Email Handle Time Measurement

Email Handle Time Measurement is the set of methods by which a workforce function measures handling time per item for email, cases and other asynchronous work, and the standard it adopts among them. The problem it answers is that no platform event marks an email the way a telephony platform marks a call: there is no answer, no release, no after-contact state, and so no measured handle time falls out of the system of record. Five methods are in use, each measures a different quantity under the one name, and a function that has not chosen among them is planning on a number whose instrument it cannot name. The page is part of the Workforce Data Definitions Framework. Average Handle Time keeps its definition on its own page and in The Definitions Register; this page is about the instrument for the asynchronous row of that definition, and the quantity it must produce is the one AHT-02 (RG-008) defines and Workload Model for Asynchronous Channels applies: handling time per item as the sum of touch durations, landing in the in-work tier of the Agent Time Accounting Model and sourced from ST-007.

Two things should be said about the evidence. The research behind this cluster found no peer-reviewed study that quantifies handle-time measurement error by method in contact centres, so the biases below are stated by direction and never by size. And the methods differ in standing: time-and-motion study and work sampling are established industrial-engineering methods with textbook treatments, while the desktop and inference methods, and the recommended standard itself, are the practitioner model, not yet validated in the literature.

What the instrument must measure

The target quantity is handling time per item: the sum of the bounded spans of agent attention the item received between open and close. Three quantities are routinely substituted for it and are not it. Elapsed time, open to close, includes queue, pend and idle time and is the customer's clock. State time, minutes in an activity code named for the channel, includes whatever the agent did while the code was set. Productive time divided by items closed is a ratio of two other measurements, not a measurement, and inherits both their errors. Each has a legitimate use; none is handling time.

The five methods

Method	What it captures	What it misses	Bias direction	When acceptable
1. Agent-toggled state timer — the agent sets a channel state on the telephony or workforce platform, which times it	minutes in the state per agent per interval; cheap, universal, already in the ledger at tier 4	which item; touch boundaries; work done in another state; the toggle the agent forgot	both directions: high when the state stays set through idle time (idle-in-state); low when email is worked in an available state (availability bleed); sensitive to incentives, since pace responds to what the state is used to judge	as the ledger's state input and a ceiling on handling time; as a per-item figure only when divided by items closed in-state and graded [E]
2. System-derived open-to-close — the work-item tool's status time stamps	elapsed time per item, measured [M] for what it is; the ageing profile	everything inside the span: queue, pend, interruptions, overnight; whether anyone was working	high, by a large and variable amount; batch closing stacks many closes at one instant and yields elapsed values either trivially short or absurdly long	for the time-to-close goal and the ageing profile, its proper use; never as handling time
3. Desktop activity capture — a capability class, named generically: software that records which item has focus and when input occurs	focus time per item; touch boundaries; interruptions and switches; items open at once	off-screen thinking, collaboration by phone or in person, work in an application it does not see; attribution when two items are open together	low, by the off-screen share; otherwise set by the attribution rule the function publishes	as the primary instrument, with the attribution rule published and a reconciliation to state time; inferred practice
4. Time-and-motion study and work sampling — an observer times work, or records the activity under way at random instants	a calibrated estimate by item type, including off-screen work; the mix of activities in the day	continuous measurement; the long tail; multitasking, which the time-motion literature handles inconsistently	observer effect; small-sample error; a study of one week describes that week	to calibrate the other instruments and to set a standard time per item type where no instrument exists; periodic, never continuous
5. Statistical inference — productive or state time regressed on items closed across agents and days	a marginal minutes-per-item from history with no instrument; a cross-check	item mix; multitasking; everything that lands in the intercept	inherits its inputs' biases; censored by work availability, since pace slows when work is scarce; confounded by availability bleed	when counts are reliable and mix is controlled; as an [E]-grade input where nothing else exists, labelled as such

The biases

Multitasking and interruption. Interrupted work is completed faster and at higher stress than uninterrupted work, so a handling time captured on a day of heavy voice interruption is not the same quantity as one captured on a quiet day. A practitioner source warns of lower output and higher error rates when email is mixed with calls; the finding on speed and stress is established, its application to measurement is inferred.

Pended items. An item waiting on a customer or third party sits open for hours with no touch. Method 2 counts the wait as handling; method 1 counts it only if the agent leaves the state set; methods 3 and 4 do not. The pend status is the fix.

Batch closing. Agents who work several items and close them together produce a cluster of close stamps at one instant. The count is right; the per-item elapsed is wrong in both directions; and any method that attributes time to the item open at the moment of close attributes it wrongly.

Idle-in-state. A state left set through a break or a huddle inflates method 1 and, through it, method 5. Its size is the part of state time minus focus time not explained by the sampled off-screen allowance once method 3 is running.

Availability bleed. The practitioner definition of availability is time the agent is available in the phone system, including auxiliary time, as a direct result of idle time, so an email worked while waiting for a call is booked as idle availability by the telephony platform and by nothing else. Method 1 cannot see it because no email state was set; method 5 is confounded by it because items closed rise while the email state does not. Production Accounting for Blended Agents carries the detection tests; the consequence here is that a handling time from method 1 or 5 on a blended team is biased low by an amount test T-06, the availability-bleed test, must bound before the number is used. These four mechanisms, pended items, batch closing, idle-in-state and availability bleed, are inferred; no peer-reviewed size estimate exists.

Incentive-driven pace and censoring. When handling time is estimated from throughput history, the history is censored by how much work was available and shaped by the agent's response to scarcity; this is the one bias with peer-reviewed evidence on pace estimation, and it argues for measuring the touch rather than inferring the pace.

