How to calculate the ROI of conversation intelligence

Short answer

The ROI of conversation analytics has three separate layers, and mixing them up is the most common mistake. The first is time saved: hours people spent reading and counting conversations by hand. The second is a changed decision: one that would have been made differently without the analytics. The third is a proven outcome: revenue or savings measured after that decision.

Only the first layer can be calculated precisely. The second and third need careful attribution. An honest ROI report shows the three figures separately and states plainly which was measured and which was estimated.

Why ROI is hard here

Conversation analytics sells nothing and cuts no cost directly. It provides information, and value comes from the work done with it. That sets it apart from a chatbot or automated calling, where there is a direct measure such as "how many conversations the bot closed". Here the chain is long: insight → decision → change → outcome, and other factors act on every link.

Because of this it is easy to fall into one of two extremes: not calculating ROI at all, or crediting all sales growth to analytics. Both are wrong.

Layer 1: time saved

Before analytics, did anyone read conversations to learn what customers think? Usually yes: a sales lead skims chats for a few hours a month, a marketing manager copies complaints into Excel, a support lead counts repeat contacts for a report. Record those hours.

1. 1Who was readingList the roles and the hours each spent on this per month.
2. 2Hourly costWork out each role's hourly cost (salary plus taxes).
3. 3What remainsReading evidence still takes time after analytics — subtract it.
4. 4The differenceHours saved per month × hourly cost = layer one.

Layer 2: a changed decision

This layer is measured with records, not numbers. Over each quarter, keep a log of decisions made on the basis of analytics: date, decision, which insight it rested on, what the alternative was, who decided. At the end of the quarter, ask: which of these decisions would have gone differently without the analytics?

The decision log is the most reliable part of an ROI report because it rests on leaders' own statements, not a model calculation. One or two genuinely changed decisions a quarter is already a sign of value.

Layer 3: a proven outcome

If a decision changed, its outcome can be measured. For example, the follow-up rule was changed based on loss reasons and the share of followed-up leads that bought went up. Two questions: how big is the increase, and how much of it belongs to this change?

• Strongest evidence: the change was applied to one group and not another, and the two were compared.
• Medium evidence: the same metric was compared before and after, with season and campaigns accounted for.
• Weak evidence: "sales went up after the change" — nothing else was checked.

Leave figures based on weak evidence out of the ROI report, or label them separately as "unverified".

A simple ROI table

The figures below are illustrative and belong to no real company or Vexvon customer. They only show the structure of the calculation.

1. 1CostPlatform, setup, field maintenance and the time spent reading evidence.
2. 2Layer 116 hours × hourly cost × 3 months.
3. 3Layer 3Measured extra revenue or savings, only with strong or medium evidence.
4. 4Layer 2Shown separately through the decision log, not as a number.
5. 5ROI(Layer 1 + Layer 3 − cost) / cost. Layer 2 is added as commentary.

The attribution rule

Analytics produced an insight, the sales team changed the script, marketing updated the ad, sales rose. How much of the rise belongs to analytics? The honest answer: it cannot be fully separated. A practical rule: record the outcome as the result of the decision, not of the analytics, and note the analytics' role as "this decision was based on its insight". Do not invent percentage splits.

Mistakes that inflate ROI

• Crediting all sales growth to analytics.
• Counting time for work nobody used to do as "saved".
• Presenting a vendor's or another company's results as your own ROI.
• Leaving out the human time needed to read evidence and maintain fields.
• Multiplying one month's result by twelve — season and chance are ignored.

When ROI will be low

• Conversation volume is small and reading by hand is easy anyway.
• Results have no owner and no decision changes.
• Everyone already knows the problem; there are simply no resources to fix it.
• Fields change every month and no comparable data builds up.

How to present it to leadership

Present the ROI report once a quarter on one page. At the top, three figures: the money value of layer one, the number of changed decisions, and the measured outcome with its level of evidence. Below, two or three lines from the decision log: which insight, which decision, who made it, when the outcome will be checked. At the bottom, the cost and one target question for next quarter. This format shows the CFO the number and the CEO the decisions, without making any figure look bigger than it is. The first quarter's ROI may be negative — setting up fields and calibrating takes time; say so in advance.

What Vexvon provides for ROI

Vexvon does not promise an ROI figure and does not calculate one in its reports. The raw material for ROI is in the panel: counts and trends of field values by period, comparisons of two periods, the list of conversations behind each value, and lead statuses and loss reasons from the CRM. That is enough to back the decision log with evidence and to compare the same metric before and after a change. More: Vexvon analytics.

Next step

Before a pilot starts, open a decision log and record today's reading hours for layer one. To choose the pilot question, see conversation analytics pilot; we can build the ROI model with your own numbers in a demo.

Further reading on this topic: customer insight dashboard for CEOs.