Nathan Smith is the founder of Arrim, an AI creative system for ecommerce brands. Smith's work focuses on using AI to produce ad concepts and building better systems to test which creative ideas perform.
Nathan Smith is the founder of Arrim, an AI creative system for ecommerce brands.
In their LinkedIn posts, they argue that better advertising comes not from producing more AI ads, but from testing ideas carefully and learning from results.
They write about how brands can build a steady creative pipeline, isolate variables such as ad hooks, and avoid mistaking noisy results for reliable winners.
Follow their work for practical views on ecommerce ad testing and how AI changes creative production.
Smith connects creative strategy to operating decisions.
For example, they recommend changing only an ad’s opening hook while keeping the rest constant, so a test can show whether the hook made a difference.
They also argue for setting a clear hypothesis, allowing enough spend to gather a meaningful signal, and resisting daily reactions to volatile results.
This is more specific than a general call to use AI or make more content: it explains how teams might turn production into a learning process.
Their posts also frame creative as a supply chain, with concepts entering tests on a planned cadence rather than relying on one winning ad.
These are recommendations, not independently verified performance results.
Nathan Smith's public writing returns to one priority: ecommerce ad creative should be run as a learning system.
Read more
Their argument is that AI has made ad production cheap enough that volume is no longer the advantage; the scarce skill is deciding what to test, what each result means, and what to do next. Around that core, they argue for hook-first diagnosis, creative operations managed like a supply chain, and AI products whose durable value sits in accumulated data and outcomes rather than generation.
Volume is not the advantage; interpretation is
Smith argues that brands can now generate many AI UGC variants quickly, but most lack rules for reading the results.
Read more
Testing dozens of creatives with fragmented spend produces noise: no variant gets enough conversions for a reliable read, losers are killed on a bad day, and a lucky winner gets scaled.
The fix is fewer meaningful tests, each built around one hypothesis, with enough spend to gather a signal, hook categories tested first, and a fixed read window.
In their phrasing, a volume-only testing system is a 'content factory wearing a media buying costume.'
The worked failure is twenty creatives with spend split across them, none producing statistically significant conversions, followed by scaling a winner after two lucky conversions.
When a UGC ad underperforms, Smith says brands often reject the whole thing: product, script, offer, and face.
Read more
The body may be fine; the first three seconds may be the problem.
Their method is to lock the body, vary only the opening line, ship five to ten hook variants, and scale the one that holds attention.
Static ads can validate a new angle faster and cheaper than video, making them a front-end test.
Use the same product demo and offer with different opening lines in five to ten variants. The test identifies which hook won instead of declaring the creative a failure.
Smith contrasts brands that ask 'What's our best performing ad right now?' with scaling brands that never ask it.
Read more
A winner is a snapshot of last week, not a durable strategy.
Their operating shifts: measure inputs, such as how many new concepts entered testing; forecast creative demand ahead of need; let cadence drive demand instead of last month's numbers; and accept a low hit rate by design.
If eight of ten ads fail, they argue, the system is working, not having a bad month.
Forecast creative demand weeks ahead so the pipeline never runs dry, rather than waiting for a winner to fade before briefing replacements.
Producing fifty concepts once took weeks of filming, editing, and creator management; now it takes a fraction of that time.
But because everyone can generate, the advantage is choosing which products to advertise, which audiences to target, which hooks stop the scroll, and which creatives deserve spend.
Arrim is described as a system that connects a product link or store, learns the brand, creates content, connects to Meta Ads, and uses 'Arrim Intelligence' to learn what converts.
Arrim's five steps: paste a product link or connect a store, learn the brand, create content, connect to Meta Ads, and feed conversion learning back into more content.
The AI video post frames a tension, not a settled prediction: whether AI will replace creative teams or change how they work.
Build AI where the value compounds, not where it can be copied
In a build update, Smith reports Arrim reached six figures ARR in under four months, solo and without funding, then says revenue is not the main point.
