The PREDICT framework.

I've spent the better part of a decade inside CRM organizations across hospitality and restaurant brands. Eleven brands at Darden. Vacation ownership at Travel + Leisure. Hotel tech at Cendyn before that. Different verticals, different scales, different tech stacks. The same problem, everywhere.

The problem is signal.

Not data, every team I've ever joined had too much data, drowning in it from the moment of onboarding. The problem isn't data. It's the inability to tell, with confidence, which data points actually predict behavior and which ones are just noise dressed up in dashboards.

A team can sit on a billion-row customer database, run nightly model refreshes, ship segmentation strategies to four channels, and still not be able to answer the simplest question an executive can ask: "Why did this campaign work?" Or, more painfully: "Why didn't it?"

This isn't a tooling problem. It's a framework problem. And it's the reason most CRM organizations underperform what their data should let them do.

I built a framework to fix this. I call it PREDICT. It's not academic, I've used it inside real organizations to ship real attribution, build real segments, and drive real revenue. It's seven steps, each of which most teams skip or do badly.

Here it is, the first time it's been written down publicly.

---

P — Pattern (not prediction)

Most teams start at the wrong end. They start with the prediction (will this person convert?) and back into the data they think predicts it. That's how you end up with 47 propensity models that all use the same five obvious features and somehow disagree with each other.

Start with the pattern instead. Look at the historical behavior of customers who actually converted. Not the model's guess about who will convert. The receipts.

You're looking for the shape of the behavior, not the score. Did they engage with three emails before booking? Did they hit the site twice in a 30-day window? Did they call the contact center? What did the sequence look like?

This sounds basic, but most CRM teams I've seen have never done a pure pattern audit. They jump straight to modeling. And then they wonder why their models don't generalize.

The pattern is the source of truth. Everything else is hypothesis.

R — Receipts (not assumptions)

The receipts are what actually happened. Not what the customer said in a survey. Not what the persona document claims. Not what the model predicts. What they did.

This is where most teams reveal a deep cultural problem. They have access to receipts (purchase history, browse logs, support tickets, email engagement) but they treat receipts as one input among many, weighted equally with the survey data and the focus group output and the segmentation deck from the agency.

That's wrong. Receipts are not one input. Receipts are the truth, and everything else is interpretation of the truth.

When the receipts disagree with the persona, the persona is wrong. When the receipts disagree with the brief, the brief is wrong. When the receipts disagree with the executive's intuition, the executive is wrong. The receipts are not negotiable.

This is the hardest sell inside most organizations because it requires deference to data over politics. The framework lives or dies on whether leadership can hold this line.

E — Edges (not averages)

The average customer doesn't exist. The average customer is a statistical fiction useful for board decks and useless for strategy.

What exists is the edges, the small groups whose behavior deviates from the mean in interesting ways. The customers who bought twice in 30 days. The ones who stopped engaging after a specific touchpoint. The ones who churned after their second support ticket.

Edges are where the actionable signal lives. The average tells you the system is healthy. The edges tell you what to do tomorrow.

Most CRM dashboards optimize for the average (open rates, conversion rates, ARPU, NPS). These are scoreboards. They are not strategy inputs. To find strategy inputs, you have to systematically look at the edges, and most teams don't because finding edges requires segmentation discipline most orgs don't have.

The edges are where the work is. The average is where the comfort is.

D — Drift (not snapshots)

Customer behavior changes. Always. Constantly. The model you trained six months ago is already wrong about the customer who's about to convert today, because that customer has read three new reviews, seen two new ads, and has $40 less in checking than they did when the model was built.

But most CRM teams treat models like furniture. They build them, they install them, and they assume they'll keep working. Then six months later, the model is mysteriously underperforming, and nobody knows why.

The answer is almost always drift. The world changed. The customer changed. The model didn't.

PREDICT requires drift monitoring as a first-class operational discipline, not a quarterly review item. Every model needs a drift dashboard. Every segment needs a recency check. Every prediction needs a "confidence today vs. confidence at build" comparison.

If you don't watch drift, you're not running a CRM organization. You're running a museum of past insights.

I — Intervention (not observation)

A CRM team that only observes is a research department in disguise. The whole point of identifying patterns, validating with receipts, finding edges, and tracking drift is to do something about it.

The intervention is what closes the loop. Send the email. Trigger the offer. Route the call. Hold the seat. Whatever the behavior is, the team's job is to bend it, to make the trajectory of the customer better than it would have been without intervention.

This is where most CRM organizations get politically stuck. The intervention requires partnership with marketing, ops, sales, and product. None of those functions report to the CRM lead. So interventions get watered down, delayed, or canceled. And the team retreats into observation, because observation is safe and observation doesn't require asking other people to do anything.

The framework dies at the intervention step in most organizations. This is also the step that separates real CRM leaders from CRM analysts with leadership titles.

C — Counter (not victory lap)

When the intervention works, the temptation is overwhelming to declare victory and move on. Don't.

After every intervention, run the counter. What's the second-order effect? Did the email that boosted conversion this week cannibalize next week's purchases? Did the offer that closed the deal train the customer to wait for offers next time? Did the seat-save intervention work for this guest but cost goodwill with the other three who showed up on time?

Counter-effects are how short-term wins become long-term losses. They are also almost never measured because measuring them requires waiting, and waiting is hard when there's a quarterly review looming.

The counter discipline is what separates a CRM team that looks effective from one that is effective. Anyone can find a number that went up. The harder question is: what went down that nobody bothered to look at?

T — Teach (not hoard)

The last step is the one most analytics leaders skip entirely, and it's the most important for the long-term health of the function.

Teach the framework to the people who need it. Marketing partners. Ops teams. Account managers. Executives who are about to ask you the same question for the 40th time. Anyone whose decisions get better when they understand how you think about signal.

If the framework only lives in the CRM team's heads, it dies when the team rotates. If it lives in the company's collective vocabulary, it compounds. Every conversation gets sharper. Every brief gets better. Every executive question gets more interesting.

Teaching is not about giving away the magic. It's about making the magic bigger. The CRM team that teaches well becomes indispensable, not because they hoard knowledge, but because they elevate the conversation around them.

---

So that's PREDICT.

Pattern → Receipts → Edges → Drift → Intervention → Counter → Teach.

It's not the only way to run a data-driven CRM organization. It's the way that has worked for me, across very different brands, at very different scales. I'm putting it down here because I've been carrying it in my head for years, and at some point a framework has to leave your head and meet the world.

If you're an operator working on this kind of problem, I'd love to know what resonates and what doesn't. The framework is a draft, like everything is.