LensReading which lens this session carries.

Experience telemetry

What the throughput numbers cost the people on the other side of them

18 measures across 17 audiences, split into the people who work inside the estate and the people the estate serves. They sit here rather than inside a function dashboard for one reason: cycle time and unit cost are easy to improve by making somebody else absorb the difficulty, and this is the page where that shows up.

7 moved the wrong way

Health

Direction of travel, not level

A measure is red because it is worse than its own baseline, regardless of whether the absolute number still looks acceptable.

Moved the wrong way

7 of 18

3 employee · 4 customer. Every one of these sits behind a throughput gain reported elsewhere in the product.

Improved

10 of 18

1 flat. Improvement here is not the same as improvement in throughput and is not counted twice.

Asked a person

9 of 18

Collected by survey rather than inferred from system events. Total sample 1,254,541 responses.

Carrying a stated gap

5

Collection window Jul 21, 2026 to Aug 17, 2026.

Operations

Employee, then customer

Each row shows the current reading, the baseline it is measured against, how it was collected and how many people it is based on. A small sample is not hidden.

People who work inside the estate

9 measures · 0 degraded · 215,489 responses

MeasureAudienceBaselineNowMoveSampleMethodCollected
Reached a person on the first attempt, when they asked for oneDegraded

The clearest case of the trade the estate is making. Touchless handling went up 29 points and the share of people who got a human when they asked for one went down 12.8. Both numbers are real and they point in opposite directions.

Survey of 2,410 employees against a population of roughly 41,000. Response skews towards people who had a problem.

Employees raising an HR case71.258.4-12.82,410surveyAug 11, 2026
Satisfaction with the case outcomeDegraded

Cases close nearly five times faster and satisfaction fell two tenths of a point. Small, consistent across four survey waves, and not noise.

Employees whose HR case was closed4.34.1-0.23,180surveyAug 11, 2026
Said the reason for the outcome was explained clearlyFlat

Unchanged. Two thirds is not a good number and it has not moved in a year, which suggests explanation quality is nobody’s objective.

Employees who received an automated outcome66.066.00.02,410surveyAug 11, 2026
Time to a usable answerImproved

Measured from the platform, not asked. Session start to the first answer the user did not immediately rephrase.

Counts an answer as usable when the user stops asking. Some of those users gave up.

Employees using self-service8.73.2-5.5184,000telemetryAug 17, 2026
Confidence in the decision without re-checking the sourceImproved

Rising, slowly. The gap between a trust score of 87.4 and a manager confidence of 3.4 out of 5 is the distance between what the system believes about itself and what its supervisors believe about it.

Managers reviewing agent-produced decisions3.13.4+0.3640surveyJul 28, 2026
Automated outcomes sent back for rework after a challengeDegraded

More work is automated and a larger share of it comes back. In absolute terms rework rose from about 470 to about 1,010 items a week.

Employees who challenged an automated outcome6.28.9+2.711,400telemetryAug 17, 2026
Satisfaction with the resolutionImproved

The counter-example. In IT, speed and satisfaction moved together. Password, access and provisioning work is where nobody wanted a conversation in the first place.

Employees whose IT incident was resolved without a person4.04.4+0.48,920surveyAug 11, 2026
Described the automated close steps as understood, not opaqueImproved

Improved after the station-level evidence trail was published. Transparency moved the number more than accuracy did.

Finance staff on the monthly close58.072.0+14.0310surveyAug 4, 2026
Hours a week spent handling exceptions, per personImproved

Time on exceptions fell per person while the number of people fell too. Total exception hours across the estate fell from about 41,500 to about 13,500 a week.

Derived from work-unit timestamps rather than from time recording. Time a person spends thinking about an exception away from the system is invisible.

Operations staff across all towers11.4 h6.1 h-5.32,219telemetryAug 17, 2026

People the estate serves

9 measures · 0 degraded · 1,039,052 responses

MeasureAudienceBaselineNowMoveSampleMethodCollected
Resolved without needing a second contactDegraded

The second clear inversion. Disputes close three times faster and one in nine more customers has to come back a second time to get there. Faster is not the same as finished.

Customers raising a billing dispute74.863.1-11.74,780telemetryAug 17, 2026
Time to order acknowledgementImproved

Every order in the week, not a sample. The largest single experience improvement in the estate.

Customers placing an order252.011.0-241.0107,549telemetryAug 17, 2026
Customer effort scoreImproved

Lower is better. Simple queries got easier, which is what the automation was aimed at.

Customers with a billing query3.42.9-0.51,960surveyAug 7, 2026
Reached a named person rather than a queue or a botDegraded

The third inversion and the sharpest. Cost to serve fell by two thirds and fewer than half the customers who asked for a person got one. This is the number a regulator would ask about first.

Customers who asked to speak to someone69.044.2-24.83,140surveyAug 7, 2026
Decision explained in writing with the reason statedImproved

Automation made explanations cheap, so they got issued. A rare case where handing the work to a machine improved the paperwork rather than removing it.

Measures that an explanation was sent, not that it was understood. Nobody has tested comprehension.

