no goal for it. Buying engagement to get saves buys the population most willing to tap something, and then you retarget them for two years.

Sales is a flat no and this trade states it twice over: neither remodeling batch contains a Sales ad, which makes nine of the sixteen batches published here Sales-free and four trades — dentistry, real estate, roofing and remodeling — with none in either batch. Nobody buys a kitchen in a checkout, and the deposit that can be taken online is a lead with a card attached. App promotion is the same answer as everywhere: a kitchen is remodelled roughly once a generation.

The sorting rule on this page is not the Facebook page's, and it follows from the mechanism above. There, an offer belongs on the page if its argument is paperwork, because paperwork reads as evidence in that feed. Here an offer belongs if the ad is worth opening again in eleven months — because the behaviour this account is built to accumulate is the save, and a save is a promise to come back to something. Ask of every creative: if a stranger files this in a folder called Kitchen and opens it next spring, is there anything in it? A finished room fails that test. A dimensioned plan, a permit card, a priced sample board, a nine-day schedule, a cancellation notice and a year with four slots on it all pass.

That test also sorts the trade's jobs. The single-room work — the hall bathroom, the one wall, the kitchen refresh — is what this app's strongest bands actually buy and what a single crew can actually schedule, so it earns more of the budget here than it does on the other page. The whole-house and addition work is not absent; it is slower, and on this app it is a two-year relationship rather than a campaign.

Each row pins four decisions: 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 this batch used exactly once. Two do not, and an absence is more honest than borrowing a neighbouring ad.

One warning before the table, and it is the same one the Facebook page gives because the policy does not care which app you are on: the aging-in-place row sells on the strongest emotional argument in the trade and every natural sentence for it is the prohibited one. Meta's personal attributes standard bars content that asserts or implies personal attributes, names disability and physical or mental health, and extends to implying you know "the personal attributes of their family". The Facebook batch burned that violation into an image; this batch briefed the fix — describe the room, never the person — and it held across every artefact. The creative section has the prescription.

The cage is the same one the Facebook page describes, so take it briefly. Every campaign created through the Marketing API declares a special ad category and remodeling is not one of them, so the ordinary answer is NONE and a remodeler keeps every lever: age, ZIP codes, neighbourhoods, a radius down to 0.63 miles, lookalikes, saved audiences, exclusions. Advertise the project on a monthly payment and that campaign is a financial products and services campaign, which removes location exclusion, ZIP, neighbourhood, sub-city and metro-area targeting, lookalikes and saved audiences, fixes age at 18–65+ and sets a 15-mile floor. The category is declared on the campaign and Instagram placements sit inside it, so moving a financed ad here changes nothing except the reach — Meta's own documentation says the restrictions apply "across all ad buying surfaces, including the Marketing API".

The lever this page is actually about is not in the picker at all. It is the audience you accumulate: a saved-and-visited pool with a retention of 730 days, built from the five documented Instagram business profile events, which is the only asset in a remodeling account whose memory is as long as the decision. Two rules in one audience — `ig_business_profile_ad_saved` and `ig_business_profile_visit` — is a union, because "Users that match at least one of the rules will be added to the audience". Build a second, shorter one at 30 days from the same events for the people who are moving now. The retention window is the design decision, and it is the only place in this account where you get to choose how long your business remembers somebody.

Then the question the to-do list for this page said to check rather than assume, and the answer is a genuine gap in the documentation. Meta's special ad category page lists supported and removed features explicitly: Custom Audience Inclusion, Custom Audience Exclusion, Custom Audience Expansion, Advantage+ Audience and Detailed Targeting Expansion are supported; Saved Audiences and Lookalike Audiences are removed. Nowhere on that page does the word "engagement", the word "Instagram" or the word "retention" appear, and the one eligibility check it documents — `is_eligible_for_sac_campaigns` — is described for a *customer list* custom audience. So a 730-day engagement pool is neither named as restricted nor named as permitted, and unlike a customer list there is nothing to pre-check. The practical answer: it is the only precision you can plausibly keep inside a declared financing campaign, so build it, attach it, launch small and read the error before you assume it is delivering. Error 2909035 is what comes back for a removed audience type, and 2859024 — "Certification Required" — is the business admin who has not yet accepted the non-discrimination policy.

