Know Your Customer
Personas, Segmentation, Customer Research
- Replace assumptions with real customer data fast - what people do beats what they say they'll do.
- Ask every buyer what made them buy, what almost stopped them, and where they first heard of you.
- Segment customers by what they do and the job they hired you for, not demographics.
- The second purchase is the hinge - repeat buyers are worth 3-5x, so track LTV by cohort, never blended.
On this page
- Why Customer Research Follows Foundation
- Starting With Assumptions (Pre-Revenue / Early Stage)
- Talking to Real Customers
- Survey-Based Attribution (HDYHAU)
- Jobs-to-Be-Done Interviews
- The Cookie-to-Zero-Party Shift
- Building Useful Personas (Not the Fluffy Ones)
- What a Useful Persona Includes
- From Data to a Useful Persona
- Profiling Your Cohorts at Scale
- Activity-Based Cohort Segmentation
- Outcome & Needs-Based Segmentation
- RFM & Lifecycle Segmentation
- Measuring Acquisition Quality by Cohort LTV
- Channel & Geography-Based Segmentation
- How Personas Evolve
- How This Feeds Everything Else
You can't build a great product, write compelling ads, or pick the right channels if you don't know who you're selling to. Start with assumptions, but replace them with real data as fast as you can. Talk to customers. Survey them. Watch what they actually do, not what they say they'll do. Your customer personas should be living documents that get sharper over time.
Why Customer Research Follows Foundation
Foundation comes first - PMF, beachhead market, brand DNA. But before you build product, run ads, or create content, understand who you're building for.
This is iterative. Start with assumptions and refine as you learn. Your day-one customer hypothesis will be wrong in ways you can't predict. The goal is to start with something testable and replace assumptions with real data fast.
Every section that follows depends on this one - product decisions, ad targeting, email segmentation, content strategy, IRL investments.
Starting With Assumptions (Pre-Revenue / Early Stage)
Before you have customers, you have hypotheses. Write them down. Be specific as vague assumptions are impossible to validate.
- Who is your beachhead customer? Not "everyone who likes outdoor gear." Something like: "Male road cyclists, 28-45, who commute by bike and want their phone accessible for navigation."
- What problem are you solving? Their actual problem, not your product features.
- Where do they hang out? Platforms, communities, forums, events, stores, podcasts.
- What would make them buy? Price point, social proof, features, brand alignment?
- What might stop them? Price objections, trust concerns, competing solutions, inertia?
Every assumption above should have a plan to validate it within 90 days. If you can't validate it, it's not a hypothesis - it's a guess.
Talking to Real Customers
The surveys told us what mattered at scale. But some of the sharpest research we ever did was in person, at events. We'd show up at a cycling ride or a four-wheel-drive show with our products and just talk to people. Hundreds of conversations. You learn far more in person than you ever will from analytics. What the actual objections are. What people love that you didn't expect. What confuses them. What they wish existed. Show up. Put the product in people's hands. Listen. Then take everything you learn back into the next product, the next ad, the next piece of content, the next product page design. There's no shortcut for this.
Customer Interviews: Aim for 20+ in the first 90 days. Structured conversations with open-ended questions. "What made you buy?" not "Did you like our product?"
Post-Purchase Surveys - three questions that will change your brand:
Those three are the always-on version, one short survey on every order. The other gear is the periodic deep survey: a longer one across your whole base, run every so often, that goes past the transaction into who the customer actually is, what else they own, what they do with the product, and what nearly stopped them. The quick survey keeps a finger on the pulse. The deep one is where the real persona work happens, and it routinely overturns things you were certain of. Both are surveys, and between them they're the highest-leverage customer research most brands underuse.
Support Ticket Analysis: Your support inbox is unfiltered customer insight. Themes tell you what your website isn't communicating and what your product isn't delivering.
Social Listening: What are people saying when they don't know you're listening? Reddit, Facebook groups, TikTok comments, forums. Unprompted opinions are worth 10x survey responses.
Another powerful approach is using AI to do deep research on existing products or categories - scrape all the customer reviews across the internet and surface the top five negative and top five positive themes. This gives you a clear picture of what matters to customers, what's underserved, what your product must deliver on, and the pain points that are still unaddressed. It's the fastest way to understand where the opportunity sits.
