The customer retention measures that let you act are the ones that move before a customer leaves. Retention rates tell you what you have already lost. A per-customer engagement score, built from replies, payments, bookings and bounces, tells you who is drifting while there is still time to call. This article shows how to build that score on the levelpartner.ai platform, alert the account owner when a customer falls below a threshold, and report it next to the two or three lagging measures a board needs.
Lagging measures: what you have already lost
Start with the measures that tell you how well you kept customers last year. They are not early warnings, but they tell you whether the early warnings are working.
Define "customer" and "period" first
Every retention figure depends on two decisions made before any arithmetic. What counts as a customer: a company, a contract, or a person? And over what period: a month, a quarter or a year? A firm that counts contracts will report a different rate from one that counts companies, and both will differ from a monthly figure annualised.
For most services firms, count companies or households, and measure over twelve months. Monthly rates are useful for subscription businesses with thousands of customers. For a book of 60 retainer clients, a monthly rate swings wildly with one cancellation and tells you little. Write the definitions down and keep them fixed, or your trend line will be meaningless.
Customer retention rate
This is the share of customers you had at the start of the period who are still customers at the end. New customers won during the period are excluded, because they say nothing about whether you kept the ones you had.
Suppose the agency starts the year with 64 clients, wins 14 new ones and loses 11, ending on 67. Clients retained from the starting group would be 67 minus the 14 new, which is 53. The customer retention rate is then 53 divided by 64, about 83%. Counting the 67 against the 64 instead would wrongly suggest more than 100%.
The most common error is to include new customers in the numerator, which flatters the figure. A firm that wins many new clients can show a rising count while losing a large share of its existing base.
Revenue retention: gross and net
Customer counts treat every client as equal. Revenue retention weights them by what they pay, which is closer to what the owner cares about. There are two versions.
Gross revenue retention takes the annual revenue from the starting customers, subtracts what was lost through cancellations and downgrades, and divides by the starting revenue. It can never be higher than the starting figure. Net revenue retention also adds the growth from those same customers: upgrades, extra services and price increases. It can end up higher than the starting revenue, which means your existing customers are growing faster than you are losing them.
Say the 64 starting clients are worth an illustrative £3.2m a year. If lost clients take £380,000 of that and downgrades take another £90,000, gross revenue retention would be (3,200,000 minus 380,000 minus 90,000) divided by 3,200,000, about 85%. If the remaining clients also buy £310,000 of extra work, net revenue retention would be (3,200,000 minus 470,000 plus 310,000) divided by 3,200,000, which is 95%.
For a firm with stalled revenue, net revenue retention is often the most revealing number. If net retained revenue is below the starting revenue, every new client you win is partly filling a hole. If you can push it above the starting revenue, growth from existing customers alone starts to move the top line.
Cohort tables
Retention rates for the whole book average together customers of very different ages. A cohort table splits them by the period they joined and shows how many from each group remain after one, two and three years. It reveals patterns that a single rate hides.
Imagine three cohorts. Year A: 18 clients at start, 14 after one year, 12 after two, 11 after three. Year B: 22 at start, 15 after one year, 13 after two. Year C: 20 at start, 17 after one year. The first-year column would read about 78%, 68% and 85%. A dip like year B's is the cue to ask what changed in onboarding that year, and a rise like year C's is the cue to find out what to keep.
Build the table in a spreadsheet from your CRM's start and end dates. Update it once a quarter. The first-year column is usually the most useful, because early losses point to onboarding and fit problems you can fix.
Leading measures: an engagement score per customer
The measures above lag. A customer who stops replying, pays late and cancels a review meeting has usually decided to leave weeks before the contract ends. The platform's contact engagement score turns those behaviours into one number on each contact, which goes up when a customer engages and down when they do not.
Choose the behaviours by looking back. Take the customers who left in the last two years and ask what they did, or stopped doing, in the three months before. Typical candidates:
- Replies to your emails, texts or chat messages.
- Payments received on time.
- Review meetings booked, confirmed or cancelled.
- Forms or surveys returned, such as the quarterly check-in.
- Emails opened and links clicked, weighted low.
- Bounces and unsubscribes, which subtract points.
Keep the behaviours that showed up in most departures and rarely in customers who stayed.
Set up the scoring rules
- Go to Settings and open Manage Scoring.
- Leave the toggle on Draft while you build the rules, so live scores are not affected.
- Click Add New Rule.
- Choose an action type, such as Contact Replied, Payment Received, Appointment Booked or Appointment Status Changed, and set the points to add or subtract.
- Repeat for each behaviour, then review the Action and Calculation columns in the scoring rules table.
- Switch the toggle to Publish to start scoring all contacts.
Weight the behaviours by how strongly they predicted departure in your own history, not by how easy they are to measure. A reply from the main contact is worth more than ten email opens. An appointment cancelled without a new date should subtract points.