Now the parameter, because a pool nobody constrains is not a pool. Advantage+ audience expands "the eligible audience pool the delivery system can search", and the things it will not expand are published: "location constraints, minimum age, language, and custom audience exclusions". Inclusions are not on the list. From v23.0 the `advantage_audience` value defaults to 1 or must be stated explicitly on a new ad set, and the change does not apply when you update an old one — so an ad set built last year carries whatever it was created with, and one built today is expanding unless you said otherwise. If the tier is meant to spend on the people who remembered you, set it to 0 and check it after every duplication.

Which leaves one genuinely useful asymmetry: exclusions are hard everywhere. Excluding the pool from the visit-buying tier is honoured even under Advantage+ audience, and so is excluding your active jobs from everything — a homeowner three weeks into your demolition should not be seeing the entry offer. In an account where inclusion is a suggestion, the exclusion list is the only instrument you fully control.

The rest is geography and it is the Facebook page's geography, with one adjustment. A remodeler's audience is a housing stock rather than a radius — the post-war capes with the galley kitchens, the subdivisions whose builder kitchens are thirty years old — and that list is worth building once from the neighbourhoods your own signed jobs came from. The adjustment is the band: this app's strength is where the single-room work lives, so keep the whole-house geography for the Facebook budget and let the offers here match what a household in its thirties and forties actually buys next. Band the age to the offer, and never name a band in the copy, because age is on the personal attributes list.

This page has a destination its Facebook twin does not, and it is the reason the funnel below has a stage that buys an audience. An ad set can name `INSTAGRAM_PROFILE` as its destination — `FACEBOOK_PAGE` and `INSTAGRAM_PROFILE_AND_FACEBOOK_PAGE` are the neighbouring values — and pair it with the profile-visit goal, so the click lands on your own grid and the visitor becomes addressable for two years. For a remodeling company that is not a consolation prize: the grid is the portfolio the trade actually has, and a household eighteen months out would rather scroll nine finished bathrooms than read a landing page.

It has one requirement and it is absolute. The profile is the landing page, so look at your last nine posts as a stranger would. If they are a logo, a holiday graphic, a hiring notice and four reposted memes, do not buy the visit — you are paying for a bounce and poisoning the only audience that matters, because the person who arrives and leaves is in the pool for 730 days either way. Nine good tiles first, then the buy.

A landing page is right for everything that argues with a document — the allowance sheet, the permit card, the day-by-day schedule, the price bands — and the rule from the other page holds without change: an ad whose premise is paperwork should produce the paperwork on the next screen. Never the homepage. Somebody who tapped an ad about allowances and landed on a rotating hero of finished kitchens has been asked to start over.

Instant Forms are for the consultation tier, and the mechanism is the Facebook guide's rather than this one's: conversion leads optimisation only works on Lead Ads, so a website form forecloses the only lead-quality instrument Meta sells for a sale that closes later. Use "Higher intent", add three or four qualifying questions, store the Meta Lead ID, and remember the asymmetry this whole page is built on — the lead is the conversion and the profile is the memory. Pointing those forms at the 730-day pool rather than at cold geography is what makes the third tier worth its budget.

The fourth option is the thread, and in this trade it earns its place for one reason: photographs arrive in it unasked. A cracked tile, a cabinet run measured badly, a phone held up at a ceiling stain. That is genuine triage and it is worth having where somebody actually answers and the conversation ends with a date. It is not a dispatch system and it is not a destination for an ad nobody is staffed to answer.

