Channel Proliferation and the No-Touch Rate

Channel proliferation is the addition of new customer contact channels — live chat, in-app messaging, SMS, social messaging — to an operation that already serves customers by voice, email and self-service. The no-touch rate is the share of transactions completed without any human handling. When a new channel is opened as a staffed, assisted-only channel, with no automation behind it, it does not itself produce any no-touch transactions: every contact it receives is a human touch. Whether it lowers cost therefore depends on how much of its volume is diverted from more expensive channels and how much is new demand that did not previously reach an agent. The research base on that split is thin. The evidence that does exist, mostly from self-service channels and customer surveys, suggests that new channels add volume at least as often as they replace it. By construction, an assisted channel cannot raise the no-touch rate until automation is added.

This page summarises the evidence, separates the components of new-channel volume, and sets out a parameter model for sizing the effect before launch. It complements Contact Deflection and Channel Shift Modeling, which covers deflection initiatives in general, and the Service Demand Rebound Model, which covers induced demand after automation. This page is limited to assisted launches and their effect on the no-touch rate. General deflection and shift forecasting is covered on the deflection page.

Why an assisted-only channel cannot raise the no-touch rate

The no-touch rate is defined over transactions, not contacts: N / T, where T is transactions in the period and N those completed with no human involvement. An assisted chat contact is, by definition, a touch. Opening such a channel can therefore leave N unchanged (if its volume comes from voice or email) or reduce it (if it attracts customers who would otherwise have completed the transaction in self-service). It cannot increase it. Any no-touch gain attributed to an assisted channel launch is either a measurement artefact — for example, counting a chat contact as "digital" and therefore touchless — or the effect of automation introduced alongside it.

The distinction matters because "digital share" and "no-touch share" measure different things and are easily conflated. A booking made through an online tool is digital and usually no-touch; a chat conversation with an agent is digital and fully touched. An operation that reports a single digital-share figure can therefore show progress while human-handled volume grows.

Evidence: substitution or addition?

Self-service channels

One of the few peer-reviewed studies of channel addition in a service setting is a field study at the call center of a large US health insurer. Use of a newly introduced web self-service portal increased telephone calls by 14% overall. Calls for unambiguous, easily retrieved information fell by 29%, but calls about ambiguous information rose substantially, as customers who had seen partial information on the portal called to interpret it. The portal was a self-service channel, not an assisted one, so the result is an analogue rather than a direct measurement of chat. It shows the mechanism that matters: a new channel raises the customer's awareness of an issue and lowers the effort of raising it, and some of that converts into contacts in existing channels.

Survey evidence from the Corporate Executive Board, later part of Gartner, found that 57% of roughly 75,000 customers surveyed had switched from the web to the phone to get an answer. Gartner's 2019 analysis of customer service leaders concluded that "while customers are using newly added self-service channels, they haven't stopped using more expensive live channels," and reported that 56% of leaders were then adding channels or features.

Self-service resolution rates explain much of this. In a 2023 survey of 5,728 customers, Gartner found that only 14% of service issues were fully resolved in self-service, and only 36% even for issues customers described as "very simple." A failed self-service attempt is not a deflected contact. It is a delayed one.

Multi-channel journeys

Gartner identified as a myth in 2021 the belief that "customers will readily adopt digital channels if provided and promoted"; in practice "customers often revert to assisted channels used in the past, leading to avoidance and abandonment of self-service." A 2023 survey found 62% of channel transitions to be high effort. Each additional channel adds another route by which a single issue can produce more than one contact. That cross-channel repeat volume is invisible to channel-level reporting, because each channel sees only its own contacts.

What is not known

No published study was identified that measures, for an assisted chat or messaging launch, the share of the new channel's volume that is diverted from voice versus net-new. Widely repeated figures such as "chat reduces call volume by 10–30%" appear in vendor and practitioner blogs without a stated method or source, and are not suitable planning inputs. The incremental share is the single most important parameter in sizing a channel launch, and it currently has to be estimated from an operation's own before-and-after data on contacts per customer or per transaction (see Contact Rate).

Components of new-channel volume

The volume C that a new assisted channel receives at steady state can be decomposed into four sources:

Component	Symbol	Description	Effect on human workload	Effect on no-touch rate
Diverted from assisted channels	sv	Customers who would otherwise have phoned or emailed	Replaces existing work; net effect depends on relative handle time	None
Touch on a no-touch transaction	sn	Contacts about a transaction that would otherwise have been completed without help: customers who would have self-served but find an agent easier to reach, and newly asked questions about otherwise untouched transactions	Adds work	Reduces
Cross-channel repeat	sr	The same issue also raised, or later raised, in another assisted channel	Adds work	None
Latent demand	sl	Newly asked questions about transactions that were already being touched, previously unasked because the existing channels involved too much effort	Adds work	None

The shares sum to one. Only s_v substitutes for existing work. The other three are additive, the same mechanism the Service Demand Rebound Model describes for automation. The split between s_n and s_l follows the transaction a contact attaches to, not the customer's motive, because that is what determines the effect on the no-touch rate.