Read more
Their argument is that traction without defensibility is a countdown: anyone can ship a polished AI product, and a surface on someone else's model is a feature waiting to be absorbed. They cite Anand Arivukkarasu's 'Supply Chain of Intelligence' framework and name Jasper, Gamma, and Chegg as 'great products, wrong layer.' The durable value, they argue, sits below generation, in proprietary data, outcomes, and learning that compounds with use.
Smith cites Anand Arivukkarasu's framework rather than presenting it as their own idea, and frames Arrim's bet as owning data and learning that a foundation model cannot reproduce by adding a button.
Answers the two questions a newcomer will have first: what has this person built, and what do they believe about it?
Read more
Here, Smith states the thesis behind Arrim.
A build update that turns into an argument.
Smith reports taking Arrim to six figures in annual recurring revenue in under four months, solo, organically, and without funding, then says revenue is the less important part.
They argue that AI products built as surfaces on top of foundation models have no moat: the model layer keeps getting cheaper, and a polished wrapper is "a feature waiting to be absorbed."
They cite Anand Arivukkarasu's "Supply Chain of Intelligence" framework and name Jasper, Gamma, and Chegg as "great products, wrong layer."
The claim that endures is that durable value sits below generation, in proprietary data, outcomes, and learning that compounds with use.
For brands planning a campaign
Explains why Smith positions Arrim as a learning system rather than a generator: accumulated conversion data, not raw output, is the product’s differentiator. Useful when comparing Arrim with prompt-to-video tools that lack a learning loop.
Open it before weighing the ideas to understand what Smith built.
A short mission post.
Smith describes the process in five steps: paste a product link or connect a store; Arrim learns the brand; creates content; connects to Meta Ads; and "Arrim Intelligence" learns what converts so it produces more of what works.
They call the company "the creative engine for modern e-commerce brands" and say they are building "far beyond an AI video generator."
The post sets out intent rather than demonstrated results.
For brands planning a campaign
Names the exact handoff points a brand would test in a pilot: brand ingestion, content generation, the Meta Ads connection, and the feedback loop. Ask for evidence at each step rather than accepting the end-to-end claim.
Answers what Smith thinks is broken in ecommerce creative: volume without decision rules.
Read more
This is their sharpest diagnosis.
An argument that creative volume is the wrong goal.
Smith walks through a familiar failure: generate dozens of AI UGC variants overnight, spread spend across twenty creatives, and none gets enough conversions to be statistically real.
Brands kill losers off a bad Tuesday and scale a winner that got lucky twice.
They argue that access to volume "used to be the moat" and everyone has it now, so the scarce skill is interpretation.
The alternative is fewer, cleaner tests: one clear hypothesis per test, enough spend for signal, hook categories first and depth on the winner later, and results read on a three-to-seven-day window instead of daily panic.
The closing line — "a content factory wearing a media buying costume" — captures the argument.
For brands planning a campaign
A testing checklist brands can use: one hypothesis per test, a spend floor for significance, test hook categories before refining a winner, and a fixed read window. The three-to-seven-day window is a concrete default.
Answers the practical question behind every underperforming ad: when it fails, what should change?
Read more
This is the hook-first method in detail.
A short method post.
Smith reframes underperformance as a "diagnosis problem": brands shoot one UGC ad, it underperforms, and they kill the product, script, offer, and face together.
They argue that half the time the body was fine and the first three seconds were the problem.
The recommended process is deliberate isolation: lock the body, change only the hook, ship five to ten variants, and scale the one that holds attention.
Same product demo, same offer, different opening line.
A failed ad becomes a signal about which hook won.
For brands planning a campaign
A reusable production rule: hold the body constant and vary hooks in batches of five to ten, so results can be attributed to the hook rather than the whole ad. It is low-cost to adopt and fits Arrim's AI-variant approach.
Explains how scaling brands run creative operations differently, in five operating shifts.
Read more
A good next read after the diagnosis posts.
A five-point list built around one observation from discovery calls: every brand asks which ad is performing best, but scaling brands never do.
