‘Does customer service pay off?’ sounds like a question about money, but it is almost always a question about time. An owner sees how many hours each week go into replying to people who have not bought anything yet and wants to know whether that effort comes back as sales. The answers circulating online fall into two groups: vague claims about building loyalty and other companies’ reports that cannot be verified against your own business. This article is about something else—what you can genuinely calculate from your own data, what you cannot, and why the latter does not mean the effect is not there.
Three questions that sound like one
A conversation about the profitability of customer service usually falls apart in the first minute because the participants answer three different questions using the same words. Each question requires different data and produces a different level of certainty.
- Does customer service pay off?A question about the entire function. There is no calculation for it, because there is no version of your business in which nobody responds to customers that can be compared with the present one.
- Does this channel pay off?A question about live chat, callbacks or forms individually. An answer is possible, but only in relative terms: which channel closes more cases for a similar investment of time.
- Did this conversation lead to a sale?A question about one contact. The data can be conclusive here, but the conclusion applies to one situation and cannot automatically be extended to the whole function.
Confusing these questions causes two errors. First, someone shows several conversations that ended in an order and draws a conclusion about the entire department—even though these particular cases were already heading towards a purchase. Second, and more dangerously, the absence of hard evidence at the level of the whole function is treated as evidence of no effect, even though it results from the lack of a comparison.
Start with the narrowest question, not the broadest. Individual contacts that ended in a goal provide hard evidence, a channel comparison provides context, and the conclusion about the whole function is a conscious decision made with incomplete data—which is worth stating plainly.
What can be attributed directly
The strongest available evidence is a contact followed by an order or ticket in the same session. One person, one visit, events in a known sequence—nothing has to be inferred from aggregate reports.
The condition is that the website must tell the system that something happened. Business goals are defined in advance in the panel, and the website records them with a call in its own code—only then do reports show the funnel to the goal and let you move from a metric to specific conversations and sessions. If the website does not send the order value as the goal value, no system knows it and has no basis for guessing it. We discuss choosing the events themselves in our article on website micro-conversions.
Scroll the table sideways to see all columns.
| What the data shows | What you can conclude | What it does not prove |
|---|---|---|
| A conversation followed by an order in the same session | The contact was the final step before the decision | That the order would not have happened without the conversation—some people would have bought anyway |
| A conversation followed by a ticket in the same session | The contact turned anonymous traffic into a case that can be managed | That the case will end in a sale—it is only the beginning |
| An order with no prior contact | The path worked without customer service involvement | That customer service was unnecessary—it may have helped during an earlier visit by the same person |
| A conversation with no goal recorded in the same session | The contact did not close a sale during that visit | That it had no value—the effect may be delayed or entirely unmeasurable |
This method has one limitation: session-level attribution understates the contribution of customer service because it sees only contacts that close the case immediately. A conversation followed by the customer returning two days later falls out of the calculation entirely. The number in this section is therefore a lower bound, not an answer—and it should be labelled accordingly.
A goal defined in the panel, recorded by the website, a funnel to that goal, and a path from the metric to individual conversations and sessions. Start for free.
What can be attributed indirectly—and how not to overstate it
The second kind of evidence is a contact followed by a customer returning and buying several days later. You can see it when one person’s successive visits form a single history, together with the traffic source and links to the conversation, callback and ticket. This is still data, not a guess—but the relationship between the conversation and the later purchase is evidence, not proof.
- Tuesday, search traffic
A live-chat question about delivery time, an answer, no order
- Thursday, direct traffic
The same person returns to the product page and leaves without making contact
- Saturday, visit from a bookmark
An order placed independently, without a conversation
- Two stories, one sequence
The conversation resolved a doubt about timing, or the customer would have returned anyway—the data does not determine which story is true
- Set the return window in advance and do not change it after seeing the result.
- Keep indirect attribution in a separate column; never combine it with direct attribution in one total.
- Check what proportion of orders happen with no contact at all—this is the baseline for everything else.
- Reject links where the conversation topic has nothing to do with what the customer eventually bought.
- Do not invent sources for sessions that have none or attach them to a more convenient channel.
- Record how many cases you rejected next to the result—that number indicates how certain the attribution is.
We explain the rules for linking visits and reading traffic sources in our article on the customer journey on a website. One rule matters most for this profitability calculation: sessions with no recorded source are marked as uncovered instead of being assigned a source by intuition. This tells you how much of the story you do not know, so you can state it next to the result.
If you collect a rating after a conversation, keep it outside the sales calculation. A high rating does not prove an impact on an order, and a low one does not prove the absence of impact—it is separate information about contact quality, useful for entirely different decisions. Combining it with value attribution produces a number nobody knows how to interpret.
An honest calculation of the customer service cost
The other side of the calculation is often treated even more casually. Someone enters the tool’s subscription fee and considers the matter closed, even though this is usually the smallest component. The real cost of customer service is people’s time—including the parts that are easy to forget because they do not have their own invoice.