Every rule on this page comes back to one test, because the account is built on an event you cannot buy: will a stranger open this again in eleven months? That is what a save is, and it is why the creative here is not the Facebook creative reformatted. A finished room fails the test — it is beautiful, it is the category default, and there is nothing in it to return to. A document passes: a dimensioned plan, a permit card, a priced sample board, a counted schedule, a cancellation notice, a calendar with four slots on it. Seven of this batch's ten are objects a person would keep, and those are the ads worth the render.

The second consequence of the save is that the ad is read with no context. A saved post is private, the poster is never told, and collections can be shared, so the version that eventually matters is the one opened in a folder next spring, possibly by the other decision-maker, with no landing page, no targeting and no memory of the campaign. The Facebook page reaches the same instruction from forwarding and screenshots; here it arrives from the folder. Put the company name, the offer and the number the argument rests on inside the frame.

Then the batch's own experiment, which was a counting test and produced the most precise result in this corpus. Six of the ten briefs pinned a countable and closed it against a successor — nine strips of tape and no tenth, five samples and no sixth, four inspection sign-offs, six numbered rows and no seventh, four sticky notes and no fifth, two aisle dimensions and no other number in the frame. Three came back exact. One held where it was pinned and not around it: the planner carried its four notes and the prompt added three more above the calendar, so pin the object, not the frame. One refused the premise in an instructive way — asked for five samples on a table, the concept stage drew one tile and labelled it SAMPLE 2 OF 5, which means the pipeline does not merely render typography more reliably than objects, it converts counts into typography when it can. And one lost outright: an FAQ card briefed for exactly three numbered lines came back with one expanded row and three collapsed ones, because a count of the picture's own structure loses to the visual direction, which owns the structure.

The tenth separated the two halves cleanly and gives the rule its final form. Briefed with two doorways, one relocated appliance and two aisle dimensions, it rendered exactly two doorways and exactly one appliance — and replaced both dimensions with a different set of its own. The shape of a count survives this pipeline; the numbers inside it do not. Brief what must be in the frame, then read what came back before you believe a figure in it.

Where a brief gets rewritten is the stage nobody looks at, and meta.json proves it per ad. Every prohibition this batch lost was lost at the image-prompt model and then rendered faithfully: "no tools in the frame" against a prompt that wrote a marker onto the hallway floor, "callouts to the four sign-offs and nothing else" against a prompt that wrote five, "no figures or dates" against a prompt that added a date line. It can also replace the argument outright — a briefed comparison between a four-inch curb and none came back as before-tile-versus-after, with the comparison gone. Two other stages invent differently: the image model promotes a prompt's "initials" into four full surnames on a government permit and letters sticky notes with real town names, and the copy stage reads the render and asserts those inventions as fact about the business. An invention that enters as pixels leaves as a service area two stages later. Read the prompt in meta.json before blaming the image model.

Arithmetic has a rule now and this batch supplied the control for it. The floor-plan ad's numbers were written by the image-prompt model — PATH 38", DOOR SWING 35", and a flagged 3" CLEAR — 36" REQUIRED — and the copy stage, generating separately, read the render and wrote "You need 33 inches of walkway back". Both subtractions are right. The cost-per-day ad had every figure supplied and returned all four exactly in three separately generated artefacts, and the arithmetic closes: 22 × 365 = 8,030, and 8,030 × $8.00 = $64,240. Set that beside the Facebook twin's financing ad, which computed an amortisation and rendered $609 against a true $626.17 four inches tall, and the line is not "it cannot do arithmetic" — a subtraction lands, a formula does not. Do anything with a rate in it on paper and put the answer in the brief.

Two prescriptions from other trades replicated here and are worth stealing outright. Name the advertiser's role and close the doors it might walk through — the design-build company that will do the work, not a consumer group, an inspector, a lender or a review site — and the comparison ad that came back as a neutral analysis service on Facebook comes back in the contractor's own first person here. And brief the room rather than the person: the curbless-shower ad carries no person in the image and none in five headlines and five primary texts, which is the compliant version of the trade's best emotional offer and the fix for the violation the Facebook batch burned into a frame.

