Pricing Guide

AI customer service ROI: how to build the business case for chat and email

Build an AI customer service ROI business case for chat and email: the formula, a worked model, sensitivity tables and the costs teams forget.

Author
By the Open Team
|Updated November 17, 2026|9 min read

AI customer service ROI comes down to one comparison: what a resolved chat or email costs you today with a human, against what it costs when an AI agent resolves it. Multiply that gap by the number of conversations the AI resolves, discount for savings you can't realise straight away, then subtract the AI's cost and your internal effort. This post turns that into a formula, a worked model and a sensitivity table you can take to finance.

Every input in the worked example is labelled as an example. Replace them with your own numbers, or run a first pass in Open's ROI calculator.

TL;DR

  • Formula: monthly net savings = resolved conversations × (human cost per ticket × realisation factor − AI cost per resolution) − internal running cost.
  • Resolution rate moves the result most, not the AI's price. In the example, a 15-point change in resolution rate is worth about $10,000 a month; a 15-cent change in price is worth about $2,000 at most.
  • Use resolution, not deflection. A customer who gave up isn't a saving. Count only conversations closed without a human that stayed closed.
  • Apply a realisation factor. Human cost per ticket only turns into money when it changes hiring, overtime or outsourcing. Start at 50 to 70 percent.
  • In the example model (20,000 chat and email tickets a month, labelled inputs), net savings land around $39,000 to $41,000 a month at Open's $0.75–$0.90 per resolution.

Why chat and email need their own business case

Voice and call-centre cases run on minutes, carrier fees and handle time. If that's your channel, read AI vs human call centers: the real cost and ROI. Chat and email differ in three ways that change the maths:

  1. Cost per ticket is lower, but volume is higher. A chat usually costs less to handle than a phone call, so each resolution saves less, but there are usually many more of them.
  2. Resolution rates differ by channel. Written questions are easier to classify and answer from a knowledge base. Email adds long threads and attachments, so it can resolve at a lower rate than chat. Model the two separately.
  3. Carrier costs are close to zero. There's no per-minute telephony bill to pass through, which keeps the AI side of the equation simple.

For context, Open's ROI calculator assumes AI handles 50 to 70 percent of tickets for companies with mainstream helpdesks and decent documentation. At Trunkrs, Open automates 80 percent of delivery support across email and chat. For what drives those rates up or down, see how much customer support AI can actually automate.

The AI customer service ROI formula

For each channel c:

Resolved_c        = V_c × R_c
Realised saving_c = Resolved_c × C_c × F
AI cost_c         = Resolved_c × (P + K)

Monthly net savings = Σ Realised saving_c − Σ AI cost_c − O
ROI                 = Monthly net savings ÷ (Σ AI cost_c + O)
SymbolMeaningWhere to get it
VMonthly ticket volume per channelHelpdesk reporting, last 3 to 6 months
RAI resolution rate (closed without a human, not reopened)Pilot data; vendor benchmarks only as a starting point
CFully loaded human cost per ticketFinance (see below)
FRealisation factor: share of human cost that turns into savingsYour hiring plan, BPO contract, overtime spend
PAI price per resolutionVendor quote
KPass-through costs per resolution (messaging or email send fees)Vendor quote
OInternal monthly running cost (owner time, content upkeep)Your estimate

The formula assumes per-resolution pricing. If a vendor bills per conversation or per action, convert it first: divide the price per conversation by the resolution rate to get the cost per resolved ticket. The Agentforce pricing breakdown works through that conversion on a per-conversation price list, and the Zendesk AI pricing guide covers a per-resolution model with its own tiers.

How to calculate cost per ticket

Fully loaded cost per ticket = (agent salaries + benefits + management overhead + tooling + QA + training) ÷ tickets handled per month. Open's ROI calculator notes that most teams land between $4 and $25 per ticket. Calculate it per channel if you can, because email usually costs more per ticket than chat.

Why the realisation factor matters

If the AI resolves 10,000 tickets a month, you don't save 10,000 × your cost per ticket on day one. Agents are on contracts, and freed time first goes to backlog and harder tickets. The saving becomes real when you stop backfilling attrition, cut BPO volume, drop overtime or absorb growth without hiring. A realisation factor of 50 to 70 percent is a defensible starting point for year one. Finance teams trust a model more when it shows this discount openly.

Worked example: 20,000 chat and email tickets a month

Example inputs (replace with yours):

InputChatEmail
Monthly volume (V)12,0008,000
Fully loaded cost per ticket (C)$6.00$9.00
AI resolution rate at steady state (R)70%55%
Realisation factor (F)60%60%
Price per resolution (P)$0.75 to $0.90$0.75 to $0.90
Pass-through cost per resolution (K)$0.05$0.05
Internal running cost (O), both channels$3,000 a month (half an FTE at $6,000 fully loaded)

Step 1: resolved conversations

  • Chat: 12,000 × 70% = 8,400
  • Email: 8,000 × 55% = 4,400
  • Total: 12,800 a month (64% of volume)

Step 2: human cost avoided, then realised

  • Chat: 8,400 × $6.00 = $50,400
  • Email: 4,400 × $9.00 = $39,600
  • Gross: $90,000. Realised at 60%: $54,000

Step 3: AI cost

At $0.90 per resolutionAt $0.75 per resolution
Resolutions (12,800 × P)$11,520$9,600
Pass-through (12,800 × $0.05)$640$640
Internal running cost$3,000$3,000
Total cost$15,160$13,240

Step 4: net savings

At $0.90At $0.75
Monthly net savings$38,840$40,760
Annual net savings (steady state)$466,080$489,120
ROI on spend2.6x3.1x

Handoffs to a human aren't billed under Open's pricing, so the 7,200 conversations the AI passes to agents add no AI cost in this model.

