Cost

Cost per conversation: a metric worth tracking

How to compute what one customer conversation costs you today, and what to do with the number.

· 4 min read · Editor: Max

Most small businesses can tell you their cost per lead. Marketing spend divided by leads generated is a standard number, and agencies quote it constantly.

Far fewer can tell you what it costs to have a conversation with one of those leads once it arrives. That number sits between the marketing budget and the revenue line, and it explains a surprising amount about why a funnel underperforms.

The basic calculation

Cost per conversation is deliberately blunt: total cost of handling inbound conversations, divided by the number of conversations handled, over the same period.

The numerator. Everything it costs to run the function for a period — say the last twelve months:

  • Salaries of everyone who answers inbound messages, at the share of their time actually spent on it.
  • Employer contributions and payroll taxes on those salaries.
  • Tooling: CRM, helpdesk, messaging platforms, telephony, priced per seat.
  • Management and training time, at the relevant salary.
  • Recruitment costs incurred during the period.

The denominator. Count distinct conversations, not messages. A conversation is one customer thread on one topic, however many back-and-forths it contains. Count everything that required a human reply, including the ones that went nowhere.

Divide. The result is uncomfortable the first time, and that is the point.

Refining it until it is useful

The single blended figure tells you the scale of the problem. Splitting it tells you what to do.

By channel. Email, WhatsApp, Telegram, phone, web form. Handling times differ sharply between channels, and so does the mix of routine and complex. A channel that looks cheap by volume may be expensive per conversation.

By outcome. Separate conversations that produced a quote, conversations that produced a sale, and conversations that ended without either. Your cost per conversation is one number; your cost per conversation that reached a quote is usually two or three times higher, because the unconverted ones consumed cost too.

By type. The most valuable split. Sample a hundred conversations and sort them into three buckets:

  1. Routine. Answerable from your existing product, pricing and policy information without judgment. Shipping questions, availability, opening hours, specifications, standard pricing.
  2. Semi-routine. Needs a few facts gathered and a standard document produced. Most quoting work sits here.
  3. Judgment. Negotiation, complex configuration, complaints, anything where a person's read of the situation changes the outcome.

Multiply each bucket's share by your blended cost per conversation. You now know roughly what you spend per year answering questions that need no one.

Cost per conversation only becomes actionable when you know what share of those conversations needed a human at all.

Two numbers that make it sharper

Conversations per staffed hour. Divide conversations handled by staffed hours in the period. If the number is low, either your handling is slow or your people are spending their time on something other than conversations. Both are worth knowing.

Cost of unanswered conversations. Count the inbound messages that arrived outside staffed hours and received no reply until the next working day — or none at all. Multiply by your conversion rate and your average order value. That is not a precise figure, but it is a defensible order of magnitude, and it is usually the largest number in the exercise.

The second one tends to change decisions. Cost per conversation measures what you spend. The unanswered figure measures what you never got the chance to spend it on.

What the number is for

Once you have it, cost per conversation does three jobs.

It prices automation honestly. Any tool that handles inbound messages should be evaluated against this figure, not against a salary. If routine conversations cost you a meaningful amount each and make up half your volume, the comparison is arithmetic rather than opinion. Axelion and every alternative should be judged the same way.

It prices marketing honestly. A campaign that doubles inbound volume also doubles handling cost. If your cost per conversation is high and your conversion rate is flat, more volume makes the unit economics worse, not better. Teams that do not track this discover it only when the quarter closes badly.

It exposes where documentation is missing. A high share of routine conversations usually means the answers exist somewhere in the business but not anywhere a customer or a new hire can find them. Writing them down is cheap and reduces the numerator directly.

Doing it without a data team

You do not need instrumentation to start. A workable first pass takes an afternoon:

  1. Export the last full month of inbound threads from each channel.
  2. Count distinct conversations per channel.
  3. Take payroll and tooling cost for that month, apportioned by time spent on inbound.
  4. Divide for a blended figure.
  5. Sample a hundred threads at random and classify them routine, semi-routine or judgment.
  6. Count threads that arrived outside staffed hours.

That gives you a blended cost, a routine share and an uncovered-volume figure. Rough numbers that exist beat precise numbers that do not.

Rerun it quarterly. The trend matters more than the absolute value: if volume is growing and cost per conversation is flat, you are scaling by hiring. If it falls while volume grows, something structural has improved.

Most teams find the number lands higher than expected and the routine share lands higher still. That combination is the case for changing something — and it is also the baseline you will measure the change against.

For the underlying cost inputs, see what a 24/7 sales desk actually costs.

Axelion answers in seconds, around the clock. It replies in your website chat, recommends the right product from your pages, saves every lead and hands complex cases to your team. WhatsApp and Telegram on Business. See it on your own website →

Related articles

What a 24/7 sales desk actually costsAutomation without layoffs: moving people up the funnel

← All articles