Shapes and lengths, measured rather than assumed. Render 4:5 at 1440 × 1800 for the feed, hold the crop to 1% rather than Facebook's 3%, mind a minimum width of 500, and keep 9:16 for Stories and Reels, where three of these ten sit. Do not buy 1.91:1: the surfaces built to want it are Facebook's right-hand column and search results. Write the headline to 40 and check it against 27 if the same ad runs on Facebook Feed — every one of this batch's fifty headlines fits 40, the longest at 39, and five fit 27, though none of the ten lead headlines does. Write the 30-character description even though this app will not show it; nine of these ten fit. And expect long primary text unless the picture is carrying the whole argument: five of 400 published here fit 125 and all five are the floor-plan ad.

Three stages and a declared financing line, and the first stage is unusual enough to be the reason this section exists: it buys an audience rather than a result. Everywhere else in this cluster the top of the funnel is a teaching buy judged on whether the readers come back. Here it is a membership purchase — a tap puts somebody in an addressable pool for up to two years, which is roughly the length of the decision they are making, and whatever that tap costs you it is not what a held consultation costs.

The second stage is the asset and the third is the only one judged on money. Keep the whole structure this thin on purpose: every extra ad set divides the same small pile of evidence, and in this trade the pile is smaller than in any other on this site. Variation belongs at the ad level inside one ad set, and the thing that varies is the offer.

One structural warning that applies to the second and third stages and nowhere else. Both of them target the pool, which means both of them have to be built with expansion turned off — `advantage_audience` set to 0 — or they are cold ad sets wearing a retargeting label. Check it after every duplication, because a copied ad set is a new ad set and the default is the expanded one.

Do the arithmetic before copying anybody's number. Meta's delivery system wants 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 it is being given. Take your own cost per event and multiply by 50, and that is the weekly spend at which one ad set delivers with full information.

Now run that twice, because this is the page where the answer differs by stage and it is the most practical thing in the section. A held consultation in remodeling is expensive and rare: a company doing twenty-five jobs a year cannot produce fifty held appointments a week at any spend, so the consultation ad set will sit in learning limited more or less permanently and no budget fixes it. Run the same multiplication on your own cost per profile visit and the answer comes out somewhere a real budget can reach, because a visit is a tap rather than an appointment in somebody's kitchen. So the visit tier is the one ad set in this trade — on either app, in either direction — that can genuinely clear the bar, and that is where the optimisation weight belongs. Ask the cheap stage to be optimised and the expensive stage to be aimed.

The corollary is a structure most accounts get backwards: this is the one guide in the cluster that recommends an app-specific ad set, and only one. The visit tier is app-specific by construction, because its destination is a profile on this app; everything else should keep Advantage+ placements and pool its evidence across all thirteen surfaces. Restricting placements anywhere else is buying the learning-phase problem on purpose.

Then the thing that makes this budget different from every other one in the cluster: the asset accrues and the spend does not come back this year. A pool at 730 days is, by construction, a bet whose payout arrives after most marketing budgets have been re-decided twice. Treat the size of the pool as a reported number in its own right — how many people entered it this month, and how many are still in it — and judge the money the way the Facebook twin says to: this quarter's spend against the contracts it produced two quarters later, from your own records, because no setting in the platform will produce that report.

Two habits keep it survivable. Decide the trial length in advance and in writing, and make it longer than your own sales cycle: the first ninety days will show cost, visits, saves and appointments and no revenue, by construction, and that is exactly when accounts get switched off. And keep the seasonality offset in mind, because the lag runs both ways here — spring kitchens are autumn contracts, so the autumn budget buys the spring calendar, while the pool you fill this winter is buying a kitchen two Christmases from now.

The first four are all versions of one error — assuming the thing you want is a thing you can buy — and they are first because every one of them is invisible in Ads Manager, which will report a perfectly healthy campaign doing none of what you think it is doing.