The shifts are: stop chasing winners, because a winner is last week's snapshot; measure input rather than outcome by tracking how many new concepts entered testing; treat creative like inventory, forecasting demand so the pipeline never runs dry; decouple production from performance data, letting cadence drive creative planning instead of last month's results; and accept a low hit rate on purpose, since eight of ten ads failing is the system working.
The unifying idea: manage creative as a supply chain, with volume in and insight out, rather than betting on one hero ad.
For brands planning a campaign
Gives a brand four operational numbers to set: weekly concepts tested, pipeline lead time, briefing cadence, and target hit rate, rather than a single ROAS figure.
Answers what Smith thinks AI does and does not change for creative teams.
Read more
The clearest statement of where they draw that line.
An opinion post that separates production from judgment.
Smith notes that producing fifty ad concepts once meant weeks of filming, editing, and creator management, while now it happens in a fraction of the time.
But they argue the advantage is not generation, since everyone can generate.
It is choosing which products to advertise, which audiences to target, which hooks stop the scroll, and which creatives deserve spend.
The post ends by asking the audience whether AI shrinks creative teams or makes them more productive, a characteristic way for Smith to prompt replies.
For brands planning a campaign
Frames AI as expanding how many ideas a team can test while leaving selection as the scarce skill. Brands adopting AI creative tools should budget for evaluation capacity, not just generation.
AI lowers production cost but raises the cost of interpretation3
Diagnose the hook before killing the whole ad2
Creative operations as supply-chain management2
Defensibility in AI products2
Creative testing as a system, not a lottery
Nathan Smith argues that most ecommerce brands do not have a volume problem; they have a testing and diagnosis problem.
Read more
The posts describe brands generating dozens of AI ads a week, spreading spend across many creatives, killing losers on a bad day, and scaling a winner that got lucky.
The alternative they set out is fewer, clearer tests: one hypothesis per test, enough spend for signal rather than noise, hook categories first, and results read over a 3–7 day window.
The same theme appears in the claim that testing more beats spending more and that the best brands treat creative like a supply chain rather than a bet.
For brands planning a campaign
For a DTC brand improving Meta ad creative testing, Smith is a fit to evaluate as a systems thinker. Ask for the decision rules, test budgets, and read windows before commissioning creative: the value lies in making tests interpretable, not simply producing more ads.
One recurring method is to separate the body of an ad from its opening.
Read more
Smith writes that when a UGC ad underperforms, brands often kill the product, script, offer, and face together, even though the first three seconds may be the failing part.
The recommended approach is to lock the body, change only the hook, ship 5–10 variants, and scale the one that holds attention.
A related post adds that static ads can validate a new angle faster and cheaper than video.
Lock the body; vary only the opening line; ship 5–10 hook variants; scale the variant that holds attention; use statics to validate new angles before video production.
For brands planning a campaign
This gives a brand a concrete, low-variance way to learn which hook won instead of guessing which creative failed. It is especially useful when a product demo and offer are already working but attention is the bottleneck.
Smith describes scaling brands as treating creative like inventory rather than a single hero ad.
Read more
The shift is to stop chasing last week's winner, measure inputs such as how many new concepts entered testing, forecast creative demand weeks ahead, decouple production from performance data, and accept a low hit rate on purpose. In this view, a “winner” is a snapshot of stale data, and waiting for results before briefing next month's ads keeps a brand a full cycle behind.
Track new concepts entering testing; forecast creative demand ahead of need; brief on cadence rather than last month's performance data; plan for a low hit rate instead of treating it as a failed month.
For brands planning a campaign
A brand can evaluate Smith on whether its creative pipeline is designed as a repeatable operation. They are a fit for accounts where testing cadence, not one-off creative quality, is the constraint.
AI lowers production cost but raises the cost of interpretation
Smith repeatedly separates AI generation from creative strategy.
Read more
AI can produce many ad concepts quickly, but the advantage lies in knowing which products to advertise, which audiences to target, which hooks stop scrolling, and which creatives deserve spend.
They describe AI UGC volume without decision rules as a content factory wearing a media buying costume.