Sizing model

The model below sizes the effect of a launch before it happens. All inputs are planning assumptions to be replaced with an operation's own data.

Inputs. T transactions per period; current no-touch rate NT₀; baseline voice contacts V; steady-state new-channel volume C; component shares s_v, s_n, s_r, s_l; voice handle time AHT_v; chat handle time AHT_c; effective chat concurrency m.

Outputs.

• Change in human-handled contacts: ΔH = C × (s_n + s_r + s_l)
• Change in no-touch transactions: ΔN ≈ −C × s_n, so the new no-touch rate is (NT₀T + ΔN) / T
• Change in agent workload (minutes): ΔW = C × AHT_c / m − C × s_v × AHT_v. This assumes diverted volume comes from voice. Where some comes from email, the email share is credited at email handle time instead.

Ramp. New-channel volume can be represented by a logistic (S-shaped) adoption curve, the standard form for technology diffusion, C(t) = C_∞ / (1 + e^{−g(t − t₅₀)}). No published source gives benchmark values for the growth rate g or the midpoint t₅₀ for service channels, so both should be fitted to the operation's own early weeks. The four components need not ramp together. Gartner's finding that customers revert to familiar assisted channels suggests that diversion from voice builds more slowly than latent demand, which arrives as soon as the channel becomes visible. If so, the first months after launch will understate the eventual voice reduction and overstate the incremental share.

Worked example

The figures below are illustrative assumptions, not benchmarks. Consider an operation with 1,000,000 transactions a year, a no-touch rate of 60%, and 400,000 voice contacts (one per touched transaction). An assisted chat channel is launched and reaches 80,000 contacts a year, 20% of the voice baseline. Voice handle time is 9 minutes, chat handle time 12 minutes, and effective concurrency 1.5.

Scenario	sv / sn / sr / sl	Added human contacts	No-touch rate	Net workload change
Substitution-led	0.70 / 0.05 / 0.10 / 0.15	+24,000 (+6%)	60.0% → 59.6%	+136,000 min (+3.8%)
Mixed	0.45 / 0.10 / 0.20 / 0.25	+44,000 (+11%)	60.0% → 59.2%	+316,000 min (+8.8%)
Additive	0.25 / 0.20 / 0.25 / 0.30	+60,000 (+15%)	60.0% → 58.4%	+460,000 min (+12.8%)

Workload change is measured against a baseline of 3.6 million voice minutes. Three features of the example follow from the model's structure rather than from the particular inputs:

1. Workload rises unless diversion is very high. At these handle times a chat contact costs 8 agent-minutes (12 ÷ 1.5) against 9 for voice. Workload falls only if s_v exceeds 8 ÷ 9, or 0.89. In general the condition is s_v > AHT_c / (m × AHT_v).
2. The no-touch rate falls, though slowly. Its decline is driven only by s_n. Where several assisted channels are launched, each contributes its own s_n, so the effects are additive across launches.
3. Concurrency is the main operational lever. The break-even concurrency at which a chat contact costs the same agent time as a call is AHT_c / AHT_v, which is 1.33 here. ContactBabel's UK data on whether chat is cheaper than voice is inconsistent between editions, with some showing chat at about half the cost of a call and others at parity. That inconsistency fits the model: the cost advantage depends almost entirely on the concurrency achieved.

The Kumar and Telang result of a 14% rise in calls after a self-service launch measures a different quantity, phone calls after a self-service launch rather than total human contacts after an assisted one. It is not a calibration point for the table. It does show that net additions of this order are plausible in practice.

The automation overlay

The no-touch rate begins to move only when the new channel gains the ability to resolve contacts without an agent. If a fraction c of channel volume is resolved by automation (true resolution, not merely containment), the change in no-touch transactions becomes ΔN ≈ C × [c × s_v − (1 − c) × s_n]. Resolving diverted contacts converts previously touched transactions to no-touch, while unresolved s_n contacts still erode the rate. ΔN turns positive once c exceeds s_n / (s_v + s_n). In the worked example that threshold is 0.07, 0.18 and 0.44 for the three scenarios. Evidence on achievable c varies by capability tier:

Capability tier	No-touch contribution	Evidence quality
Assisted only (live agent)	Zero by definition	—
Self-service generally (not bot-specific)	Customer-reported full resolution of 14% across all self-service channels; no independent bot-specific benchmark identified	Analyst survey
Generative AI assistant	No independent benchmark identified. Published figures are vendor-reported, vary widely, and often define "resolution" as the customer not returning rather than confirmed resolution	Vendor only; not suitable as a planning input
Agentic AI executing transactions	Forecast: 80% of common issues resolved autonomously by 2029	Analyst forecast, not measurement

Generative AI has independently measured effects on assisted work. A study of 5,172 support agents found a 14% increase in issues resolved per hour, and 34% for novice agents. This lowers AHT_c, and with it the workload penalty of an assisted-only launch, but it does not create no-touch transactions. Gartner also reported in late 2025 that only 20% of customer service leaders had achieved AI-driven headcount reduction, a reminder that forecast containment and realised savings differ. The relationship between containment and staffing is non-linear; see AI Containment Rate and Its Workforce Implications and Interior Optimum (containment rate).

