If your Meta ads are inconsistent, your creative is solid but your ROAS swings wildly week to week, or you feel like you’re talking to everyone and converting no one – the problem almost certainly isn’t your storefront. It’s that you haven’t defined who you’re actually selling to.
Connor Shelefontiuk, founder of More Conversions, a CRO agency working with DTC brands, calls this the Core Avatar – a precisely defined picture of your single most valuable buyer. Not a broad demographic bracket.
Not three personas you rotate through in brainstorm sessions. One person, with specific problems, specific language, and specific reasons to buy from you over everyone else. Watch Connor explain the Core Avatar framework here.
Most scaling brands don’t have this. And it’s costing them more than they realize – at a moment when the cost of getting it wrong has never been higher.
Average ecommerce customer acquisition costs have risen roughly 222% over the past eight years, with a 40-60% increase between 2023 and 2025 alone. When CAC is climbing that fast, a vague offer that converts the wrong people isn’t just an efficiency problem. It’s an existential one.
A Core Avatar is a single, deeply researched profile of the customer your brand is best positioned to serve. It goes beyond age and income bracket into the territory that actually drives conversion:
The reason scaling brands resist this isn’t negligence – it’s a deliberate strategic choice that happens to be wrong.
The logic goes: if you broaden your appeal, you expand your total addressable market (TAM). A bigger TAM means more potential customers, which means more revenue. It sounds reasonable. It doesn’t work.
When you try to speak to everyone, here’s what actually happens:
Narrow your target, and the opposite happens:
The stage of the brand matters here too. It’s not a one-size-fits-all problem:
Avatar research is not a one-time workshop exercise. It’s an ongoing data collection process. The methods depend on where you are in your growth stage, but the goal is always the same: understand your buyer in their own words, not the language your internal team uses to describe the product.
If you’re pre-scale or don’t yet have thousands of customers to survey, you’re not working with your own data – you’re working with the market’s data. Two sources stand out.
1. Reddit
Shelefontiuk calls it “an absolute gold mine” – and the reason is simple: anonymity. People say things on Reddit they’d never put in a survey or a review form. Go into the subreddits relevant to your vertical – supplements, skincare, pet care, outdoor gear, whatever your category is – and read everything Don’t post. Don’t engage as a brand. Just absorb.
You’re specifically looking for:
2. Competitor reviews on Amazon and other platforms
One-star reviews are one of the most underutilized research assets in ecommerce. A negative review tells you exactly what a customer expected, what they didn’t get, and how they articulate the gap.
That’s not just feedback. That’s:
If you can directly address what your competitors’ customers are complaining about, you’ve already built your Core Avatar’s pain point into your marketing.
The goal from both sources is the same: collect enough raw language that patterns start to emerge. The same complaints surfacing across dozens of Reddit threads and hundreds of reviews aren’t noise – they’re the signal your entire offer should be built around.
Once you have enough customers to hear from – Shelefontiuk suggests 5,000 to 10,000 as a reasonable floor – surveys become your most powerful research tool. The key is running them at three distinct moments in the customer journey, not just one.
1. Abandoned Cart Survey
Most brands send abandoned cart emails. Almost none include a survey. That’s a significant missed opportunity.
When someone abandons checkout, you have a narrow window to understand why. A simple email offering a small incentive in exchange for three quick answers can yield more actionable data than months of A/B testing.
Ask:
The answers tend to cluster. Shelefontiuk describes an 80/20 pattern: 80% of abandonment reasons fall into one or two categories. Once you see those clearly, you can test offer adjustments that directly address them. Abandoned cart rates drop. Conversion goes up. And you didn’t guess – you listened.
2. Post-Purchase Survey (at checkout completion)
This one lives on the thank-you page, immediately after purchase – when the customer is in the best possible frame of mind.
Ask:
You’ll collect the exact language your best buyers use to describe what they value – the kind of language that should be on your homepage, in your ads, and throughout your email sequences.
3. Post-Delivery Survey (7-14 days after fulfillment)
This is where you find out whether the product is actually solving the problem your marketing promised it would.
Ask:
The answers close the loop between what customers expected before buying and what they experienced after. That gap, when it exists, shows up in your return rate and LTV long before anyone flags it as a strategic problem.
