Customer service · Decision guide
Compare agent hours, review effort, per-resolution fees and setup work for human-first support, an AI agent that hands off to people, and a self-service knowledge base — with the automation share as a range you set.
Every path against today — select one to see exactly what changes
The result depends almost entirely on the share of conversations an AI agent resolves for your customers, which no vendor page can tell you. Enter it as a range; if the comparison flips inside that range, a one-queue pilot is the honest next step.
Who this is for Support leads and founders whose team answers hundreds or thousands of customer conversations a month.
Optional. Answers decide which paths meet your hard constraints; "not sure" keeps a requirement unresolved instead of guessing.
Your current approach, the alternatives and a smaller reversible pilot — all with the same inputs and horizon.
Current approach
People answer every conversation, with saved replies for common questions.
Main uncertainty: Handling time as volume and complexity change.
Alternative
An AI agent answers from your help content and hands conversations to people when it cannot resolve them or the customer asks.
Main uncertainty: The automation share and hand-off rate for your customers — the inputs that decide this comparison.
Alternative
Write and maintain help articles so customers find answers before contacting you.
Main uncertainty: How many conversations articles really prevent.
Smaller reversible pilot
Turn the AI agent on for one queue of repeat questions for eight weeks, with a rollback plan.
Main uncertainty: How well one queue predicts the others.
Person-hours are everyone's effort added up, not calendar time. Valued staff time is hours × the value of an hour: an economic cost, not a bill.
| Path | Setup effort | Ongoing effort per week | Setup cash | Ongoing cash per month | Cash over the horizon | Incl. valued staff time | Weeks to go live |
|---|---|---|---|---|---|---|---|
| Human-first support | 4 h | 47.03 h | $0 | $0 | $0 | $44,583 | 1 |
| AI triage with escalation to people | 36 h | 38.78 h | $0 | $249 | $1,247 | $40,168 | 4 |
| Self-service knowledge base | 63 h | 43.43 h | $0 | $0 | $0 | $43,790 | 6 |
| Pilot: AI agent on one queue | 15 h | 43.56 h | $0 | $67 | $399 | $42,264 | 11 |
Horizon: 6 months. Change the assumptions, compare any path with your current approach, combine steps and export a plan with the interactive tool (JavaScript).
Support · every week from week 1 · 46.03 h / week
Support · weeks 1–1 · 4 h
Support · every week from week 1 · 1 h / week
Gains: A person on every conversation; Full context and judgment; No AI provider dependency
Sacrifices: Hours grow in line with volume; Coverage limited to staffed hours
Must be true: Staffing can grow with volume
Support · weeks 1–3 · 20 h
Support · weeks 2–2 · 4 h
Support · weeks 4–4 · 12 h
Support · every week from week 1 · 46.03 h / week
Support · every week from week 5 · 36.36 h / week
Support · every week from week 5 · 0.41 h / week
Support · every week from week 5 · 2 h / week
Vendor or partner (cash only) · every week from week 5 · $249 / month
Gains: Fewer conversations per person; Answers outside staffed hours; Help content gets better
Sacrifices: Per-resolution fees; Review and content upkeep; Wrong answers reach some customers
Must be true: Automation share stays inside the range you entered; Customers can always reach a person
Support · weeks 1–6 · 60 h
Support · weeks 6–6 · 3 h
Support · every week from week 1 · 46.03 h / week
Support · every week from week 7 · 41.43 h / week
Support · every week from week 7 · 2 h / week
Gains: No per-conversation fee; Content useful for customers and staff; Customers can help themselves anytime
Sacrifices: Writing effort up front; Content goes stale without upkeep; Deflection is hard to measure
Must be true: Customers find the articles where they look for help
Support · weeks 1–1 · 5 h
Support · weeks 1–1 · 2 h
Support · weeks 2–2 · 6 h
Support · every week from week 1 · 46.03 h / week
Support · every week from week 3 · 43.45 h / week
Support · every week from week 3 · 0.11 h / week
Vendor or partner (cash only) · every week from week 3 · $67 / month
Support · weeks 11–11 · 2 h
Gains: Your own automation numbers; Limited exposure; Content for one queue gets fixed
Sacrifices: Setup work for a small share of volume; One queue may not represent the rest
Must be true: The pilot queue has enough volume to read results
The workload model follows each conversation:
Automation share and hand-off rate are your assumptions. Vendor documentation describes how a product hands off to people; it is not a measurement of your queue. If you do not know your handling time, leave it unknown: the tool will show the workload as unresolved rather than invent a number.
Service quality, customer satisfaction and the effect on trust are not modelled. They belong in the pilot's measures and in your stop rules.
If most conversations are complaints about the same product problem, fix the product first: no support model removes the cause. If conversations involve safety, health, legal or financial advice, keep a person responsible for every answer.
Specific, dated practices — not whole-company models. Paths marked as an adaptation are WeCaseStudy's adaptation for your context, not a claim that the company used this exact workflow.
Calculations: formula family support, version 1. Decision version 1. How the numbers are made