Year one with a ramp

Resolution rates climb over the first months as the knowledge base fills gaps and you add actions. Open's ROI calculator page says AI-ready teams typically handle 30 to 40 percent of volume in the first month and 60 to 70 percent by month three (the page calls this deflection, so confirm it as resolution in your own pilot). Using an example ramp of 35 percent, then 50 percent, then 64 percent from month three onward, year one resolves about 145,000 conversations. At $0.90 per resolution, that gives roughly $438,000 in year-one net savings, about 6 percent below the steady-state figure. This applies the steady-state chat and email mix to every month, which is a simplification.

Sensitivity: which inputs move the number

Same example, blended across both channels (C = $7.20, F = 60%, K = $0.05, O = $3,000):

Resolution rateNet monthly at $0.90Net monthly at $0.75
40%$23,960$25,160
55%$34,070$35,720
70%$44,180$46,280

Moving from 40 to 70 percent resolution adds about $20,000 a month. Moving from $0.90 to $0.75 adds about $1,200 to $2,100. When you compare vendors, weight demonstrated resolution rate on your own conversations far above headline price.

The break-even test

An AI resolution pays for itself when C × F exceeds P + K. With the example email cost of $9.00 and a $0.95 all-in AI cost, break-even needs a realisation factor of only about 11 percent. For chat at $6.00 it's about 16 percent. So even a cautious finance team should accept that each resolved chat or email saves money. The open questions are volume and resolution rate.

Costs teams forget to include

  • Knowledge work. Someone has to own the knowledge base, review flagged gaps and approve changes. That's the O in the formula. Don't set it to zero.
  • Integrations for actions. Order lookups, refunds and account changes are where resolution rates rise past FAQ level. Budget engineering time if the vendor doesn't build these for you.
  • QA during rollout. Plan for heavier review in the first weeks, then sample.
  • Platform and seat fees. Some vendors charge seats or platform fees on top of usage. Open has no setup, seat or per-message fees; see pricing.
  • Billing definitions. Read what counts as a billable resolution and whether handoffs are billed. That definition can move your AI cost more than the headline rate. How AI customer service pricing works covers what to check in each model.

For comparison, Intercom prices Fin at $0.99 per outcome. At the example's 12,800 resolutions, that's $12,672 a month before any seat fees, against $9,600 to $11,520 at Open's range. Fin on Intercom's own helpdesk adds $19 per helpdesk seat a month; on other helpdesks there are no seat fees and a 50-outcome monthly minimum.

Benefits to list but not monetise

Some gains are real but hard to put a dollar figure on. List them in the business case without adding them to the total:

  • 24/7 coverage on chat and email without night shifts
  • First response in seconds rather than hours on email
  • Consistent answers across languages and channels
  • Agents spending time on complex cases instead of "where is my order"

If CSAT or retention moves during the pilot, add it to the next version of the model with real data.

How to turn the business case into a decision

  1. Pull 3 to 6 months of volume by channel and agree a cost per ticket with finance.
  2. Run the numbers in the formula above, or plug them into Open's ROI calculator for a first pass.
  3. Pilot on live traffic in one channel, often chat or email, and replace the example resolution rate with measured data.
  4. Present three scenarios (conservative, expected, strong) using the sensitivity table, and show the realisation factor openly.

Open charges $0.75–$0.90 per resolution, depending on monthly volume, and handoffs to a human aren't billed. The first 120 days are backed by a refund guarantee described on the pricing page. The same Agent 5 setup that answers chat and email also covers WhatsApp, social, Slack and voice if you extend the case later.

Sources

Frequently Asked Questions

Multiply the number of conversations the AI resolves by the gap between your fully loaded human cost per ticket (discounted by a realisation factor) and the AI's cost per resolution. Subtract internal running costs to get net savings, then divide by total AI and running spend for the ROI multiple.

It depends on documentation quality and how many actions the AI can take in your systems. Open's ROI calculator assumes AI handles 50 to 70 percent of tickets for companies with mainstream helpdesks and decent documentation. Email can resolve at a lower rate than chat because threads are longer, so model the channels separately.

Deflection counts any conversation that did not reach a human, including customers who gave up. Resolution counts only conversations the AI closed that did not reopen or escalate. Build the business case on resolution, because deflection overstates savings.

Open charges $0.75–$0.90 per resolution, and handoffs to a human are not billed. Intercom lists Fin at $0.99 per outcome (checked 17 November 2026). If a vendor bills per seat or per conversation, convert the quote into cost per resolved conversation at your volume before comparing.

With per-resolution pricing and no setup fee, AI cost only arises when a conversation is resolved, so each resolution is net positive from the start if your realised human cost per ticket exceeds the AI price. The bigger timing question is when savings show up in headcount or outsourcing, which depends on your attrition and hiring plan.