The 4.2-Month Warning Nobody Reads

A Leeds logistics company lost 25 clients last year. When we mapped the data retroactively, a pattern emerged. It was not subtle.

The accounts that left didn't leave suddenly. They withdrew gradually. Over 4.2 months, on average, across the same five dimensions:

Order volumes declined 15-30%. Gradual, not sudden. A 5% drop in one month is noise. A 5% drop per month for four months is a 19% decline. Month by month, unremarkable. Cumulatively, the client was disengaging.

Complaint frequency increased. From zero or one per quarter to two or three. Not dramatic escalations. Small upticks in minor issues that, taken individually, looked routine.

Email response times lengthened. The client who used to reply the same day started taking three days. Then five. The communication cooled without anyone announcing it.

QBR engagement dropped. Meetings that had been substantive discussions (questions, plans, shared priorities) became polite nods and early adjournments.

Payment terms stretched. From net-30 to net-40, then net-45. Not enough to trigger credit control. Enough to signal reduced commitment (a distinction that credit control processes are not designed to detect, and therefore don't).

Each signal, in isolation, looked like normal business variation. Together, across the same account, over the same period, they described withdrawal. Eighteen of 25 churned accounts followed this pattern. The commercial team spotted 7. The other 18 were, in the commercial director's assessment, "not on our radar."

Not on the radar. For 4.2 months. While the signals sat in four separate systems that nobody was looking at simultaneously.

The specific signals vary by industry. The pattern is consistent. We've now seen it across logistics (Tom, Leeds) and SaaS (Sophie, London), and the shape is recognisable in every B2B service business with recurring client relationships.

Signal 1: Volume or usage decline. In logistics: order volumes. In SaaS: login frequency and feature adoption. The first signal, usually visible 3-4 months before cancellation. Always gradual. A sudden drop gets noticed. A steady 5% monthly erosion doesn't, because each individual month is within normal variance. The decline becomes visible only when someone tracks the trend over 8-12 weeks, which requires someone to be tracking trends over 8-12 weeks, which requires a system, because humans track events, not gradients.

Signal 2: Complaint or support pattern shift. The small increase in minor issues. From zero per quarter to two. The shift from silence to occasional grumbles. Not the dramatic escalation that triggers a crisis response. The quiet uptick that triggers nothing because each individual complaint is resolvable and minor. Month 2-3 before churn.

Signal 3: Communication cooling. Response times lengthen. Meetings become shorter. The client who used to share plans and ask questions becomes polite but distant. This is a relationship signal, not an operational one, which means it lives in email threads and meeting notes rather than in any structured database. Nobody's job is to measure the warmth of client emails, which is why nobody does. Month 2-3.

Signal 4: Payment behaviour change. Terms stretching without formal renegotiation. The Leeds logistics firm discovered this was a stronger early predictor than complaint frequency. Clients who complain are still engaged (they want the problem fixed). Clients who silently stretch payments are reducing commitment without confrontation. The quiet withdrawal is more predictive than the loud one. Month 1-2 before churn.

Signal 5: Engagement withdrawal. In logistics: QBR participation drops, attendee seniority falls. In SaaS: feature adoption stalls. In consulting: the client stops bringing new problems. In agencies: the client stops requesting forward-looking work and only responds to deliverables. The final behavioural signal before the formal decision to leave. Month 0-1.

The consistent finding across both the Leeds and London data: each signal is explainable on its own. "They're probably just busy." "Volumes dip seasonally." "That complaint was resolved." Together, five explainable signals across the same account over the same period tell a story that nobody is reading because the story is distributed across four systems and one person's instinct.

Healthy accounts show the inverse pattern. Mapping it is valuable for the same reason: it tells you what "good" looks like, specifically, for your client base.

Volume stable or growing. Complaints low, and when they arise, the client's tone signals trust (they expect the problem to be fixed, which is the tone of someone who plans to stay). Communication warm and responsive. Payment consistently on time. Engagement active: asking questions, sharing plans, attending reviews with senior staff.

