Customer churn analysis is the practice of measuring which customers stopped buying from you, or cut back sharply, over a period, and working out why and which signals came first. In B2B, customers rarely announce they are leaving; orders simply thin out and baskets shrink. The point of the analysis is to see that decline before the customer has gone.
Why B2B churn goes unnoticed
In a subscription business churn is obvious: the contract is not renewed and the account closes. Wholesalers, manufacturers, distributors and service firms work differently. The relationship runs on orders and has no end date, so when a dealer that used to order every quarter starts ordering twice a year, nothing flags it. The account is still open and the customer still looks "active".
This silent churn tends to hide in three places. The first is headline revenue: growth from new accounts masks the erosion of existing ones. The second is the sales rep's memory: the rep usually knows why a customer is buying less, but that knowledge never reaches a system. The third is fragmented data, with orders in the ERP, complaints in email and service visits in a separate spreadsheet, and nobody joining them up at customer level.
Churn analysis looks at the opposite end of the CRM and sales pipeline. The pipeline measures how you win customers; churn analysis measures when and why you lose the ones you have already won.
The cost of not measuring churn
Many businesses in Türkiye are not yet equipped to ask the question at all. Take-up of CRM and business intelligence, the two basic tools for tracking customer behaviour, remains low:
According to TurkStat, only 12.0% of Turkish enterprises with 10 or more employees used CRM software in 2025, and only 6.5% used business intelligence (BI) software. (TurkStat ICT Usage Survey in Enterprises 2025)
Even where the data exists, its quality can wreck the analysis. Data quality expert Thomas C. Redman estimates the cost of bad data at 15% to 25% of revenue for most companies (MIT Sloan Management Review, 2017). One company held under three separate customer accounts turns into three "lost customers" in a churn report, or into a loss nobody sees. That is why the work starts with cleaning up duplicate records.
Listening to customers also appears to go hand in hand with growth. Bain, which created the Net Promoter Score, states that Net Promoter System leaders on average grow at more than twice the rate of their competitors (Bain & Company, Net Promoter System: Measuring Your Net Promoter Score). That is the method owner's own claim and should not be read as proof of cause and effect, but it is a useful reminder that listening to customers and measuring loss belong in how growth is managed.
How to measure churn: four core metrics
A single "churn rate" figure is misleading. In B2B you need at least four metrics read together:
| Metric | What it measures | How it is calculated | What it can hide |
|---|---|---|---|
| Customer (logo) churn | How many customers left | Customers lost in the period / active customers at the start | Treats a small account the same as a large one |
| Gross revenue retention | How much existing revenue you kept | This period's revenue from the same customers, excluding increases / last period's revenue | Ignores upsell and expansion |
| Net revenue retention | Whether the existing base is growing | This period's total revenue from the same customers / last period's revenue | Growth at a few big accounts can mask many small losses |
| Partial churn (contraction) | Customers who stay but buy less | Number and share of customers whose order frequency or value falls below a set threshold | Raises false alarms if seasonality is not separated out |
The last row matters most. Outside subscriptions, you decide what "lost" means. A rule measured against each customer's own habits, such as "at risk after twice their usual order interval, lost after three times", is far more accurate than applying the same fixed period to everyone.
When choosing which of these reach the management dashboard, apply the principle from our guide on how to set KPIs: every metric needs an owner and a decision it feeds.
Customer churn analysis in six steps
- Define churn. Put full churn, partial churn and "at risk" into written rules, agreed by sales and finance together. Otherwise every meeting argues about a different number.
- Bring customer data into one place. Order and invoice history, returns, late payments, service visits and complaints must all meet on the same customer key. The structure that joins these sources is usually a data warehouse with ETL; if marketing and campaign data are also in scope, a customer data platform (CDP) is worth considering.
- Run a cohort analysis. Group customers by the period in which they were won and check how many in each group are still active after 6, 12 and 24 months. If losses cluster among customers won in a particular period or through a particular channel, the problem may be how you acquire customers rather than what you sell.