Scroll the table sideways to see all columns.
| Cost component | How to calculate it | Common mistake |
|---|---|---|
| Time spent in conversations | The hours actually spent responding, costed at the rates of the people doing the work | Counting only the typing, without reading the history or checking orders |
| Standby time | Hours when someone must be available even if no enquiries arrive | Ignoring it because it is not visible in conversation statistics |
| Switching between tasks | Add several minutes for every interrupted task when customer service is an addition to another role | Assuming that a reply written in one minute costs one minute |
| Implementation and maintenance | Time spent on templates, tags, goal configuration and conversation reviews, spread across months | Treating it as a one-off cost that never returns |
| The tool | The amount on the invoice, added at the end | Starting the calculation with this item and ending there |
It is also worth calculating the cost of maintaining a short response time, because that determines staffing. We explain how to measure it in our article on first response time in live chat, and cover the range of tool and implementation costs in our article on how much live chat costs. One rule is enough here: calculate the cost in hours first, then convert those hours into money.
Instead of estimating the workload from memory, take it from reports: the distribution of enquiries over time, the number of cases per person, and the specific conversations that took the most time. See how it works.
Effects you cannot calculate but that still exist
Most of customer service’s impact on business performance leaves no trace in any system because it consists of something not happening. There is no ‘complaint that never happened’ event. There is no ‘basket the customer did not abandon because they received an answer’ event. An absence of evidence is not evidence of no effect—it is simply the nature of negative events.
- Complaints avoidedAn explanation before the purchase replaces a case after the purchase. Fewer complaints are visible, but there is no way to identify the specific complaints that never arose.
- A question that did not become an abandonmentA person who received an answer and stayed on the website generates no event. Only the person who leaves generates one.
- A recommendationA customer recommends the business away from your website—in conversation, in a group or in a message. The resulting traffic is classified as direct or remains uncovered.
- A return after a long timeA contact from several months ago returns as an order when the need arises. No reasonable attribution window can include it.
In a budget discussion, these effects should be named rather than priced. A figure invented by feel ruins the entire calculation because it turns the discussion into an argument about the multiplier instead of the decision. An honest format is simpler: a directly calculated number, an indirectly attributed number alongside it with the uncertainty marked, and below them a list of effects with no numerical value. Such a document is harder to challenge than a statement in which everything has a price.
There is one exception. You can actually inspect where people hesitate before asking or leaving—in session recordings and heatmaps. This does not turn an unmeasurable effect into a number, but it does turn it into concrete material for discussion. We explain how to include it in a regular review in our article on the customer service report.
Session recordings and heatmaps reveal hesitation, repeated returns to the same place and exits without a question. Explore behaviour analytics.
A simple calculation you can complete in a spreadsheet in an hour
The purpose of this calculation is not to produce one number for a slide, but to separate what you know from what you assume. Choose one complete month that was neither campaign-driven nor a holiday period. The entire exercise consists of five steps and a six-row spreadsheet.
- Step one: count the hours.Add up the time spent on customer service during the selected month, including standby time and task switching. Convert it into money only after adding up the hours.
- Step two: list contacts with a goal in the same session.This is your hard number. If the website sends the goal value, use it; if not, stick to the number of orders and tickets.
- Step three: list contacts followed by a return within the chosen window.A separate column, a separate total, and a description of the method in the adjacent cell. Reject cases where the conversation topic is unrelated to the purchase.
- Step four: find the baseline.Count orders placed without any contact. Without this number, you cannot separate the impact of customer service from ordinary traffic that converts on its own.
- Step five: add what you did not calculate.A short list of effects with no numerical value, plus the share of uncovered sessions. This is not a footnote—it is part of the result.
The table below illustrates only the structure of the calculation; it is not data from any business. The figures are invented and show only which fields to complete. If a field remains empty, that is information in itself—it usually means no goal has been defined or the website is not recording it.
Scroll the table sideways to see all columns.
| Spreadsheet field | Where the value comes from | Illustrative example; figures are invented |
|---|---|---|
| Customer service hours per month | Schedule plus added standby and interruption time | 38 hours |
| Cost of those hours | Hours multiplied by the rates of the people providing support | An amount calculated from staff rates, not the tool invoice |
| Contacts with a goal in the same session | The funnel to the goal and drill-down to individual conversations | 17 contacts |
| Contacts followed by a return within the window | One person’s visit history after rejecting unrelated topics | 9 contacts, 4 rejected |
| Orders with no contact | The same report, showing paths with no conversation, callback or ticket | The baseline for the two fields above |
| What was not calculated | A descriptive list plus the share of uncovered sessions | Deliberately left without a numerical value |
After an hour, you have something no external report can give you: a result calculated from your own traffic, with the limit of its reliability clearly marked. Repeat it after a quarter using the same method and you will see the direction of travel. And if the first attempt shows that you have nothing to calculate because the website records no goals, that too is a result—and the best possible first step.