They also describe how Arrim connects a product link or store, learns the brand, creates content, connects to Meta Ads, and uses Arrim Intelligence to learn what converts so it can create more of what works.
For brands planning a campaign
If a brand wants AI to support faster iteration, Smith's work suggests evaluating the feedback loop, not the generator alone. Ask how the system selects hooks, allocates spend, and feeds conversion learning back into the next round of creative.
Smith writes about building Arrim to six figures ARR in under four months, solo and without funding, and why traction without defensibility is a countdown timer.
Read more
Their argument is that generation commoditizes: a surface on someone else's model is a feature waiting to be absorbed. Durable value sits underneath generation, in proprietary data, outcomes, and learning that compounds the longer a product runs.
For brands planning a campaign
For a brand assessing AI vendors, this is a useful lens: ask what proprietary learning or outcome data the product accumulates, not just what it can generate. It also positions Smith as someone who thinks about product durability, not only campaign tactics.
Smith’s publicly documented business relationship is Arrim, the AI creative system they say they built and run for ecommerce brands.
Read more
The posts also show Arrim working with Meta Ads as a connected ad platform.
They don’t identify sponsorship, affiliate, or client relationships with outside brands; ask about those before booking.
Several brands appear only as passing examples in their commentary.
Arrim
Owned business
Smith says they built Arrim, an AI creative system that learns a brand, generates content, connects to Meta Ads, and helps ecommerce brands test and scale ad creative.
Read more
They say they started it to solve a problem they faced in ecommerce and reached six figures in annual recurring revenue in under four months, solo and without funding. The posts present Arrim as their own product, making the relationship one of ownership rather than client work or sponsorship.
Smith says Arrim connects to Meta Ads so the system can learn what converts.
Read more
Their writing repeatedly uses Meta ad accounts to illustrate creative-testing problems. Meta is a product integration and recurring subject of their work, not a stated sponsor or formal partner.
Nathan Smith's posts make a running argument about how ecommerce brands should decide what to test, rather than presenting a feed of finished ads.
Read more
The clearest entry points split an ad into hook and body, treat creative as a supply chain with forecasting and a planned low hit rate, or distinguish AI generation from the judgment about which ideas deserve spend.
The Arrim build updates explain where Smith sees durable value in an AI creative product, adding context to the testing advice.
For the testing thinking, start with the hook and supply-chain posts; for the product story, start with the Arrim updates and work back.
The tags below describe the subjects and format, and the comparable accounts that follow cover overlapping ground, each with a stated difference.
Soriano also refuses to throw away an ad when one part stops working.
Read more
They use hook rate, hold rate, CTR, and conversion to locate where viewers leave, then re-cut the opening using footage the brand already paid for. It's the same hook-first move Smith describes: lock the body and vary the opening line.
Soriano's posts are diagnostic and service-shaped: a numbered check against the brand's own account average, plus an offer to reply with a rewritten hook. Smith writes frameworks for testing cadence and creative operations, connecting the same hook idea to AI production and a product they are building.
Novinsek makes the same cost argument for hook variants on an existing body: one creator-shot body costs roughly $500 and a brief cycle, while ten hook variations on that body cost minutes in an editor.
Read more
That supports Smith's case for changing only the opening before killing an ad. Novinsek also argues that the popular claim that hook testing no longer works is wrong, matching Smith's position that hook-level learning is the cheap test.
Novinsek's evidence comes from agency operations.
Their posts cover UGC managers breaking at a client count, unpredictable creator pricing, and where in Ads Manager to run a test and which metric to read.
Smith writes from the brand side about designing a testing system and reading results, with added coverage of AI production and product strategy.
Shakoor also treats the account as a pipeline of tested angles rather than one winning ad, arguing that a single winning creative carries the account until it tires and that brands should already have the next test running when the current winner fades.
Read more
That echoes Smith's supply-chain framing of creative demand and their warning that a winner is a snapshot of last week.
