Instant delivery of marketing messages to customers, such as emails, ads or Web recommendations, triggered upon customer behaviors in real-time, is at the cutting-edge of today’s marketing tech. Watching Web activity, social interactions, and vendor engagements generates a lot of data that might indicate customer preferences and readiness to buy something.

There are a couple problems, however.

One is a tendency to confuse behavioral events with facts about customer needs, intent, or preferences. Just because someone views a page of golf clubs on an e-commerce site, does that mean they want to buy golf clubs? Maybe. It could also mean they are doing research for someone else, checking out the cost of a friend’s clubs, holiday browsing, are just curious about what this retailer offers or any number of other things.

If you monitor third-party data, like home sales, does that mean the new homeowner will welcome emails or ads for new refrigerators? It might. But if you get it wrong, you risk annoying your customers with unwanted solicitations, or worse, making them suspect you are stalking them.

Have you ever received an email or ad with recommendations or offers that has annoyed you in this way? I have, frequently.

The other problem is you have to wait for customers to behave a certain way before you communicate a relevant message based on that triggered event. What do you do with good customers who have not engaged for a while, who have opted out of your behavior monitoring schemes or for whom good data is not readily available?

Ultimately, marketing’s job is to get people to buy your products and services. Real-time behavior monitoring can provide valuable insights when it comes to knowing what to offer a customer and when, but it is at best supplemental to the kind of analytics that matter most.

If you want to predict future purchasing behavior, the most reliable and accurate predictor is past purchasing behavior. Despite where they surf, what links they click, or what they say, people vote with their wallets when it comes to real buying intent. Purchasing intent, in terms of when people will buy and what they will buy, develops and may be sustained over longer periods before the actual purchase occurs, often unrelated to or only loosely related to specific behaviors detectable at a given moment in time.

Therefore, any marketing communication that attempts to deliver highly relevant or individualized messages to customers, whether or not in response to a real-time trigger, should have at it’s foundation an analysis of past purchasing behavior.

By analyzing and recognizing the patterns in purchasing behavior across an entire customer base, it is possible to identify where any given customer is in their loyalty lifecycle and what they are most likely to buy over time, such as the next 7-, 30-, 60- or 90-day periods, despite what they do today.

Purchasing data is not only the most accurate predictor of purchases, but it is usually the most reliably available and unambiguous data available for every customer. Using it to target customers and individualize messages is not controversial. It’s been used for that purpose by companies for, who knows, 100 years?

Since nothing guarantees success, the best approach is to use both the big picture of purchasing behavior and in-the-moment behaviors to target and individualize offers to customers. While you cannot always count on existing customers identifying themselves and leaving traces of their behaviors that you can collect, integrate, interpret and use in real-time, you can always use the purchase transaction history you already own, and as it turns out, it’s more effective anyway.