- Choose your early warning signals. Look at the final months of customers you have already lost. Did the order interval widen, did the product range narrow, did payments slip, did service requests rise? These become the inputs to a risk score.
- Produce a risk score. Start with a simple rule set, then move to a statistical or machine learning model once you have enough history. As with demand forecasting, the model must be back-tested on past periods.
- Attach an action to every risk. For a high-risk customer, who does what, and by when? A rep visit, a call from a director, a technical support review or a look at payment terms. Record the outcome so you can measure which interventions actually work.
Skip step six and the analysis stays a report. If the same risk list lands on the same rep every week and nobody acts on it, even the best model will not win back a single customer.
Early warning signals: where to look
Churn signals usually build up outside the sales team. This checklist shows which system holds each signal and what to watch for:
| Signal | Source system | What to watch for |
|---|---|---|
| Order frequency | ERP / e-commerce | Usual interval stretching noticeably |
| Basket breadth | ERP | Fewer product groups bought |
| Payment behaviour | Accounting | Growing overdue balance |
| Service and faults | Field service system | Repeat faults, longer resolution times |
| Complaints and requests | Email, phone, chatbot | The same complaint coming back |
| Relationship | CRM | Change of account manager, long gaps without contact |
For companies selling through dealers, most order data is generated in a B2B dealer portal, and how often a dealer logs in is itself a strong signal. Fault and resolution-time data sits in field service management software.
To make complaints measurable, incoming requests need to be logged by topic; on web and messaging channels a customer service chatbot can handle that first capture. An NPS survey is another signal source. By Bain's definition it is the percentage of customers scoring 9–10 on a 0–10 scale minus the percentage scoring 0–6 (Bain & Company, Net Promoter System). A customer whose score drops may be telling you something before their orders do. For survey design and closing the loop, see our Net Promoter Score guide.
Two products from our own family can feed this table directly. If online sales run on the SmartSales e-commerce platform, the order-frequency signal can be tracked in the store's Orders and Customers reports, and the customer list can be exported as CSV or JSON into the analysis layer. If the SmartOperator AI customer agent runs on your website, the number of conversations handed over to your team is reported in its panel against the previous period; a rise in handovers can point to more topics the automated answers cannot resolve.
How we build churn analysis at Digital Bridge
We do not sell an off-the-shelf package. We start by understanding where your data lives today, then confirm scope, phases and cost in a written proposal.
- Discovery and needs analysis. We review which fields your ERP, accounting, CRM and service data sit in, how many years of history you have, and whether the customer key matches across systems. We write the churn definition together with your sales and finance teams.
- Data layer. We consolidate scattered sources in a data warehouse and match duplicate customer records. If customer contact and interaction data is missing or scattered, we set up a CRM around your process.
- Pilot. We start with one product group, region or customer segment. We back-test how many of the customers you lost in the past the model would have flagged in advance, and show the result side by side with your current approach.
- Model and dashboard. We build the risk score with the same approach we use for demand forecasting and predictive analytics: several methods are tried on the same data, the one with the lowest error is chosen, and accuracy is measured regularly. Results flow into a BI dashboard that reps and managers actually use.
- Action loop. At-risk customers become CRM tasks for the account manager, and each intervention and its outcome is recorded. Over time you can measure which actions keep customers.
Customer data may include personal data, especially in fields such as contact names, phone numbers and email addresses. In Türkiye this falls under KVKK, the Personal Data Protection Law (Law No. 6698), which broadly mirrors GDPR in requiring a defined purpose and transparency towards the people whose data you process. We cover the points to settle before modelling in our article on AI and KVKK personal data. For dashboard design in general, see our BI dashboard and KPI guide.
Your next step
You do not need a big project to start. From the last two years of sales data, list the customers who ordered regularly a year ago but no longer do, and go through that list with your sales reps. The length of the list and the variety of reasons will tell you how much the analysis is worth to you.
Then get in touch with us. We will review your data with you and pick the right customer segment for a pilot. For more guides on data-driven decisions, visit our Data & Analytics topic page.