Shakoor's posts are practitioner accounts: what working on DTC brands and Meta ads looked like across three years, plus a reflective piece about stepping back and choosing a direction. Smith's posts are argument-led, with a stated method; the Arrim build updates also offer a view on AI product defensibility.
Nathan Smith's posts target DTC ecommerce brand owners and operators running Meta ads, with topics including creative testing systems, hook variants, AI-generated UGC, and the cost of interpreting test results.
Read more
That topical match is close to the stated goal, but does not establish who reads the posts.
Reach is modest: 5,632 LinkedIn followers recorded on October 9, 2026.
The posts include no impression, reach, or engagement figures, so they don't show how many people saw a given post, which formats travelled furthest, or whether readers are the brand-side operators they address rather than other agency people.
They are a reasonable pick for a peer-to-peer argument about creative testing aimed at ecommerce operators, and a weaker pick for broad reach to a validated audience of a particular size, seniority, or geography.
Ask for recent performance on comparable ecommerce and Meta creative topics—reach and engagement on the last few months of posts—and any audience survey or page analytics showing job roles, company size, seniority, and geography.
No partnership or sponsorship formats are described in the posts here.
Read more
Nathan Smith’s posts promote Arrim, the AI creative system they founded, and several close with an invitation to work with the company rather than with them as an independent creator.
A September 10, 2026 post pitches brands “struggling with creative bottlenecks” (“We can help”); a September 8, 2026 post has a similar close; and a September 20, 2026 post frames creative testing as a service problem Arrim addresses.
These are sales calls to action for a software and services business, not a defined creator offering such as sponsored posts, paid placements, or brand ambassador arrangements.
For a DTC brand planning a campaign, collaboration would likely run through Arrim as a vendor relationship.
Ask directly about rates, deliverables, exclusivity, disclosure practices, and the partnership contact route; the posts do not specify them.
Fit if the brand treats any engagement as a vendor relationship through Arrim, not a paid creator placement, and accepts a modest, unmeasured audience.
Read more
Brief
A DTC ecommerce brand running Meta ads wants to rebuild its creative testing system and use AI to iterate faster, starting with the testing framework itself.
Reasons
The brief addresses the exact problem in this material. Nathan Smith argues that most ecommerce brands have a diagnosis problem, not a volume problem: they generate dozens of AI ads a week, spread spend across twenty creatives, kill losers after a bad Tuesday, and scale a winner that got lucky twice. That is the failure mode the brief aims to fix.
The framework breaks down into testable steps: set one clear hypothesis per test; spend enough to get signal rather than noise; test hook categories first, then go deep on the winner; read results over a 3-7 day window instead of reacting daily; and, for an underperforming ad, keep the body fixed and change only the opening line, shipping 5-10 variants so the result reflects the hook rather than the whole ad. Static ads come first as a cheaper, faster way to validate a new angle before video spend.
Creative is treated as an operating system, not a bet on a hero ad. The recommended shifts are to stop chasing last week's winner, measure inputs such as how many new concepts entered testing, forecast creative demand weeks ahead, brief on cadence rather than last month's performance data, and accept a low hit rate by design: eight of ten ads failing means the system is working.
AI is framed around campaign needs. Production has become cheap and interpretation expensive: anyone can generate an AI video, so the advantage lies in choosing which products to advertise, which audiences to target, which hooks stop the scroll, and which creatives deserve spend. The Arrim process connects a store, learns the brand, creates content, connects to Meta Ads, and feeds conversion learning into the next round of content.
The writing targets DTC ecommerce brand owners and operators running Meta ads. Reach is modest and unmeasured: the only audience figure is 5,632 LinkedIn followers recorded on 9 October 2026. Ask for impression, reach, engagement, and demographic data before relying on this as a reach buy. This is a peer-argument fit, not a guaranteed-reach fit.
Limits
Arrim's product claims, including the Meta Ads connection and learning loop, come from Nathan Smith's own posts. They describe no product documentation, customer reference, or independently tested result. Before commissioning creative, the brand should ask for the decision rules, test budgets, spend floors, and read windows in writing.
