Understanding churn signals in loyalty data before the member leaves

Most brands find out a member is leaving at the exact moment it's too late to do anything about it. The cancellation or lapse is the endpoint of a process, not the starting point. It begins weeks or months earlier, as behaviour quietly shifts and no alert fires.

At Livewall, we design and build loyalty programmes and platforms for brands including HEMA, Decathlon, Rituals, and Just Eat Takeaway. Across those programmes, one pattern holds: the data needed to predict churn is almost always already there. The question is whether anyone is reading it at the right time.

This article covers the signals that matter, how to organise them into a system that works, and which interventions actually move the needle.

The three earliest signals

1. Declining session frequency

A member who normally opens the app three times a week dropping to once a week is showing something significant. Not once, but as a trend across two to three weeks. One quiet week means nothing. A sustained drop in sessions means everything.

The key is to measure against the individual baseline, not the programme average. A member who has always been a light user is performing normally. A member who was previously highly active and is now pulling back is on their way out.

2. Reduced transactional behaviour

Within loyalty programmes, purchases or other point-earning actions are the clearest indicator of engagement. When transaction frequency drops while the member still has some app activity, you're in a middle zone: they're still present but no longer buying.

That's a different problem from full inactivity, and it calls for a different intervention. Discount activation typically outperforms content or gamification in this zone.

3. Ignoring push notifications or emails

Open rates and click rates are lagging indicators when viewed per campaign. But at the individual level they're an early warning. A member who hasn't opened the last four communications is showing that relevance has slipped.

Important note: communication fatigue is as dangerous as churn itself. Sending more messages to someone who has already stopped responding will accelerate their departure, not prevent it.

From individual signals to a workable system

Isolated signals are useless without combination. The power comes from aggregating behavioural indicators into a single readable score per member. We call this an engagement health score, and in its most practical form it looks something like this:

• Recency — when was the last session or transaction?
• Frequency — how does activity over the last 30 days compare to that member's personal historical average?
• Depth — does the member go beyond the surface? Are they visiting product pages, sharing content, responding to challenges?
• Communication responsiveness — how actively does the member engage with brand communications?

Each dimension carries a weight calibrated to the specific programme. A daily grocery app has different weights than a seasonal campaign.

The score works best when it is calculated continuously and when threshold values automatically trigger an intervention flow. No waiting for a monthly report. Acting at the moment someone crosses a boundary.

What actually works as an intervention

Not every intervention works for every churn stage. This is the distinction that makes the biggest practical difference:

Early signal (activity dip, member still engaged) Use challenges and gamification to re-engage. A specific challenge tied to past behaviour outperforms a generic bonus. Show the member that the programme knows who they are.

Mid-stage (communication response dropping, transactions declining) Personalise the reward structure. Offer something relevant to that member's specific purchase history or behavioural pattern. Time-limited offers work well here, provided they are not generic.

Late signal (member nearly inactive) A win-back mechanic with a clear value proposition. Not five messages per week, but one well-timed moment with a relevant reason to return. A temporary tier upgrade sometimes works. A simple 'we miss you' message can be enough if it is genuinely personalised.

In our retention strategy work, we always start by mapping behavioural segments before building an intervention plan. A one-size-fits-all retention campaign rarely delivers.

The data you actually need

A common objection: 'Our data isn't in good enough shape.' That's rarely entirely true. In most cases the baseline data exists but is spread across systems that don't talk to each other.

The minimum dataset you need to detect churn signals reliably:

• Session data per member, with timestamp
• Transaction history with date and category
• Communication data: opens, clicks, and unsubscribes per member
• Programme behaviour: challenge participation, point-earning actions

You don't need an advanced machine learning model to start. A well-designed loyalty platform with the right data structure and simple rule logic already produces significant results. You can add complexity later once you understand the baseline patterns.

What you do need: someone with ownership over churn monitoring. Data without action is worthless. It's the combination of a good system and a team that acts on it that changes outcomes.

What changes in practice

Brands that pick up churn signals proactively see three things shift. First, average member lifetime increases because interventions happen while they can still have an effect. Second, CRM costs per retained member improve because you stop sending everyone the same expensive win-back offer. Third, the programme's data quality grows because you start understanding which behaviours are genuinely predictive of departure.

This is not about perfect prediction. It's about systematically acting earlier than you do now. Even a shift from reactive to proactive with 60% reliability delivers measurable results for virtually any loyalty programme design.