DEV Community: Sahil Arora

Inheriting an Ad Account: From Two Weeks to an Afternoon

Sahil Arora — Mon, 21 Sep 2026 05:09:27 +0000

Every agency has a version of the same photograph: a new buyer, three monitors, forty browser tabs, and the thousand-yard stare of someone two days into an account somebody else built.

Inheriting an ad account is archaeology. The structure was designed by a person who is no longer reachable. The naming convention made sense to them. There are campaigns that exist "for a reason" nobody can name, creatives that have run for a year without anyone defending them, and one ad set whose budget is suspiciously protected. Your job in week one is not to improve any of it. It is simply to understand what you have been handed before you touch a dial.

For years that took two weeks. Last quarter it took me an afternoon, and the difference was not effort.

What the two weeks actually went on

Very little of that fortnight was analysis. It was reconstruction.
You export campaign data as far back as the platform will give you. You screenshot the structure so you can see it all at once. You build a spreadsheet mapping which creative ran where, because the naming will not tell you. You ask the client questions they cannot answer, because the person who knew left in March. You find a tracking discrepancy and lose a day deciding whether it is real. Somewhere near the end, you form a view.

The output of two weeks was not a strategy. It was a rough mental model of an account that had been sitting there, fully documented in its own data, the entire time. Every answer you needed already existed. It just could not be asked a question.

Asking the account instead of rebuilding it

The shift is that an inherited account can now be interrogated directly rather than reassembled by hand.

The first hour of my last handover went like this, inside Hawky's copilot. What has actually driven results in this account over the last year. What do the winning creatives have in common. What was tried and quietly abandoned. Which spend is concentrated in things that are no longer performing. Each answer came back with the rows underneath it, so I was not taking anyone's word for the state of an account I had just been handed, including the software's.

By lunch I had what the fortnight used to produce: a map of what is load-bearing, a list of things running on momentum rather than merit, and three specific questions for the client that were actually worth their time.

The part I did not anticipate was the confidence that came with it. Walking into a handover call able to say "this campaign is carrying 60% of your results and this one has not justified its budget since spring" changes the relationship immediately. You stop being the new agency asking for context and start being the one that already read the account.

What the account cannot tell you

This is where honesty matters, because the data has a hard boundary.
The account can tell you what happened. It cannot tell you why the last team chose it. There is always a reason behind the strange ad set: a founder's preference, a retail partnership, a legal constraint on a claim, a burnt campaign nobody wants named. That knowledge is tribal, it lives in people, and no amount of analysis will surface it. Ask the humans, and ask early.

You also inherit the previous team's measurement, not the truth. If their tracking was broken or their attribution windows were unusual, every answer you get faithfully reflects a flawed record. Audit the plumbing before you trust the story, especially on any conversion number that looks better than the client's own revenue suggests it should.
And a year of history describes the world as it was tested, not as it is. An angle that never appears in the account may be missing because it failed, or because nobody thought of it. Only one of those is useful to know, and the data cannot distinguish them.

A handover checklist worth stealing

Four questions, in this order, on the first afternoon. What is carrying the results. What is spending without earning. What did they try that stopped. What is different about the markets or segments that perform.
Then two questions for the humans, which the data will never answer. What is untouchable and why. What has been promised to someone that is not written down anywhere.

That is a full handover. It used to be a fortnight.

Closing Thoughts: the account was always talking

The thing I keep returning to is that none of this information is new. It was all sitting in the account on day one of every handover I have ever done. The fortnight was never spent discovering it. It was spent building a machine capable of hearing it.

We do not need that machine anymore. Which means the first week of a new client relationship can be spent on the part that was always more valuable and always got squeezed: deciding what to do about what you found.

What Is AI Competitor Ad Analysis? A Complete Guide for Marketers

Sahil Arora — Wed, 26 Aug 2026 06:30:00 +0000

I used to keep a folder on my desktop called "competitor screenshots." Every few weeks I'd scroll the Meta Ad Library, grab whatever my competitors were running, and file it away with the vague
intention of "learning something." That folder grew to 400 images. The number of decisions it ever changed: roughly zero.

That folder is the before picture. AI competitor ad analysis is the after, and the difference isn't the collecting; it's the comprehension. Let me walk you through what the term actually means, how the technology works, and what it changes in practice.

The Definition, Without the Fluff

AI competitor ad analysis is the use of machine learning to systematically collect, deconstruct, and interpret the ads your competitors run, at a scale and depth no human team can match
manually. The raw material is public: ad libraries from Meta, Google, and TikTok expose what brands are running. What the AI adds is the layer humans never get to: analyzing hundreds or thousands of those ads at once and extracting the patterns underneath.

The distinction that matters: old-school "ad spying" tells you what a competitor is running. AI analysis tells you how it's constructed and what's working: which hooks they lean on, which formats they've scaled, which offers they've quietly stopped promoting, and where the gaps in their coverage are.

What the Machine Actually Sees

This is the part that surprised me most when I first watched it work. A good system doesn't treat an ad as one blob. It deconstructs every ad into elements. I've heard it called creative DNA, and the name fits.
Computer vision reads the visual layer: format, color palette, whether there's a face, where the product sits, how prominent the logo is. Language models read the copy: the hook type (fear,proof, discount, curiosity), the tone, the call to action, the offer. Metadata fills in the rest: how long an ad has been live, how many variants of it exist, which platforms and placements it runs on.

Longevity and variant count are the closest public proxies for success: nobody keeps scaling a loser for ninety days. When an AI clusters 800 competitor ads and shows you that the three
longest-running campaigns in your category all lead with customer testimonials in the first two seconds, that's not a screenshot folder. That's intelligence.

What Changed in My Actual Week

Three concrete shifts, from someone who lived the manual version.

Research went from days to minutes. A category scan that used to be an intern-week is now a query. The first time I ran Hawky's competitor analysis on my category, it came back with a few hundred competitor ads already deconstructed into hooks, formats, and offers, plus one finding I still quote: the two longest-running campaigns in my space both led with the same testimonial structure, something four months of manual screenshotting had never surfaced. And because it watches continuously instead of producing a one-off report, the map stays current: you catch a competitor's new angle the week it launches, not the quarter after.

Creative briefs got evidence. "Make something scroll-stopping" became "our top three competitors all use discount hooks; the testimonial angle is unoccupied, so test that." Writers and
designers, it turns out, love being handed a map instead of a mood.
We stopped copying. Counterintuitive, but real. When you can only see a competitor's five bestads, the temptation is to imitate them. When you can see their entire portfolio mapped, you see
the crowded ground and the open ground, and the open ground is where the cheap clicks live.

The Honest Limitations

AI analysis reads public signals, not private dashboards. It infers success from longevity and investment patterns; it cannot see a competitor's actual ROAS, and anyone who claims otherwise
is selling something. It also can't tell you whether a rival's strategy is smart. Sometimes the whole category is scaling the same mistake together. The machine finds the patterns; deciding which
patterns deserve respect is still your job.

And a warning from experience: this intelligence is a compass, not an autopilot. Teams that mechanically chase whatever competitors do end up perpetually one step behind by design.

Closing Thoughts: The Folder Is Gone

I deleted the screenshot folder last year. Not because competitive intelligence stopped mattering, but because collecting stopped being the bottleneck. The bottleneck now is what it always should
have been: having the taste and nerve to act on what the analysis shows you.
If you're still scrolling ad libraries manually, you're not behind because you lack discipline. You're behind because you're competing against teams whose machines read a thousand ads before breakfast. Close that gap first; the strategy conversations get a lot more interesting afterward.