Together, these three surveys give you a real-time feedback loop on who your Core Avatar is, what drives them to buy, and whether your offer is delivering on the promise that converted them.
Here’s what inconsistent Meta performance actually looks like:
The natural assumption is creative fatigue, budget fluctuations, or algorithm changes. But Shelefontiuk’s read – after working with DTC brands across a wide range of spend levels – is that the root cause is almost always avatar ambiguity.
Why Meta struggles without a clear avatar
Meta’s ad platform is a machine learning system. It learns who to show your ads to based on who converts. When your offer speaks clearly to one type of person, Meta builds an increasingly accurate model of that person and finds more of them efficiently. Your acquisition cost stabilizes. Your results compound.
When your offer converts too many different types of people – 25-year-old men, 62-year-old women, 37-year-old moms, 45-year-old dads – Meta can’t build a coherent model. It’s looking for “anybody who buys,” which is not a targetable audience. The result:
As Shelefontiuk puts it: “Meta doesn’t know who to target consistently because you don’t know who you’re talking to consistently.”
Meta has shifted toward AI-driven targeting. Advantage+ campaigns – now the default for most ecommerce advertisers – are built on the premise that the algorithm will find your buyers if you give it good conversion signals. But the algorithm is only as good as the data it’s fed.
A vague offer that converts a wide range of people gives the algorithm a wide range of signals – and wide signals mean imprecise targeting, not better reach.
The research backs this up: Lebesgue’s analysis of Meta ad performance found that broad targeting delivered 49% higher ROAS compared to lookalike targeting. But that lift doesn’t happen automatically. It depends on the creative doing the filtering – speaking so clearly to one specific type of person that the algorithm learns exactly who to find. Without a defined Core Avatar, broad targeting just means broad results.
The fix isn’t a new creative strategy. It’s offer and messaging clarity upstream. Every time Shelefontiuk’s team has helped a brand tighten its Core Avatar and align the offer to that person, Meta performance stabilized. The algorithm found its footing because it finally had consistent data to learn from.
The downstream effect of a well-defined core avatar isn’t just better ads. It’s a fundamentally more efficient funnel.
Shogun’s Q1 2026 Ecommerce Conversion Rate Benchmark Report – drawn from 747 active stores across 11 industries – puts the median ecommerce conversion rate at 1.56% and the cohort mean at 2.41%. But the more useful number is the spread:
That spread is not primarily a design problem or a checkout friction problem. The stores at the top of their industry distribution have, in most cases, made a deliberate decision about who they serve and built everything around that person:
The stores at the bottom are often still trying to appeal to everyone.
Shogun’s data shows that mid-market merchants ($15M-$49M GMV) were the most resilient segment of the year, holding essentially flat on conversion while smaller and larger cohorts declined by double digits.
The structural reason is straightforward. Mid-market brands tend to have:
When 2026 got harder – rising ad costs, traffic mix shifting toward higher-bounce paid channels – those advantages compounded. The brands that knew their buyer had something to fall back on. The ones that didn’t, felt every fluctuation.
The path to the top quartile rarely runs through a redesign or a new CRO tool. It runs through getting clearer on who you’re actually selling to – and building the offer and the funnel around that person.
Avatar research doesn’t end with surveys and Reddit threads. It becomes the input for a systematic testing process.
Shelefontiuk’s framework: research → hypothesis → test
Compile every insight from your research into a list of hypotheses. For example:
Then test them – individually, not all at once. Each hypothesis gets its own test. You’re looking for what Shelefontiuk calls customers “voting with their dollars.” When one version of the offer consistently outperforms the others, you’ve learned something real: that specific element resonates with your Core Avatar in a way the control did not.
They do the research, feel like they have a better understanding of their customer, and then implement a dozen changes at once – a new headline, a repositioned offer, different imagery, updated copy – and call it a “refresh.”
When performance improves, they don’t know why. When it doesn’t, they don’t know why either. Either way, they’ve learned nothing they can build on.
According to research by Convert, only 31% of companies have a structured approach to testing and optimization. The majority are making funnel decisions without the data to know what’s actually working.
They do it incrementally. Clear hypotheses, tied directly to what customers told them, tested one at a time. Avatar research gives you the raw material. Systematic testing tells you which insights are actually true at scale.