Survey-Based Attribution (HDYHAU)
The third question in that post-purchase survey - "where did you first hear about us?" - is doing more work than the other two combined. With third-party signal degraded, the platforms over-claim and under-count in ways that contradict each other, and the customer's own answer has quietly become the most reliable attribution method most brands have. It's called HDYHAU, and the thank-you page is the right place to ask it because that's where people actually respond.
The move isn't to trust HDYHAU blindly over the platforms - it's to triangulate. Put the customer's stated source next to what Meta, Google and your MMM claim, and the gaps are the insight:
Platform-only attribution- Meta claims 60% of conversions
- Google claims another 45%
- The two add up to more than 100%
- Podcasts, word-of-mouth and "a friend told me" show as zero
- You over-fund the channels that report best, not the ones that work
HDYHAU triangulated- Customers say 35% heard via a friend
- Podcast shows up at 12% - invisible to every platform
- Meta's self-reported share lands well under its claim
- Discrepancies flag where the platforms are double-counting
- You fund what customers actually credit, dark social included
Where the survey and the platform disagree, the survey is usually closer for top-of-funnel and word-of-mouth, and the platform is closer for last-click. Use each for what it's good at. And the answers aren't only an attribution number - when a customer tells you they bought to film their rides, that use-case answer feeds straight into your outcome segments. One survey, two payloads: where they came from, and why they bought. The attribution methodology itself - triangulating the platforms, MMM, holdouts - is Section 27: Measurement & Data's job; what earns its place here is that second payload, the use-case signal feeding your segments.
NPS Surveys: The score matters less than the follow-up: "Why did you give that score?"
Jobs-to-Be-Done Interviews
The 20+ interviews above are only as good as the questions. The trap is asking about your product. "What do you think of the mount?" gets you politeness. The job is what they hired the product to do, and it has three layers: the functional job (what it does), the emotional job (how they want to feel), and the social job (how they want to be seen). For a bike mount the functional job is "holds my phone". The real job is "feeling secure that my phone won't fly off at 40mph on a $1,500 bike". You build product and write ads for the second one.
The output you want is a solution-free outcome statement: a sentence that names the job without naming your product, so it survives even when the product changes. A weak one bakes in the solution: "I want a magnetic phone mount." A strong one stays solution-free: "Minimise the time it takes to get my phone mounted and navigation running when I stop at a light." The first locks you into a feature. The second is a job you can win three different ways - and it's the sentence your ad headline gets written from.
The interview is the high-fidelity version of this, but it doesn't scale past the few dozen conversations you can sit through, and a few dozen is where it can quietly mislead you. One articulate customer with a strong opinion can tilt a small sample, and you walk away certain of something that was really just one loud voice. The survey is how you run the same discovery across thousands of customers, and that scale is the point. When a job comes back from twenty people it's a hypothesis. When it comes back from thousands, in the same words, over and over, you can be fairly sure it's real and not a handful of one-offs. This is the same post-purchase survey from earlier in this section doing a third job: alongside the trigger and the objection, one open-ended question asks for the job itself - "what were you trying to do when you bought this?" or "what was going on that made you start looking?" You lose the flinch and the body language, but you gain statistical weight. Read the verbatims, not just the tallies. That is the whole payoff: you stop guessing which features and benefits to lead with and which objections to defuse, and you start knowing, cohort by cohort. Every ad, product page and sales video then gets written from what the customer told you, in their words, not from what the team assumed in a meeting.
Run the survey to find the jobs at volume, then use in-person conversations to go deep on the ones that matter. The events research earlier in this section was JTBD discovery before we had a name for it - putting the product in someone's hands and watching where they hesitate surfaces the emotional and social jobs a survey question only hints at, the flinch, the raised eyebrow, the "oh that's clever" you didn't expect. One caution across all of it: weight what customers do over what they say they will. People will tell you they'd pay $100 and then baulk at $49, or swear they want a feature they never touch. Stated preference is the starting point; purchase data, usage data and return reasons are the check on it. Survey for breadth, conversation for depth, behaviour as the tiebreaker.