Ask how many variants entered testing last month, how far ahead the creative pipeline runs, and how conversion data changes the next brief. Those answers show whether the claimed system exists in practice.
When to choose a specialist
Poor fit for a campaign seeking a commercial creator offering and a measurable audience; neither is documented in Nathan Smith's posts.
Read more
Brief
A national consumer brand wants a paid creator partnership for a category launch: a fixed fee for a defined set of sponsored posts, exclusivity and disclosure terms, and delivery to a large verified audience with demographic guarantees.
Reasons
Nathan Smith's posts document no sponsorship format, rates, deliverable specifications, exclusivity terms, disclosure practice, or partnership inquiry route. Their commercial calls to action point to Arrim as a software and services vendor, not a creator offering; one post closes with a service pitch offering to help brands with creative bottlenecks. Ask for the placement terms before planning a fixed-fee campaign.
Audience reach and composition are unmeasured. The only figure available is 5,632 LinkedIn followers recorded on 9 October 2026. There are no impression, reach, or engagement figures, audience survey, or breakdown by job role, company size, seniority, or geography. Ask for audience analytics before planning a launch around audience size or makeup.
The subject matter is narrow relative to a general consumer launch. The posts cover creative testing systems, hook variants, AI-generated UGC, Meta ad operations, and AI product defensibility. A category launch aimed at general consumers or a non-ecommerce audience has little overlap with that material.
Seven short LinkedIn posts are enough to judge the ideas and the method. They are not enough to underwrite a paid placement with reach or performance guarantees.
Limits
Smith's posts do not provide sponsorship information. This is not a claim that Smith never accepts sponsorships or never works with brands on commercial terms outside Arrim. If the brand wants a paid placement, the first step is to ask directly for a rate card, deliverable list, disclosure policy, and audience analytics; Smith's posts do not answer those questions.
A worked campaign example
Angle
Run a hook-isolation sprint. Treat the last quarter of ad failures as a diagnosis exercise, not a verdict on creative quality, then rebuild testing so each batch answers one question about the opening three seconds. The worked campaign angle is a six-week program with four moving parts: a diagnostic, cheap angle validation, hook-isolation batches, and fixed read rules.
Brief
The same DTC ecommerce brand, six weeks, goal: a repeatable Meta creative testing system that uses AI to increase variant volume while keeping decision rules human.
Deliverables
A hook library of tested opening lines, each with the read-window data behind it.
A written set of decision rules: spend floor per variant, 3-7 day read window, kill criteria, escalation criteria, and the number of new concepts entering testing each week.
A monthly testing calendar that runs on cadence rather than last month's performance data, with statics used to screen new angles before video spend.
Handoff
Before commissioning creative, ask for the decision rules and test budgets in writing, which conversion data the system retains, and how that learning changes the next brief. Without those answers, treat the engagement as creative production with commentary, not as a testing system.
Steps
Week 1 - diagnosis. Pull the last 20-30 ads and sort failures into two piles: ads where the product, offer, or body was weak, and ads where only the first three seconds broke. Group the second pile into hook categories such as problem call-out, result-led, objection-handling, and comparison. This creates a hypothesis backlog rather than a list of dead ads.
Week 2 - validate angles with statics first. Take the strongest three or four new angles and test them as static ads, which the methodology says can validate a new angle faster and cheaper than video. Promote only the angles that earn a place in video production.
Weeks 3-5 - hook-isolation batches. Keep the product demo, offer, and body fixed. Change only the opening line and ship 5-10 variants per batch, so the result shows which hook won rather than whether the whole ad worked.
Weeks 3-5 - set spend and read rules. Set spend per variant high enough to generate a signal, use one hypothesis per test, and read results within a fixed 3-7 day window. Don’t kill a variant after a single bad day. Expect most variants to lose: eight of ten failing means the system is operating, not that it’s been a bad month.
Weeks 1-6 - track input metrics: the number of new concepts entering testing each week, how far ahead the creative pipeline runs, and how many tests produced a usable learning, even when the ad lost. These numbers show whether the system is working better than a single ROAS figure for one ad.
