Why most chat ROI math gets laughed out of the room
The usual pitch multiplies every chat by an invented conversion lift and calls it revenue. A CFO dismantles that in one question: "would those customers have bought anyway?" A framework survives scrutiny when it states its costs fully, claims its benefits conservatively, and names its own caveats before the CFO does.
The cost side (count everything)
Cost per resolved conversation = (agent time on chat x loaded hourly cost + software cost) / conversations resolved. Two honest inclusions people skip: the time agents spend BETWEEN chats staying available, and the software line. The second is where flat-priced tooling quietly changes the math — a per-agent, per-feature stack grows the denominator's cost with every hire; a flat plan does not.
Benefit 1: chat-assisted revenue (with honest attribution)
Count orders or signups where a chat happened BEFORE the purchase, in the same visit or within a defined window — your transcripts plus order timestamps give you this. Then say the honest sentence out loud: "some of these would have converted anyway." Claim the assisted number as an upper bound and let the trend line, not the absolute, carry the argument: if assisted conversions grow when you staff chat better, the channel is working.
Benefit 2: deflection value (the sturdiest number)
Every question your knowledge base and AI absorb is a conversation that did not consume agent minutes. Deflected volume x your measured cost-per-ticket is real, defensible saving — it survives scrutiny because both factors come from your own books. Track it monthly; it is usually the number that grows fastest as your deflection layers mature.
Benefit 3: the concurrency dividend
Compare cost per resolved conversation across your channels. Chat's advantage is structural — one agent, several conversations — so the same demand handled in chat instead of phone shows up directly as cost avoided per conversation, using entirely internal numbers.
What NOT to claim
- Industry-average conversion lifts applied to your traffic — borrowed numbers, instantly challenged.
- Every chat-touched sale as chat-caused — state assisted, not caused.
- Satisfaction-score deltas as dollars — keep CSAT as a health signal, not a revenue line.
The one-page version
Costs: agent chat hours x loaded rate + software. Benefits: assisted revenue (upper bound, trend), deflection savings (volume x own cost-per-ticket), channel cost advantage (chat vs phone per resolution). Caveats stated in the same document. Your operating metrics feed the model; transcripts and the resolution dashboard are the audit trail. On MyLiveChat's flat pricing, the software line stays constant as the team grows — which, in this framework, is simply a denominator that stops moving.
Show the range, not a single number
Every chat ROI calculation rests on assumptions — what a deflected contact would have cost,
what share of assisted revenue chat can claim, how much of an agent's time is really chat. A single
confident figure invites an argument about whichever assumption the reader distrusts most, and the
argument is usually won by whoever is most senior rather than most correct.
Present three scenarios instead: conservative, likely and optimistic, with the assumption that
changes between them named explicitly. The conversation then shifts from whether your number is right
to which assumption the business believes, which is a far more productive discussion and one you can
actually resolve with data later.
Make the conservative case the headline. If chat pays for itself under assumptions your sceptical
colleague chose, the case is made; if it only works under the optimistic case, you have learned
something important before staking your credibility on it.
The costs that appear after year one
First-year ROI cases are usually too kind because the costs that arrive later are invisible at the
start. Budget for them and the case survives its second review.
The recurring ones are content maintenance — canned replies and help articles go stale and
someone has to keep them true — and the management overhead that appears once more than two
people answer chat: scheduling, reviewing, onboarding replacements. Neither is large, both are real,
and neither appears in a first-year spreadsheet.
Turnover is the one people miss entirely. If the role has any churn, the cost of recruiting and
training a replacement is a genuine cost of running the channel, and it is far larger than the software
line that everybody scrutinises.
When the honest answer is that chat does not pay
Sometimes the numbers do not support it, and the credible thing is to say so. Chat struggles to pay
for itself on a site with very low traffic, in a business where the average order value is small and the
questions are rare, or where nobody can realistically be available during the hours visitors are
active.
In those cases the answer is usually not to abandon the channel but to change its shape: narrow the
hours to when it converts, restrict the widget to the pages where questions actually arise, or run it
offline-first as a capture form with a fast reply. All of these preserve most of the benefit at a
fraction of the cost.
Being the person who reports honestly that a channel does not pay is also how your positive numbers
get believed later. An analysis that always concludes in favour of the thing being analysed is not
read as analysis for very long.