The Cookie-to-Zero-Party Shift
For a decade you could infer who your customer was from the trail third-party cookies left across the web. That trail has gone cold - third-party cookies are already blocked or unreliable in most browsers, the signal keeps degrading, and the platforms know less about your buyer than they used to. The replacement isn't better tracking. It's just asking. Zero-party data is what a customer tells you on purpose: their use case, their vehicle, their experience level, what they're trying to achieve. They hand it over because you give something back - a better recommendation, a more relevant email, the right size first time.
The post-purchase survey is one collection point, but it's the last one. Pull the questions forward and you learn the customer before they've even bought:
Because the customer hands it to you directly, you carry the obligation to collect it with explicit consent and a documented lawful basis (GDPR, CCPA, and the rest). Log what they agreed to and when. A preference centre that records consent is the difference between an asset and a liability the day a regulator or an acquirer's lawyers ask how you got it.
Building Useful Personas (Not the Fluffy Ones)
Most persona exercises produce beautifully designed PDFs that never get opened again. Good personas are built from data, not imagination, and include information you can act on.
Bad Persona- "Sarah, 34, marketing manager"
- "Loves yoga and sustainable living"
- "Shops online and values quality"
- "Active on Instagram"
- Based on: imagination and stereotypes
Good Persona- "Road cyclists, 28-45, commute + weekend rides"
- "Buying trigger: saw it on a cycling forum/Strava"
- "Main objection: 'will it fit my bike?'"
- "Channels: Strava, cycling subreddits, YouTube reviews"
- "AOV: $85, repeat rate: 42%, ~2.1 orders per customer over 24 months = $180 revenue Lifetime Value (LTV)"
- Based on: purchase data + post-purchase surveys
AI tools can now analyse survey responses at scale, extract themes from support tickets automatically, and build data-driven segments from purchase behaviour. The customer research cycle that used to take months can happen in weeks. Use AI to compress the learning loop, but don't skip the real conversations.
You can now ask an LLM to roleplay your customer and react to a price, a headline, or a positioning angle - and ~89% of researchers are already using or trialling AI tools, with ~83% planning to spend more on them. It's genuinely useful for one thing: a fast, cheap directional read before you put anything in front of real people. Use it to narrow ten message options to three. Never use it to decide. A synthetic persona has never had your product fail at 40mph, never baulked at your price, never told you the objection you didn't see coming. It reflects the internet's average of a customer, not yours. The June 2025 ICC/ESOMAR code now formally governs synthetic data in research precisely because the line between "pre-test" and "finding" gets blurred so easily. Treat LLM output as a hypothesis to validate against the real conversations this whole section is built on - the moment it replaces them, you're back to guessing with extra steps.
What a Useful Persona Includes
| Persona Element | What to Include |
|---|---|
| Demographics | Age, gender, location, income (from purchase data, not assumptions) |
| Psychographics | Values, lifestyle, identity markers, community |
| Buying triggers | What moment made them search/buy? (from surveys) |
| Objections | What almost stopped them? (from surveys + support tickets) |
| Channels | Where they discover, research, and buy |
| Price sensitivity | Full price buyer, deal hunter, or "I'll wait for a sale"? |
| Purchase frequency | One-time, seasonal, or loyal monthly? |
From Data to a Useful Persona
The table is the skeleton. Building the persona is four moves: pull the raw material from the research above (interviews, surveys, support tickets, purchase data), cluster it into the three-to-five groups that genuinely behave differently, fill each group out against the elements above, then pressure-test it. That last step is the one that matters - does the persona change a real decision? If two personas would make the same call on every channel, every creative, and every product question, they're one persona wearing two hats.
Here is the shape, worked through for a generic placeholder - "The Considered Upgrader". Swap in your own category and your own data:
- Who they are: an existing category user, 30-45, replacing a starter product they've outgrown
- Buying trigger: they hit the limits of the cheap version, then saw someone they trust using the premium one
- Main objection: "is it really worth several times the price of what I already own?"
- Where they research: YouTube reviews, Reddit threads, your own comparison content
- Economics: AOV, repeat rate and orders-per-customer pulled from your own data, never guessed
- The decision it changes: lead the ads with a side-by-side against the cheap version; sell the upgrade, don't open with a discount
That last line is the whole point of the exercise.
A note on depth: at Quad Lock our personas were extremely detailed, but that was a consequence of our category and the years of behavioural and purchase data we'd accumulated, not a goal in itself. The right level of detail is whatever makes the persona useful for the decisions actually in front of you, which depends on your category and how much real data you have. Early on, a scrappy five-line persona that genuinely changes how you brief an ad beats a twenty-page document nobody opens. Build for usefulness first, then add depth as the data and the decisions demand it.