Week 6 - close the loop. Record which hook categories held attention, which angles failed on first exposure, and what the next batch should test. Use AI for volume and asset variation, not decision-making: production is cheap now, and the scarce work is interpreting which hook deserves more spend.
Choose John Soriano when a specific ad is already running and losing, and the immediate need is a diagnosis rather than a new testing system.
Read more
The material gives a four-number check to find where viewers leave: a low hook rate means the first three seconds failed and the body can be kept; a good hook rate with a low hold rate means the middle drags; a good hold rate with low click-through means the offer or call to action is the problem; and good click-through with low conversion points to the landing page.
It also gives a five-cut re-edit list that turns one existing raw UGC clip into five testable ads: swap the opener, lead with the problem, product, or proof, tighten the pace, and rewrite the on-screen captions.
That is the better choice when the brand has already paid for footage and wants more tests without booking another creator, or wants someone to read one ad and return a rewritten opening line and first frame.
The material works at the level of a single ad and its edit.
It assumes a functioning account and existing assets, but does not provide the per-batch decision rules, spend floor, fixed read window, input metrics, or AI-assisted production loop in Nathan Smith’s material.
A brand that needs to design the whole testing calendar or use AI to increase variant volume gets less from these two posts than from Smith’s material.
Choose Sebastian Novinsek when an agency or in-house team’s UGC operation is breaking under volume, or the brand needs tool mechanics rather than a framework.
Read more
One post describes the pattern of agency UGC scaling: manual processes crack at around eight to ten clients, creator pricing becomes unpredictable and margins shrink, and surviving past that point depends on treating UGC as a system rather than a set of individual freelancer relationships.
The other post argues that Andromeda did not kill hook testing and points to the Creative Testing Tool inside Ads Manager, which forces separate audience cohorts per variant, splits budget across the test set, and concentrates spend on the winner after the window.
It also flags a default key metric that needs switching to cost-per-purchase and puts numbers on the method: about US$500 for one creator-shot body, against roughly ten hook variations on that body in about ten minutes of editing.
That cost comparison gives brands a concrete basis for deciding how to spend the next creative budget.
The strongest material here covers operational and supply-chain economics, aimed at agencies managing many clients and teams deciding how many creator shoots to book.
It is thinner than Smith’s on what the brand decides per batch and how conversion learning feeds the next campaign, and it does not describe an AI production loop.
The posts discuss tool behaviour and make Andromeda claims.
Choose Ujla Shakoor when the specific worry is concentration risk: an account built around one winning creative, hook, angle, or message, so that when the creative tires, the whole account feels it.
Read more
The argument is that the goal is not to find one winning ad but to build a pipeline of winners, and that the question to keep asking is what is being tested next across different hooks, angles, customer problems, and formats. That framing suits a brand whose spend is concentrated in a single winner and that wants to widen the testing base before the winner dies.
This is the narrowest of the three on the same buyer need.
The pipeline argument comes without spend floors, read windows, kill criteria, input metrics, or an AI component; one of the two posts is a personal reflection on a career pause that does not address the buyer’s task.
A brand that accepts the pipeline idea but needs the rules and AI-assisted iteration gets considerably more from Smith.
Nathan Smith's seven LinkedIn posts cover Arrim product updates and ecommerce ad creative testing, including hook testing, creative operations, AI production economics, and AI product defensibility.
The public profile page shows 5,632 followers as of 9 October 2026.
None of the posts carry publication dates, so their ideas are not placed on a dated timeline.
Nathan Smith's posts do not include videos, transcripts, interviews, product documentation, performance data, a rate card, or independent third-party coverage, and the growth and revenue figures for Arrim come from Smith's own posts without independent verification.
The similar-creator section compares three creators, with two posts each: Sebastian Novinsek, dated 28 April and 5 May 2026; Ujla Shakoor, dated 3 and 18 September 2026; and John Soriano, dated 2 and 5 October 2026.
