Wiki:HTML Packs/Channel Launch Sizer

An HTML pack is a deployable analytical tool published through this wiki: a single-file HTML tool, the data contract it reads, and an assistant kit for a Claude project. See Wiki:HTML Packs.

The Channel Launch Sizer estimates what opening a new assisted contact channel, such as live chat, messaging or SMS, does to human workload and to the no-touch rate, the share of transactions completed without any human handling. It compares a human-only launch with the same launch backed by automation, gives the break-even points between them, and shows the range of outcomes the inputs allow. It implements the model on Channel Proliferation and the No-Touch Rate, and its Evidence tab sets out the research behind every input.

When to use it

Use it before a channel launch, when the decision is how to launch rather than whether: staffed only, or held until intent routing or automation is ready. It suits a launch forced by customer, client or commercial demand where the automation will not be ready on day one, and a review of a recent launch in which human-handled volume rose rather than fell.

Do not use it to size automation added to an existing channel with no new entry point; AI Containment Rate and Its Workforce Implications and Interior Optimum (containment rate) cover that case. Do not use it for interval staffing. The sizer gives annual and monthly workload, which then feeds an Erlang or simulation model at interval level.

What it shows

• Sizer: the scenario inputs, each with an evidence grade, and where the new volume comes from, as an editable list of sources. The headline numbers are added human-handled contacts, the change in no-touch rate, net FTE and net cost for a human-only launch. A comparison table sets these against the launch with automation. Four break-even thresholds follow: the automation resolution that restores the pre-launch no-touch rate, the automation resolution for zero net workload, the concurrency for a workload-neutral human-only launch, and the share of volume that would have to be diverted. Two monthly charts show the ramp and the step when automation goes live.
• Range: a seeded Monte Carlo over every ranged input. It gives P10, P50 and P90 for net FTE, the value of launching with automation and the change in no-touch rate, plus the probability that a human-only launch adds workload. It also shows a histogram, and a grid of net FTE by volume mix and concurrency against the wiki's substitution-led and additive bracketing cases.
• Evidence: the research base for each lever, with each source's quality (peer-reviewed, regulatory filing, analyst, trade press, or gap) and what the evidence means for planning.
• Data: load a reference case or the synthetic sample, import and export the shape file, export a Markdown note, copy a share link, and read the validator findings.

Every headline number carries the weakest grade among its inputs: [M] measured, [C] computed, [E] estimated or [A] asserted. In the reference case the source shares are graded [A]. No published study measures the diverted versus net-new split for an assisted launch, so the tool shows that as the main uncertainty rather than hiding it.

The contract

The tool reads and writes one shape-file envelope with schemaId wfm.shape.channel-launch.

payload.launch_sources has one row per source of new-channel volume:

Column	Meaning
source	Name of the source, code-named (for example "Voice", "Email", "Would-be self-service")
kind	diverted: moved from an existing assisted channel, relieving its work. notouch: touches a transaction that would otherwise have had no human handling. repeat: the same issue also raised in another assisted channel. latent: a newly asked question about an already-touched transaction
share, share_low, share_high	Most likely, low and high share of new-channel volume, 0–1. The share column sums to 1
aht_sec	Handle time of the work a diverted row relieves, in seconds

Assumptions are held under launch.*, each with a value, a range of low, most likely and high, a grade and a status. They are: transactions per year, the pre-launch no-touch rate, contacts per transaction, new-channel volume as a share of contacts, new-channel handle time, effective concurrency, paid hours, shrinkage, occupancy, cost per FTE, automation true resolution, ramp midpoint and rate, automation go-live month and horizon.

Declared conventions. Handle times are on a worked-time basis, and concurrency is effective concurrency, not the configured maximum. Automation resolution means issues actually resolved, not contacts contained. No column scales: a shape-only export leaves the file unchanged, because the operation's size lives in launch.transactions.

Validator rules.

• Fail (the tool refuses the scenario): shares do not sum to 1; a share is outside 0–1; a diverted row has no handle time; a fraction is outside 0–1; or a volume, handle time or concurrency is not positive.
• Warn: a range does not contain its most-likely share, or concurrency is above 4.
• Info: there is no notouch row.

How to deploy

1. Open channel-launch.wfmlabs.com, or save the page as a single HTML file and open it locally. It runs entirely in the browser and makes no network calls.
2. Start from Reference case (wiki) or Load sample (Estate-1) on the Data tab, then replace the inputs with your own, and regrade each one as you replace it.
3. Enter the sources of new volume. Where email or another channel is being withdrawn, give it its own diverted row with its own handle time.
4. Read the Range tab before quoting a number. If the P10 to P90 range crosses zero, the volume mix decides the answer, and the mix is the thing to measure.
5. Optional: create a Claude project, paste Block 1 into its instructions and upload Blocks 2 and 3 as knowledge. The assistant then interviews the planner and writes the shape file.

Block 1 — Project instructions

# Channel launch sizing — scenario intake

[ROLE]
You help a workforce planner build a scenario file for the Channel Launch
Sizer: what opening a new assisted channel (chat, messaging, SMS) does to
human workload and the no-touch rate, human-only vs with automation.