Profiling Your Cohorts at Scale
The deep survey from earlier is where customer understanding stops being a sketch and becomes a map. Past a certain size you don't have "a customer," you have a portfolio of cohorts who buy for genuinely different reasons - and the deep survey is how you profile each one. Run it across your whole base, repeat it on a cadence so you can watch the base shift over time, and profile every cohort on the same grid so they're comparable:
- Who they are: demographics, location, devices, and what else they already own
- What they do: how, when and how often they actually use the product
- Why they bought: the job they hired you for, and the decision factors behind it
- What nearly stopped them: the objection you had to get past
- What else they need: the adjacent products that map their next purchase
At Quad Lock the survey was how we actually knew our customers rather than guessed at them, and we ran it across the whole base, year after year. We read it cohort by cohort, because a customer was never one person - the adventure rider, the cruiser rider and the daily commuter bought for genuinely different reasons, and the survey was the only thing that told us how. The headline was always the job, not the product: people weren't buying a mount, they were buying navigation, getting where they were going with the phone in view and their hands where they should be. So that is what we led with. But underneath the job sat the thing that had to be won first - trust. People had to believe the thing would not drop their phone before they would rely on it at all, that they could set off and forget it was even there. Access mattered, but trust was the gate, so we didn't say "holds your phone," we said "stop worrying about your phone and get on with the ride."
The pattern generalises, and it's why the deep survey earns its place: it does three things a dashboard can't. It names the dominant job per cohort, so you lead with the job and not the feature. It exposes the belief underneath the job - the thing the customer has to trust before the job even counts - which is what your messaging has to win first. And it shows what else each cohort needs, which is your cross-sell and where most of your LTV actually comes from (see Section 21: Customer Retention & Loyalty). Get that read wrong and you can push the wrong next product for years, the trap the cyclist example further down describes.
The deliverable is simple: a one-page profile per cohort, a cheat sheet, that your ad writers, product pages and video scripts all pull from, so every piece of content is written for a specific customer instead of blasted generically at everyone. It is also what tells you where to put the money, because the budget should follow your best cohorts (the category-by-geography method in Section 14: Meta Ads).
Activity-Based Cohort Segmentation
Group customers by what they DO, not just who they ARE. Far more actionable than demographics alone.
At Quad Lock, the customer base included cyclists, runners, motorcyclists, drivers, and more. Each cohort had different needs, consumed different content, attended different events, followed different influencers, and responded to different ad creative. Restructuring around customer activities rather than product categories transformed the business (Section 3: Brand DNA). Each cohort got its own creative, its own ambassadors, and its own event calendar. The product was the same. The way we spoke to people was different.
Activity-based personas enabled Quad Lock to create resonant content per group, target ads with cohort-specific creative, choose relevant ambassadors and partnerships, and prioritise investment based on cohort economics. What's your brand's "activity-based" personas?
Start with 3-5 personas max. If you can't remember them without looking them up, you have too many.
Outcome & Needs-Based Segmentation
Activity tells you what someone does. It doesn't always tell you why they bought. Two cyclists can buy the same mount for completely different reasons - one wants to navigate safely on a commute, the other wants their phone rock-solid while filming a descent. Same product. Same activity. Different job. The job they hire your product to do is what your messaging should speak to.
The sharpest way to find the gaps worth chasing is Opportunity Scoring. For each outcome a customer wants, you ask two things: how important is it, and how satisfied are they with what's on the market today. Plot the two against each other and the high-value gaps fall out - important outcomes that nobody is serving well.
Importance high, satisfaction low. That quadrant is where the unmet demand sits - the job customers care about that the market is fumbling. It's also where your highest-LTV segments usually hide, because solving a job nobody else solves earns loyalty, not just a sale.