The practical consequence for sequencing is that an assisted channel launched ahead of its automation spends its first period as a pure cost and demand-creation event. When automation is added later, its resolution rate must first exceed the threshold above before the no-touch rate recovers to its pre-launch level. Separately, the s_r and s_l volume the channel created continues to add workload until it is automated or designed out.

Travel industry illustration

Travel management and online travel show a pronounced asymmetry between booking and servicing. Booking automation is mature. In a 2019 survey of 202 corporate travel managers, 92% had an online booking tool, but only 59% reported adoption of 70% or more, despite 81% having a booking mandate. Only 11% offered chat or instant messaging at the time, and 35% planned to add it. A digital-native travel management company reported in its 2025 registration statement that 90% of bookings were made online or on mobile.

Servicing is less automated and less often disclosed. No public travel management company reports a servicing automation rate for changes, cancellations and refunds comparable to its booking automation rate. The same filing reported that its virtual agent handled about half of user interactions without a live agent, with customer satisfaction of 78% for virtual-agent interactions against 96% overall. It also stated as a risk factor that "when large numbers of our customers experience delays or cancellations, our support costs tend to increase." At the scale of an online travel agency, one operator reported more than 250 million service interactions a year, over half of them self-served. Another reported customer service cost per reservation falling about 10% in a year when bookings grew about 10%. Total service cost was therefore roughly flat. The source does not report contact volume, so whether this reflects fewer contacts per booking or lower cost per contact is not established.

No public source reports contacts per trip, or the net-new contact effect of a messaging launch, for any travel company. The airline and travel messaging case studies identified for this page were vendor-authored.

Planning implications

• Forecast the new channel as demand, not as a transfer. Model C with an explicit incremental share, rather than subtracting it one-for-one from voice. See Demand Forecasting for Digital and Async Channels.
• Track contacts per transaction across all channels. Channel-level volume reports hide cross-channel repeats. The rate of contacts per transaction or per customer is the only metric that shows whether a launch added or moved demand.
• Report no-touch and digital share separately. An assisted chat contact should never count toward a touchless metric.
• Set a concurrency floor. The economics of an assisted-only channel depend on sustained concurrency above AHT_c / AHT_v. Very high concurrency carries its own costs; see Task Switching Costs in Multichannel Operations.
• Instrument the launch as an experiment. Staggered rollout by client, region or customer segment lets an operation estimate its own s_v instead of borrowing an unsupported industry figure.
• Contain the channel's entry points. Placing chat entry points after self-service attempts rather than before them, and routing chat to specific intents, reduces s_n, the only component that directly erodes the no-touch rate.

Interactive tool

The model on this page is implemented in the Channel Launch Sizer, published as the HTML pack HP-WFM-003. It runs in the browser with no network calls. It holds the four-component decomposition, extended to any number of diverted channels, each with its own handle time. It also holds the worked example as a reference case, the four break-even thresholds, and a seeded Monte Carlo over ranged inputs that reports P10 to P90 outcomes. The tool's Evidence tab restates the research above against each input it informs. Defaults are the synthetic worked example; real figures entered into the tool stay on the machine where they are entered.

Maturity Model Position

At Level 2, channels are typically added in response to customer or client demand, and each is forecast in isolation. The incremental-demand effect shows up as unexplained forecast error. At Level 3, forecasting incorporates contact rates and cross-channel volume, so the launch effect can be measured and the component shares estimated. At Level 4, launches are staged as controlled experiments and sequenced with automation capability, so assisted channels open with at least intent routing and a defined containment roadmap. At Level 5, channel and automation decisions are planned together in a unified human-and-AI capacity model, and the no-touch rate is a planned output rather than a residual.

See Also

• Wiki:HTML Packs/Channel Launch Sizer — the interactive tool (HP-WFM-003)
• Contact Deflection and Channel Shift Modeling
• Service Demand Rebound Model
• Contact Rate
• Digital Messaging
• Multichannel Workforce Management
• AI Containment Rate and Its Workforce Implications
• Interior Optimum (containment rate)
• Cost per Contact
• Hospitality and Travel Workforce Management