[RULES]
1. Never ask for or record client, employer, vendor or person names. Use
   code names (Estate-1, Client-A, Vendor-1). Say so if a name is offered.
2. Ask first: what decision will this change? Do not continue without one.
3. Interview one block at a time: the operation (transactions, no-touch
   rate, contacts per transaction); the new channel (volume share, handle
   time, effective concurrency); capacity and cost; automation (TRUE
   resolution, go-live month); then where the new volume comes from.
4. For every number, ask for low / most likely / high and its source.
   Grade it: M measured from a system of record, C computed from measured
   inputs, E estimated, A asserted. When the planner has no number, offer
   the reference value from evidence-and-defaults.md, grade it A, and say so.
5. Sources of new volume: split them into the four kinds in
   channel-launch-contract.md. Push back on any mix with more than 70%
   diverted unless it is measured; no published study supports it.
   Where a channel (often email) is being withdrawn, give it its own
   diverted row with its own handle time.
6. Handle time is on a worked-time basis; concurrency is effective, not
   configured. Automation resolution is resolution, not containment.
7. Check before writing: shares sum to 1; each diverted row has aht_sec;
   fractions are 0–1.

[OUTPUT]
End with the shape file as one JSON code block following
channel-launch-contract.md exactly (schemaId wfm.shape.channel-launch,
provenance.generator "claude-desktop", provenance.codeNames true), then a
five-line brief: the decision, the three weakest inputs, and what to
measure first. Tell the planner to load the JSON on the tool's Data tab.

Block 2 — channel-launch-contract.md

# Contract: wfm.shape.channel-launch (1.0)

Envelope keys: $schema "https://wfmlabs.com/schemas/shape/1.json",
schemaId "wfm.shape.channel-launch", schemaVersion "1.0.0",
scale "scaled", title, created (ISO), note, provenance, basis,
assumptions, grades, payload, questions. Unknown keys are ignored;
missing keys take defaults.

## payload.launch_sources (one row per source of new-channel volume)
- source: string, code-named
- kind: diverted | notouch | repeat | latent
  - diverted: moved from an existing assisted channel; relieves its work
  - notouch: touches a transaction that would otherwise be no-touch
  - repeat: same issue also raised in another assisted channel
  - latent: newly asked question about an already-touched transaction
- share, share_low, share_high: 0–1; share sums to 1 across rows
- aht_sec: seconds; required (>0) for diverted rows, 0 otherwise

## assumptions (each: {value, range:[low,likely,high], grade, status, note})
launch.transactions            per year
launch.no_touch_rate           0–1, before launch
launch.contacts_per_txn        assisted contacts per transaction
launch.new_volume_pct          0–1 of baseline assisted contacts
launch.aht_sec                 new-channel handle time, seconds, worked
launch.concurrency             effective sessions per agent
launch.paid_hours              per FTE per year
launch.shrinkage               0–1
launch.occupancy               0–1
launch.cost_per_fte            per year
launch.auto_resolution         0–1 TRUE resolution
launch.ramp_mid_month          month at half of steady-state volume
launch.ramp_rate               logistic rate per month
launch.auto_start_month        month automation goes live
launch.horizon_months          months shown

grades: "launch_sources.share" and "launch_sources.aht_sec" (M/C/E/A).
status: default | estimated | confirmed.

Block 3 — evidence-and-defaults.md

# Evidence and reference values

Reference case (wiki worked example, synthetic): 1,000,000 transactions;
no-touch 60%; 0.40 contacts per transaction; new channel 20% of contacts;
chat 720 s, voice 540 s; concurrency 1.5 (range 1–3); shrinkage 30%;
occupancy 80%; automation true resolution 30% (range 10–50%).

Reference mixes (share of new-channel volume):
- Substitution-led: diverted 70 / notouch 5 / repeat 10 / latent 15
- Mixed:            diverted 45 / notouch 10 / repeat 20 / latent 25
- Additive:         diverted 25 / notouch 20 / repeat 25 / latent 30

What the research supports:
- New channels add volume at least as often as they replace it. A new web
  portal raised calls 14% overall (Kumar & Telang 2012, ISR). Customers
  revert to familiar assisted channels (Gartner 2021).
- Self-service fully resolves 14% of issues (Gartner 2024), so would-be
  self-service is a real source of new assisted volume.
- An assisted chat contact is cheaper than a call only above break-even
  concurrency = chat AHT / relieved handle time per chat contact.
- Agentic AI resolving 80% of common issues by 2029 is a forecast (Gartner
  2025), not a measurement. Vendor "resolution" often means non-return.
- No study measures the diverted vs net-new split for an assisted launch.
  Grade the shares A until measured by a staggered rollout.

Full sources: wiki page "Channel Proliferation and the No-Touch Rate".

Usage notes

• Confidentiality. The tool makes no network calls and stores nothing. A share link carries the whole scenario in the part of the URL after the #, which browsers do not send to a server, so a link is as sensitive as the file itself. Use code names in titles, notes and source names.
• Transport. The tool is one HTML file. It can be saved and opened locally inside a data boundary, so real figures never leave the machine they are entered on (see The Shape File Bridge).
• Reading the result. Quote the range, not the point estimate, while the source shares are graded [A]. Re-run after the first weeks of a launch, once the contacts per transaction across all channels can be compared before and after.
• Versioning. The version is in the tool footer. A newer version is a new file, and this page's change history says what changed.

Change history

Version	Date	Change
1.0.0	2026-10-08	First release. Sizer, Range, Evidence and Data tabs; seeded Monte Carlo; contract wfm.shape.channel-launch 1.0.

See also

• Channel Proliferation and the No-Touch Rate, the model and the evidence
• Contact Deflection and Channel Shift Modeling and Service Demand Rebound Model
• The Shape File Bridge, the pattern HTML packs implement
• Wiki:HTML Packs, the register