Map each job to a segment and a message, and the work gets concrete fast:
| Job to be done | Segment | The outcome they're buying | Message that lands |
|---|---|---|---|
| Stay safe and navigate hands-free | Commuter | Confidence it won't move in traffic | "Locked on, eyes on the road" |
| Capture the ride without missing a beat | Content creator | A stable mount they never think about | "Film it, don't fumble it" |
| Trust it on rough terrain | Adventurer | Gear that survives where they go | "Built for the worst you can throw at it" |
The content-creator job is usually the smaller segment by headcount but the higher LTV by customer - they buy more accessories, replace more often, and tell a louder story. Size the segments by lifetime value, not just count, before you decide where the budget goes (see Section 21: Customer Retention & Loyalty for the LTV mechanics).
The opportunity isn't where YOUR customers are unhappy - they already bought. It's where the whole category leaves people underserved. The AI review-mining approach earlier in this section is built for exactly this: pull the top recurring complaints across every competitor and you've found the low-satisfaction outcomes for free.
"Wants a magnetic mount" is a feature preference. "Wants to grab the phone in one move at a red light" is the job. Build segments on jobs - features change, jobs don't, and the job is what your ad has to promise.
RFM & Lifecycle Segmentation
The most pragmatic segmentation for ecommerce isn't a survey or a persona. It's RFM - Recency, Frequency, Monetary - and it runs on data you already own: how recently someone bought, how often, and how much they've spent. No research budget, no guessing. Your store already knows.
The trap is fixed thresholds. "Spent over $200" means nothing without context - it's generous for a $40 AOV brand and trivial for a $400 one. Score on quintiles instead. Rank every customer 1 to 5 on each of R, F and M relative to your own base, so the top 20% of spenders score a 5 regardless of what your numbers actually are. The segmentation moves with your business instead of going stale.
From there, every segment maps to one lifecycle move - the score pattern tells you who, the flows do the work:
| Segment | RFM pattern | The move |
|---|---|---|
| Champions | High R, high F, high M | Reward and ask for referrals/reviews |
| Loyal / Promising | Mid-high F, rising M | Cross-sell the next activity, nudge subscription |
| One-time buyers | R recent, F = 1 | Second-purchase nudge - the make-or-break moment |
| At-risk | R falling, was high F/M | Win-back before they're gone |
| Lost | R low, no recent activity | One last reactivation, then stop spending |
The actual flows, cadence and win-back timing for each segment are Section 21: Customer Retention & Loyalty's job, with the automations in Section 12: Email & SMS. RFM is how you decide which message each segment gets; those sections run it.
A customer who buys twice is worth several times a one-time buyer - some data puts a repeat buyer at 3-5x the lifetime value. The single highest-leverage RFM segment is your one-time buyers, because moving them to two orders is what bends your LTV curve. Build that flow first.
When you don't have the data, you start with assumptions, and ours were wrong. We assumed a cyclist's next purchase would be something for a different part of their life, like a car mount, so that's what we'd put in front of them. Then the customer surveys came in and flipped it on its head. For a lot of cyclists the next thing they bought was just another cycling mount. They were enthusiasts, so they had a road bike and a mountain bike and a commuter, and they wanted the same setup on all of them. The next best product wasn't the adjacent activity, it was more of the same one. We'd have kept pushing the wrong cross-sell for years if we'd trusted the assumption instead of the data.
Re-score on a schedule - monthly is plenty for most brands - because customers move between segments constantly. A Champion who goes quiet for 60 days is now At-risk, and the win-back flow only fires if your scoring caught the shift. Stale RFM is worse than none, because you'll cross-sell people who've already churned.
Measuring Acquisition Quality by Cohort LTV
RFM tells you how to treat the customers you have. Cohort LTV tells you whether the customers you're buying are worth what you paid. The blended average is where brands fool themselves: one big-spending segment props up the mean while a channel quietly fills the top of the funnel with people who buy once and vanish. The average looks fine. The truth is hiding in the cohorts.
The fix is to stop measuring LTV as one number and start measuring it per acquisition cohort - everyone who first bought in the same month, from the same channel - and watch it accrue over time. Read the same checkpoints for every cohort so they're comparable: spend-to-date at 30, 60, 90, 180 and 365 days. A cohort that's flat after day 30 is one-and-done. A cohort that keeps climbing is where your real customers live, and it's the segment worth winning more of. You don't have to build this by hand - the cohort-analytics tools in your stack (Lifetimely, Polar) produce per-cohort LTV curves out of the box; see the analytics layer in Section 10: E-Commerce & Tech Stack.
