The Cold Start Problem: Why Most Personalisation Fails Before It Starts

You're three weeks into planning a personalisation strategy. The new platform launched cleanly. The customer experience is genuinely better. Someone's done a solid job of the migration.

Then someone asks the obvious question.

"Where's the data coming from?"

That's when it hits. In moving everything across from the old system, nobody carried the behavioural data with them. No browse history. No search patterns. No purchase data tied to customer accounts. Just a clean, empty slate - and a personalisation strategy with nothing to run on.

It won't work. Not yet. You'll have to wait.

That's the cold start problem. In my experience, it's the single most common reason personalisation projects underperform - not the platform choice, not the budget, not the team. Nobody started capturing data early enough.

What the cold start problem actually is

When you launch a new site, switch platforms, or bolt on a personalisation tool, you start from zero. You don't know how your customers browse. You don't know which categories they visit before they buy, which search terms convert, or how long their purchase cycles run. There's no behavioural fingerprint to work with.

Until that data accumulates, any personalisation you attempt is guesswork. And guesswork at scale can actively harm the experience - surfacing the wrong products, sending ill-timed emails, treating new visitors the same as lapsed ones.

Think of it like smoking a brisket. You can buy the best smoker on the market - a proper offset, a Big Green Egg, whatever you fancy - but if you skip the preparation, you've wasted the meat. The rub needs time to penetrate. The temperature needs managing across hours, not minutes. The equipment is the easy part. The patience is the hard part. Personalisation works the same way. The platform is the smoker. The data is everything that happens before you light it.

How it happens

There are three patterns I see repeatedly, and they're all avoidable.

The replatform wipe. A site migration happens and the historical data stays behind in the old system. Nobody exports it. Nobody maps it across. The new platform launches with a clean database and years of customer behaviour effectively thrown away. This is the most painful version, because the data existed - it just didn't make the journey.

The "we'll add tracking later" trap. The team gets the site live first, planning to sort behavioural tracking once things settle down. "Later" arrives six months after launch. Those six months are gone and can't be recovered. The data collection window closes quietly whilst everyone's focused on something else.

The tool-first mistake. A business buys a personalisation platform - sometimes a genuinely impressive one - before properly understanding what data they have, in what format, and whether it can actually feed the tool. They're sold on features built for mature datasets, then frustrated when those features don't deliver on day one. It's not that the platform lied. It's that the platform assumes a foundation that wasn't there.

The honest conversation

When clients come to TSD ready to personalise, the first question we ask is: what data do we have to work with?

Sometimes the answer is reassuring. An established site with proper tracking, purchase history tied to accounts, years of search data - we can get moving quickly. More often there's a gap. The data exists but it's messy, or it's scattered across systems that don't talk to each other, or there simply isn't enough of it yet.

The honest answer is that you need at least 45 days of clean, consistent data before meaningful personalisation becomes possible - ideally 90, depending on your site's traffic volume. Not possible as in technically achievable - possible as in actually useful. Patterns only emerge over time. Seasonal behaviour, purchase cycles, the difference between a browsing session and a buying session - none of that is visible in two weeks of data.

Those figures are a starting point, not a ceiling. Lower-traffic sites will need longer. Some types of personalisation need longer regardless.

Tracking has to be part of the spec, not an afterthought

The fix is straightforward, even if it requires discipline. When we're configuring an e-commerce store for a client now, tracking is in the spec from day one. Not a post-launch task. Not on the "nice to have" list. A build requirement.

In practice, that means three things.

Behavioural event tracking from launch - what people search for, which categories they browse, what they add to the cart and abandon, how long they spend on product pages. This is the raw material of personalisation.

Purchase history linked to customer accounts, not just anonymous sessions. A purchase you can't tie to an individual is much less useful than one you can.

Customer segments tagged as they emerge - new versus returning, B2B versus B2C where relevant. Early segmentation creates the foundation for everything that comes later.

And here's the bit people find counterintuitive: don't wait until you know exactly what you'll do with the data before you start capturing it. Capture it anyway. The use cases will follow. The data won't retrospectively appear if you delay.

What you can do during the cold start period

You're not stuck while you wait for data to build. Rule-based personalisation - new visitor versus returning, geography, referring source - doesn't need historical data. It works from signals available in the current session. Not sophisticated, but it's a start.

If you're replatforming, export whatever you can from the old system before the migration happens. Order history, customer records, even basic category affinity data. Imperfect historical data is better than none. It can be cleaned and mapped.

For clients on Shopify, TSD's sales reporting platform TrendSeam gives you a head start on purchase pattern analysis even during early collection phases. It's not magic, but it means we're not waiting quite as long to see useful patterns emerge.

For clients on Umbraco, it's worth knowing that Umbraco Engage includes a track-only analytics mode. It's considerably cheaper than the full personalisation licence, which makes it a practical option for businesses that aren't ready to personalise yet but want to start building a data foundation now. Getting that in place at the point of build - rather than retrofitting it later - means you've got solid behavioural data waiting for you when you are ready to move forward.

If a cold start is unavoidable (i.e. it's a brand new site), use the cold start period productively. Set up tracking correctly. Import what you can. Start with rule-based approaches. That way, when the data matures, you're building on something solid rather than scrambling to catch up.

The difference preparation makes

Clients who get tracking right from the start of a new build are typically running meaningful personalisation within 45 to 90 days, depending on traffic volume. Those who bolt it on later are often waiting 6 to 12 months to reach the same point.

That gap has a cost. Not just in delayed personalisation revenue, but in months of customer behaviour that could have informed better decisions across the board - merchandising, stock planning, email timing, content strategy. Behavioural data doesn't just power personalisation. It powers most of the intelligent decisions a modern retailer makes.

The cold start problem is fixable. But you have to fix it before you launch, not after. Once that window closes, you can't reopen it.

If you're planning a new build or replatform and want to make sure tracking is sorted from day one, get in touch. It's a much easier conversation before the project starts than after.