Worked example

The numbers are synthetic and graded [E]; the results are [C] on [E] inputs. One agent, one shift, on an email queue with no voice work. Productive time on the channel from the ledger is recorded at 5.6 hours [E], of which 0.6 hours is a 20-minute huddle and a 16-minute break misbooked inside the email state, and therefore inside productive time; the misbooking surfaces in the gap the reconciliation rule below publishes. Items closed, from the work-item tool, 32 [E]. The state timer shows 5.6 hours in the email state [E]. Open-to-close elapsed time averages 3.4 hours per item [E], most of it pended. Desktop activity capture attributes 4.3 hours of focus time to the 32 items [E]. A work-sampling study on the team the previous week gave 9.2 minutes per item for this item type [E]. A regression of productive time on items closed across the team gives 0.157 hours per item [E].

Method	Per-item result	Against the focus instrument
1. State timer ÷ items	5.6 h ÷ 32 = 10.5 min [C]	high by 2.4 min: idle-in-state of 0.6 h (1.1 min) plus 1.3 min of off-screen residual (0.7 h) [C]
2. Open-to-close	204 min [C]	not a handling time; the ageing figure
3. Desktop focus time ÷ items	4.3 h ÷ 32 = 8.1 min [C]	the reference for the focus quantity; 1.3 h [C] of recorded productive time is unattributed: 0.6 h idle-in-state and 0.7 h off-screen
4. Work sampling	9.2 min [E]	high by 1.1 min; includes off-screen work the desktop instrument cannot see
5. Regression	9.4 min [C]	high by 1.3 min; the intercept absorbs idle-in-state and off-screen work indistinguishably

Four of the five bracket the value between 8.1 and 10.5 minutes; the fifth is a different quantity. The focus instrument's 8.1 minutes is measured [M] for the focus quantity under the published attribution rule. The reconciliation Σ focus ≤ state ≤ productive holds (4.3 ≤ 5.6 ≤ 5.6) and publishes the 1.3-hour gap as 0.6 hours idle-in-state, which is where the misbooked huddle and break surface, and 0.7 hours off-screen residual. AHT-02 for the item type is then 8.1 minutes plus a calibrated off-screen uplift from the sampling study, 9.2 − 8.1 = 1.1 minutes, graded [E] because the uplift is sampled, not captured. The inputs are synthetic and one shift is not a standard; the standard is set on the calibration cadence below. Service times in a large published call-centre dataset are lognormal rather than symmetric, and the transfer of that shape to email is inferred, so the per-item distribution is kept, not only its mean.

The recommended standard

The standard is transaction-level capture with a focus-time instrument, reconciled to the ledger. It is the practitioner model, adopted because each part is checkable, not because the literature has validated the whole.

1. The unit is the item. The count is stamped at close in the work-item tool and graded [M]. Batch closes are detected by the clustering of close stamps and reported.
2. The time is focus time per item from desktop activity capture, with a published attribution rule for touch boundaries and for minutes when two items are open together. Graded [M] under the rule, which is part of the definition card.
3. Reconciliation runs every extract. Σ focus time on the channel ≤ state time in the channel's code ≤ productive time, per agent per day. The two gaps are published as idle-in-state and off-screen residual with a tolerance SET LOCALLY; a breach fails test T-05, item count against time, on Reconciliation Tests for Time Accounting.
4. Calibration is periodic. A work-sampling study per item type at a cadence SET LOCALLY bounds the off-screen gap and supplies the standard time where the instrument is absent.
5. Grades travel. AHT-02 (RG-008) cites the instrument and the attribution rule in its Source of Truth Register entry ST-007, whose v0 is a quarterly sample and whose v1 is the work-item timer, and a model that reads the number inherits the grade (Human Gates and Number Grades).

When only partial capture exists

Most estates start below the standard, and the ladder is explicit so the grade is honest at each rung. With a count and a state timer but no focus instrument, handling time is state time divided by items closed in-state, graded [E], and on a blended team test T-06 bounds the understatement first. With a count only, handling time is the work-sampling standard time for the item type, graded [E]. With no reliable count, the channel is marked not inventoried in the ledger and planned on an asserted item estimate graded [A], and the capacity plan says so. At no rung is elapsed time used as handling time, and at no rung is a number promoted a grade because a better instrument is on order.

Maturity Model Position

Four scales on this wiki use the word level; the launch page states which is which. On the WFM Labs Maturity Model™ a function at Level 2 has a count per item and a state timer, states the grade [E] on the handling time it derives, and can name its instrument. Level 3 is the standard above: a focus-time instrument, a published attribution rule, reconciliation on every extract and a calibration cadence, which is what lets the automation layer prioritize work on measured rather than assumed effort. At Level 4 the per-item distribution is an input to the simulation that sizes the channel, and the residuals are small enough that the capacity model and the ledger agree without a monthly explanation.

See Also

• Workforce Data Definitions Framework — the hub this page belongs to
• Workload Model for Asynchronous Channels — the quantity the instrument must produce
• Agent Time Accounting Model — the in-work tier the measurement lands in
• Agent State Taxonomy and Activity Codes — the state the toggle timer sets
• Production Accounting for Blended Agents — the bleed tests for blended teams
• Reconciliation Tests for Time Accounting — tests T-05 and T-06
• Source of Truth Register — ST-007, where the instrument and rule are cited
• Average Handle Time — the definition this page instruments
• Chat Concurrency and Handle Time — the sister problem for concurrent sessions
• Human Gates and Number Grades — the grades that travel with the number