Back
Blog / 
Strategy & ROI

Building a defensible business case for customer-service AI

written by:
David Eberle
A hand adding a cost slip to a blue balance weighing work saved against costs.

Published · Updated

A business case for customer-service AI should help you decide whether its value justifies the full cost, and which work to start with. It should also make clear what evidence you need before expanding.

Measure the work actually saved after review, corrections and handovers, using comparable requests over the same period. Separate writing time from total handling time and include software, integration and ongoing operating costs. Treat freed hours as capacity unless you can show how they reduce spending.

The method below includes reported experiences from Typewise pilots in 2026 and an explicitly hypothetical calculation. The examples explain what to measure; they are not a forecast for your team.

Step 1: agree the unit of work

Messages, tickets and cases are different things, and AI tools, help desks and finance each tend to count a different one.

  • A message is one inbound or outbound communication.
  • A ticket is a help-desk record. One customer case often spans several tickets: the original request, a follow-up, a reopen after delivery.
  • A case is the customer's problem, from first contact to the point where nothing further is needed.

An assistant that measures time saved per message says nothing about the case. A help desk that reports ticket volume says nothing about repeat contacts. Measure how many messages and tickets make up a case in your own data, and use that relationship consistently in the cost calculation. A finished case can remove human work, and a useful partial result can reduce it. Measure the effort that remains; sending a message alone establishes neither.

In one pilot review, a travel company's service lead described a mismatch in the business case. Her per-ticket handling time covered opening, reading, thinking and writing. Our measurement counted writing time only, a fraction of her figure.

The same customer case could span one ticket or several, depending on whether the customer phoned or kept writing. Nobody could say the average. We needed to reconcile those units before comparing savings.

Step 2: baseline per cohort, per channel, per season

A single "average handling time" hides most of what the business case needs.

  • Cohorts. Support team members who split their day between phone and email have a different per-ticket baseline from those who only write. Specialists handling complex tasks differ again. The travel team above had colleagues who mostly phone, colleagues who mostly write, and one who handled several times the email volume of the others; a single average would have hidden those differences. Baseline each cohort the AI will actually touch.
  • Channels. Chat, email and contact-form requests carry different volumes and different handling patterns. A successful chat pilot does not validate email; the scenarios have to be calculated separately.
  • Seasons. Volume and request types can differ between quiet and peak periods. A baseline measured in one period and applied to the whole year misstates both the saving and the capacity you need.

Ask what the current fastest path is for each ticket type before assuming AI will improve it. A logistics group's project lead told us her team had decided against turning long email templates into snippets. Finding the existing template was quicker than assembling the message; snippets stayed useful for short sections only.

That reported decision matters to the comparison. Rebuilding those emails from snippets may be slower for that team, even if it helps with short, varied replies.

Step 3: separate three kinds of saving

Keep these apart in every table you produce, because they do not add up and they are not interchangeable.

  1. Writing time. The minutes a person spends composing a reply. Check whether a dashboard measures this directly or estimates it from feature usage and session timing. Validate any modeled figure against actual handling effort before valuing it.
  2. Handling time. Everything a person does for the ticket: reading, researching in other systems, deciding, writing and updating records. If research stays manual, a large writing-time saving may produce a modest handling-time saving.
  3. Whole-case resolution. The case finishes without a person. This can remove the human handling of a case, but it needs a clear definition of "resolved", including reopens and partial handoffs. See what counts as a resolution for that definition.

In a second travel pilot review, the sponsor asked where the promised overall saving was. The team lead explained that staff still had to research most requests in the reservation system before answering. The assistant shortened the writing step only.

Some mailboxes could not use it at all because contact-form requests arrived without body text. Those limits help explain why a writing-time figure could not stand for the overall saving.

A worked hypothetical

The numbers below are invented to show the mechanics. They are not a benchmark, not a Typewise result and not a prediction for your team.

For this example only, each case is recorded as exactly one ticket. Tickets and cases coincide, with no repeat tickets inflating the count. Where your help desk records several tickets per case, convert to cases first and run the arithmetic on cases.

Assume an email cohort with 9 minutes of handling per ticket:

  • 3 minutes reading and researching;
  • 4 minutes writing;
  • 2 minutes updating records.

The cohort handles 10,000 tickets a month. Split them into mutually exclusive sets before calculating savings:

  • Autonomous AI entry: 1,500 tickets (15%) of one eligible type enter autonomous handling. In this example, every eligible ticket enters, so eligible coverage and actual entry coincide.
  • Assisted pool: the remaining 8,500 tickets. The assistant is used on 60% of these: 8,500 × 60% = 5,100 tickets.

No ticket appears in both sets.