The Leeds logistics firm discovered the sharpest retention signal was one nobody had measured. Clients who sent a director to quarterly business reviews had a 92% retention rate. Clients whose QBR attendees dropped to coordinator-only level had a 3x higher churn rate over the following 12 months. This signal was sitting in Salesforce meeting notes for years. Unstructured text in a free-form field. Unanalysed. The strongest predictor of whether a £25,600 annual contract would renew was the job title of the person who showed up to a quarterly meeting. Nobody had looked.

The London SaaS company (Sophie, Week 17) found a retention signal in Mixpanel: clients who adopted three or more features in their first 90 days had dramatically lower churn. Clients who used only the core feature beyond month three churned at 4x the rate. The signal was in the product analytics tool. For years. Nobody had aggregated it by account.

Both discoveries were retrospective. The data existed. The pattern was there. The systems held the signals. Nobody had connected them because each signal lived in a different tool and nobody's job description said "connect the dots across four systems for 180 accounts."

Three structural reasons. None involve negligence or inattention.

Signals live in separate systems. Volume in the WMS. Complaints in the CRM. Payments in Sage. Engagement in meeting notes. Each person on the commercial team sees their slice. Tom checks volume. His AMs check complaints. Finance checks payments. None of them are wrong. All of them are incomplete. The composite picture exists nowhere except in Tom's head, once a fortnight, for 180 accounts, from whatever he can recall.

Gradual decline hides in noise. A 5% volume drop in a single month is within normal variance. Nobody flags it. Nobody should flag it. But a 5% drop per month for four consecutive months is a 19% decline, and by month four the client has already started conversations with competitors. The system that would catch this needs to track trends over weeks, not assess snapshots fortnightly. Most businesses assess snapshots. The gradual decline lives in the space between snapshots, which is the space nobody examines.

Nobody's job is pattern recognition across systems. Tom's job is managing commercial relationships. His AMs' jobs are account management. None of their job descriptions say "aggregate signals from four systems and identify multi-month behavioural patterns across 180 accounts." They do it informally, from instinct, at a fortnightly meeting. Fortnightly intuitive synthesis across 180 accounts is, when stated precisely, a description of a task that would be extraordinary if it worked and unremarkable that it doesn't. (This is, applied to client health specifically, the invisible 30%: unnamed monitoring work consuming a quarter of the commercial team's time.)

The firms that catch the pattern share an approach. None of it requires unusual technology.

They connect the signals first, analyse second. The Leeds logistics agent aggregated four systems into one score. The London SaaS agent aggregated five. In both cases, the individual tools were standard. The insight was in the combination. No single system showed the pattern. All of them together did.

They weight the signals based on their own data. The Leeds firm discovered that QBR attendee seniority predicted churn more strongly than complaint frequency. Nobody would have guessed that from industry benchmarks. Their own retrospective data revealed it. Generic weighting models are starting points. Six months of actual churn data produces weights specific to your client base, your service, and your market.

They measure trends, not snapshots. A health score of 72 means nothing in isolation. A score that dropped from 85 to 72 over three months means everything. The direction and velocity of change matter more than the absolute number. Monthly snapshots show where an account is. Weekly trends show where it's going.

They act early. At month 2 of a 4.2-month decline, the account is saveable. An unprompted conversation about service quality, initiated before the client has contacted competitors, is a retention conversation. The same conversation at month 4 is a negotiation the client has already decided to lose. The difference between month 2 and month 4 is the difference between retaining a £25,600 contract and funding a replacement sales cycle that costs five times more.

Agent cost across the two Blueprints in this area: £260 to £280 per month. Revenue at risk in both cases: measured in hundreds of thousands per year. The investment case requires no persuasion. It requires arithmetic.

Your clients are signalling whether they're staying or leaving. Not with announcements or formal notice. With behaviour. Volume. Complaints. Response times. Payment patterns. Meeting attendance.

The signals are in your systems right now. Distributed across whichever combination of operational, financial, support, and CRM tools you run. The question is whether they're connected. Whether anyone is looking at all of them simultaneously, for the same account, over time.

If the answer is a fortnightly meeting where three people report from memory: the 4.2-month warning is in your data. Nobody is reading it.

The AI Workflow Diagnostic helps you identify which client health signals your business already holds and which ones are worth aggregating. Takes 10-15 minutes.

Or grab Unstuck. Forty stories. Every one had signals nobody was connecting.

by SP, CEO - Connect on LinkedIn
for the AdAI Ed. Team