That's the know-your-customer payoff: the high-LTV cohorts and jobs from the outcome work above earn a higher bid and more creative; the thin ones get cut or fixed. It tells you not just who to talk to, but how much each one is worth talking to. Turning that into an actual CAC ceiling - the max-allowable-CAC maths, the LTV:CAC gates, and where it all sits in the three-gate model - is Section 26: Finance & Unit Economics's job; feed your per-cohort LTV into the gates there, and see Section 21: Customer Retention & Loyalty for the LTV mechanics.
Channel & Geography-Based Segmentation
Everything above assumes you can see your customer. On your own DTC store you can - first-party data, post-purchase surveys, the lot. The moment you sell through Amazon, a marketplace, wholesale, or into a new country, that visibility drops, and the research method has to change with it. You can't run a thank-you-page survey on an order you never owned the checkout for.
Match the method to the channel's blind spot:
| Channel | First-party data gap | Research method that works |
|---|---|---|
| DTC paid social | None - you own the checkout | Post-purchase surveys, HDYHAU, quizzes, full RFM |
| Amazon / marketplace | No email, no survey on the order, no checkout | Review mining, Q&A scraping, marketplace feedback tools, category review analysis |
| Wholesale / retail | No end-customer data at all | Retailer sell-through data, in-store intercepts, the events approach earlier in this section |
| International | Use-case and jobs differ by market | Re-run interviews and surveys per market - don't port the home persona |
For Amazon and marketplaces, lean hard on the AI review-mining approach from earlier in this section - it was built for exactly the channel where you can't survey the buyer. The reviews and Q&A are your post-purchase survey, just written by someone else's customers as well as your own.
The same product gets hired for a different job in a different market - the commuter job that dominates one country can be the weekend-adventure job in another, with different objections, price sensitivity, and channels. Re-derive the persona per market from local research before you localise the ads; translating your home-market creative is not the same as knowing the customer. Section 24: International Expansion covers the market-entry sequence this feeds.
How Personas Evolve
Track how customers evolve, not only who they are at first purchase. That's where the real product and marketing insights live.
One of the most valuable things we did was track how customers moved between categories over their lifetime. What customers bought next varied by cohort - our cyclists, as the surprise above showed, often just bought another bike mount, but that was specific to cycling, where so many of them owned several bikes and wanted the same setup on each; other cohorts didn't follow the same pattern. What held across all of them was the longer game, and it was one of our best engines of growth: acquire a customer in the passion category that brought them in, earn their trust there, then expand them cross-category later. The cyclist who eventually buys a car mount. The customer who adds a motorcycle mount once they get a bike. Post-purchase surveys and purchase data mapped those paths - which products to surface down the line and which new ones to build - so we could land a customer where their passion was and grow the relationship across categories from there.
Review and update quarterly. Set a calendar reminder. If your personas haven't changed in a year, you're probably not learning.
How This Feeds Everything Else
| Section | How Personas Feed In |
|---|---|
| Section 5: Product | What to build, what features matter, what problems to solve |
| Section 10: E-Commerce & Tech Stack | Platform choice, app stack, tech decisions |
| Section 11: Website & Conversion Optimisation | Site UX, messaging hierarchy, conversion copy, objection handling |
| Section 12: Email & SMS | List segmentation, personalisation, flow triggers, content relevance |
| Section 13: Meta Ads - Setup & Infrastructure | Audience targeting, creative angles, lookalike sources, ad copy |
| Section 15: Google Ads | Keyword strategy, search intent mapping, ad copy variations |
| Section 17: Content & Creative | Positioning, tone of voice, visual identity alignment |
| Section 18: Social Media & Content | What content resonates, which platforms, what tone |
| Section 19: IRL Brand Building | Which events, which ambassadors, which partnerships, which communities |
| Section 21: Customer Retention & Loyalty | Cohort analysis, LTV by persona, retention strategy per segment |
| Section 24: International Expansion | Which markets have your target customers, localisation priorities |
Section 4 Checklist
Tools for this section
Free Excel tools that pair with this section:
- Customer Persona Builder - Turn 'we know our customer' into a structured doc the whole team can actually use.
Go from reading to doing.
You've read the section. Sign in to score your ecommerce brand with the Health Check, track your progress across 472 checklist items, and get your tools and history in one dashboard. Free, and always will be.