LineAssumptionResultNote
Assumed writing-time saving40% of writing time1.6 min per ticketOf a 4-minute writing step
As a share of handling time1.6 of 9 min18% per touched ticketResearch and record updates unchanged
Assisted tickets (disjoint from the autonomous set)60% of the 8,500 non-autonomous tickets5,100 ticketsSome ticket types cannot use the assistant at all
Gross writing-time minutes5,100 × 1.6 min8,160 minUnrounded
Review cost for AI drafts5,100 × 0.5 min2,550 minReading a draft is not free
Net assisted saving8,160 − 2,5505,610 min = 93.5 hoursAbout 6.2% of the cohort's 90,000 handling minutes

A headline of "40% time saved" has become about 94 hours a month of capacity. That is still worth having. It is also a fraction of what the headline implied, and the reconciliation is what makes the number survive a finance review.

Now calculate whole-case resolution on the separate autonomous set. Treat the 1,500 tickets as autonomous attempts. Each reached a verified candidate resolution: the performing system confirmed the outcome, with no pending work.

The repeat-contact window still has to pass. Suppose 150 tickets (10%) reopen inside that window. The other 1,350 become verified resolutions once their windows mature.

Keep the resolution rate and the effort saving separate:

  • Resolution state: 1,350 resolved and 150 known unresolved out of 1,500 attempts.
  • Autonomous AI-entry share: 1,500 / 10,000 = 15% of all cases. This excludes the separate assisted pool.
  • Full resolution within the autonomous AI-entry group: 1,350 / 1,500 = 90%, retaining the 150 unresolved cases in that denominator.
  • Full autonomous resolution overall: 1,350 / 10,000 = 13.5% of all cases, using the definitions in what counts as a resolution.

Calculate the human-effort saving across all 1,500 attempts. The 9-minute baseline must represent what those same tickets would have cost people over the same period. Include all residual work on both sides of that comparison.

  • Baseline effort: 1,500 × 9 = 13,500 minutes.
  • Human work on reopened tickets: 150 × 12 = 1,800 minutes.
  • Net effort saved: 13,500 − 1,800 = 11,700 minutes, or 195 hours.

That is an effort calculation over all attempts, not a claim that all 1,500 resolved autonomously. Silence establishes neither resolution nor effort saved. Keep unverified outcomes unknown; value any useful partial work only from measured effort, including the human work still required.

Add the two savings lines before rounding:

5,610 + 11,700 = 17,310 minutes, or 288.5 hours a month.

Each line has a different measurement, confidence and risk. They can be added here because the sets are disjoint and tickets equal cases. No ticket or case is counted twice.

Step 4: subtract the residual human work

Every automated or assisted ticket can leave work behind:

  • a draft that is reviewed and edited rather than sent as is;
  • a case the AI handled "partially", where a person completes the action;
  • a reopen, where the customer returns because the answer did not hold;
  • exceptions the AI routed to a queue, which someone now has to triage.

Pilot dashboards may show the share of conversations the AI participated in. That is not the share it finished. Before you value anything, inspect the partially resolved cases and the human time they consumed; early figures without a stable denominator and an audited outcome should not enter the business case.

Step 5: value the capacity honestly

Freed hours are capacity, not cash, until something changes.

  • Cash saving happens if you reduce agency or overtime spend, avoid planned hires, or avoid other spending that you can document.
  • Capacity gain happens if the same team handles growth, clears backlog, or takes over tasks from more expensive colleagues. That is valuable but it is not a cost line that disappears.

Use a fully loaded hourly labor cost: salary plus employer contributions, benefits and allocated overhead, divided by the productive work hours those costs cover. Apply it to the same kind of work and period as the time estimate.

Labor-cost equivalent = net minutes saved ÷ 60 × fully loaded hourly cost.

If we assume €40 per productive hour in the hypothetical above, 17,310 ÷ 60 × €40 = €11,540 a month of capacity at that cost. The hourly cost is an invented input, not a reported customer figure or Typewise price. This is before incremental AI software, setup and ongoing operating costs, and it becomes cash saved only if spending actually changes.

Use your actual labor costs and staffing arrangements, not a generic salary. Then name the intended use of the freed time: handling growth, reducing overtime, clearing a backlog or taking on other work. Check afterward whether that change happened. More capacity can be valuable even when payroll stays the same, but it is a different result from cash saved.

Step 6: put the costs on the same page

Show costs over the same period as the savings. Include licence or usage fees, integration, data-readiness work and evaluation effort. Also include your team's ongoing work reviewing drafts, maintaining instructions and knowledge, and running the system after the pilot. Count each cost once: the draft-review time already deducted in the example must not be charged again.

Choose and state a period, commonly twelve months. Show the net per month with writing-time, handling-time and whole-case savings still separate. The reader should be able to see which line the result depends on.

Pricing and costs differ by vendor and contract; use the actual figures you have been quoted rather than any example here.

Before presenting the case, give finance the inputs as well as the total: measured baseline, eligible volume, observed effort, residual human work, costs and the uncertainty on each. Keep assistance and autonomous handling on separate, non-overlapping sets of cases. Then agree how you will compare the forecast with actual results after the pilot.