Make the score fall when nothing happens
Scores do not decay by themselves. A customer who simply goes quiet keeps the score they had, which is exactly the customer you most need to see. Build a workflow on a schedule, for example every month, that subtracts points from contacts tagged as active customers, using the action that modifies the engagement score. Engaged customers earn the points back; quiet ones drift down.
Alert the owner below a threshold
Build a second workflow on the Contact Engagement Score trigger, set to fire when the score drops below your threshold and limited to contacts tagged as customers. Its actions create a task for the account owner to call within two working days and send them an internal notification naming the customer and the last thing they did. The score also shows on the contact record and beside each conversation, so the owner sees it in context.
Suppose the agency scores replies at plus 5, on-time payments at plus 10, confirmed review meetings at plus 10, cancelled meetings at minus 10 and bounces at minus 15, with a monthly workflow taking 10 points off every client. A client below 20 triggers a call task. A client who pays on time but never replies or books a review would gain 10 a month and lose 10, so they would sit still rather than climb. Add one cancelled meeting and they would start to fall towards the line, which is the drift you want flagged.
Set the threshold from your own data. Score every current customer for a month before switching the alert on, then set the line just above where your recent leavers sat. Scoring applies to every contact in the account, so tag customers and filter on that tag rather than writing rules only for them. For the manual health field some firms run alongside the score, see CRM retention strategies.
Which measures to report to whom
The managing director and the board need net revenue retention, gross revenue retention and customer retention rate, quarterly and annually. Client services leads need the cohort table and the leading indicators, monthly. Account owners need their own leading indicators, weekly, in a dashboard that shows which accounts need attention.
Keep the board pack to three numbers and a trend. Adding more retention measures to a senior report rarely leads to better decisions; it gives everyone something to debate without changing what anybody does.
Measures that mislead firms of this size
Churn rate is simply the mirror of retention rate, and reporting both adds nothing. Lifetime value calculations are fragile with small numbers: a single long-standing client can double the average, and the figure is easy to misuse when justifying acquisition spend. Satisfaction scores from surveys are useful as prompts but weakly linked to retention in B2B services, where a satisfied client can still leave because their budget is cut.
Monthly retention rates, annualised, are the most misleading of all for small books. One cancellation in a month of 60 customers looks like a dramatic fall. Use annual or rolling twelve-month figures instead.
Celebrating a rising customer count as evidence of good retention. Count growth can hide heavy losses if new business is strong. Always separate new customers from retained ones before you judge retention.
How the four industries adapt the measures
Agencies and consultants with retainers can use the standard measures directly. For mortgage brokers, a retention measure that makes more sense is the share of customers reaching the end of a product term who proceed with the brokerage again. For real estate firms, measure lettings retention by managed properties and landlord revenue, and measure sales retention by repeat and referred instructions from past customers. Engagement scoring suits them too, but the behaviours differ: for a broker, replies to the rate-review message matter more than invoice payments. For account-level tactics once a risk is spotted, see account retention strategies.
Troubleshooting
Scores never change. The rules are still in Draft. Switch Manage Scoring to Publish; scoring then runs for all contacts.
Every customer sits above the threshold. Nothing subtracts points. Check the monthly decay workflow is published and only targets tagged customers.
The alert fires for leads, not customers. The threshold workflow is missing the customer tag filter. Add it, and remove the tasks it created by mistake.
The trade-off
Accurate retention measurement needs clean dates and values in the CRM, consistent definitions and someone who owns the calculation. That is unglamorous work, and early figures may be uncomfortable: net revenue retention often comes out lower than the owner assumed.
You also give up a single headline number. The honest picture needs at least three measures, read together. The reward is knowing where revenue is leaking and whether your efforts are working, which a single satisfying percentage cannot tell you. For what to send a customer once the score flags them, see retention campaign ideas.
Frequently asked questions
What is a good customer retention rate for a services firm?
There is no reliable universal benchmark, because definitions, contract lengths and industries vary so much. The more useful comparison is with your own previous years and between your own cohorts. Aim for steady improvement and investigate any cohort that performs worse than the rest.
Which customer retention measures predict a customer leaving?
Behaviours, not rates: replies, on-time payments, review meetings kept and cancelled, and bounces. Combine them in an engagement score, test the weights against customers who left, and alert the owner when a customer falls below the line.
Should we measure retention by customer or by revenue?
Both. Customer retention tells you how many relationships you keep, and revenue retention tells you how much money you keep. A firm losing many small clients but keeping its largest will show low customer retention and high revenue retention, and needs to decide whether that is acceptable.
How often should we calculate retention measures?
Calculate lagging measures quarterly on a rolling twelve-month basis. The engagement score updates continuously, so the threshold alert is the weekly measure. Annual figures go into the yearly plan. Calculating monthly rates for a small book creates noise rather than insight.
What counts as a lost customer if they simply stop buying?
Set a rule based on your normal buying cycle. For retainers, a customer is lost when the contract ends without renewal. For project or transactional businesses, a customer who has not bought within a defined period, such as two normal buying cycles, can be counted as lapsed for measurement purposes.