Customers receiving a credit decision62.091.0+29.06,210telemetryAug 17, 2026
Satisfaction with the onboarding experienceImproved

Barely moved despite the cycle time collapsing. Suppliers care more about being paid than about being onboarded quickly.

Suppliers being onboarded3.53.6+0.1480surveyJul 21, 2026
Query answered correctly on the first responseImproved

Improved because the agent can see the full document trail in one place, which the previous shared mailbox could not.

Suppliers querying an invoice68.281.4+13.29,400telemetryAug 17, 2026
Complaints per 10,000 transactionsDegraded

Rate up 45 percent on a volume that more than doubled. In absolute terms complaints rose from roughly 420 to roughly 1,280 a week. Both the rate and the count moved the wrong way.

All service recipients9.814.2+4.4900,433telemetryAug 17, 2026
Net promoter scoreDegraded

Six points down while the cost to serve fell by two thirds. This is the headline the estate would rather not print and it belongs next to the saving, not away from it.

Modeled. It is a weighted reconstruction from the survey and complaint measures on this page rather than a directly fielded NPS study, and it should be read as a direction rather than a score.

All service recipients28.022.0-6.05,100modeledJul 31, 2026

Actions

Where throughput was bought with experience

Each of these is a measure that got worse while the corresponding volume, cost or cycle time improved. That trade may be acceptable. It has to be made deliberately.

Reached a person on the first attempt, when they asked for one

71.2 → 58.4

The clearest case of the trade the estate is making. Touchless handling went up 29 points and the share of people who got a human when they asked for one went down 12.8. Both numbers are real and they point in opposite directions.

Employees raising an HR case · survey · 2,410 responses · paired with H2R touchless share, 70 percent, up from 41

Satisfaction with the case outcome

4.3 → 4.1

Cases close nearly five times faster and satisfaction fell two tenths of a point. Small, consistent across four survey waves, and not noise.

Employees whose HR case was closed · survey · 3,180 responses · paired with H2R mean handling time, 4.2 hours, down from 19.6

Automated outcomes sent back for rework after a challenge

6.2 → 8.9

More work is automated and a larger share of it comes back. In absolute terms rework rose from about 470 to about 1,010 items a week.

Employees who challenged an automated outcome · telemetry · 11,400 responses · paired with R2R touchless share, 72 percent, up from 38

Resolved without needing a second contact

74.8 → 63.1

The second clear inversion. Disputes close three times faster and one in nine more customers has to come back a second time to get there. Faster is not the same as finished.

Customers raising a billing dispute · telemetry · 4,780 responses · paired with O2C dispute cycle time, 3.1 days, down from 9.4

Reached a named person rather than a queue or a bot

69.0 → 44.2

The third inversion and the sharpest. Cost to serve fell by two thirds and fewer than half the customers who asked for a person got one. This is the number a regulator would ask about first.

Customers who asked to speak to someone · survey · 3,140 responses · paired with C2L cost to serve, $9.86 per transaction, down from $28.90

Complaints per 10,000 transactions

9.8 → 14.2

Rate up 45 percent on a volume that more than doubled. In absolute terms complaints rose from roughly 420 to roughly 1,280 a week. Both the rate and the count moved the wrong way.

All service recipients · telemetry · 900,433 responses · paired with Estate volume, 900,433 transactions a week

Net promoter score

28.0 → 22.0

Six points down while the cost to serve fell by two thirds. This is the headline the estate would rather not print and it belongs next to the saving, not away from it.

All service recipients · modeled · 5,100 responses · paired with Estate cost to serve, $9.1M a week, down from $27.9M

Live observability

Every measure against its own baseline

The bar is the current reading as a share of the better of the two readings, so a longer bar always means a better experience regardless of which direction the raw number moves.

Movement since baseline

18 measures · 10 up · 1 flat · 7 down

Reached a person on the first attempt, when they asked for one

employee · Employees raising an HR case

71.2 → 58.4

Satisfaction with the case outcome

employee · Employees whose HR case was closed

4.3 → 4.1

Said the reason for the outcome was explained clearly

employee · Employees who received an automated outcome

66.0 → 66.0

Time to a usable answer

employee · Employees using self-service

8.7 → 3.2

Confidence in the decision without re-checking the source

employee · Managers reviewing agent-produced decisions

3.1 → 3.4

Automated outcomes sent back for rework after a challenge

employee · Employees who challenged an automated outcome

6.2 → 8.9

Satisfaction with the resolution

employee · Employees whose IT incident was resolved without a person

4.0 → 4.4

Described the automated close steps as understood, not opaque

employee · Finance staff on the monthly close

58.0 → 72.0

Hours a week spent handling exceptions, per person

employee · Operations staff across all towers

11.4 h → 6.1 h

Resolved without needing a second contact

customer · Customers raising a billing dispute

74.8 → 63.1

Time to order acknowledgement

customer · Customers placing an order

252.0 → 11.0

Customer effort score

customer · Customers with a billing query

3.4 → 2.9

Reached a named person rather than a queue or a bot

customer · Customers who asked to speak to someone

69.0 → 44.2

Decision explained in writing with the reason stated

customer · Customers receiving a credit decision

62.0 → 91.0

Satisfaction with the onboarding experience

customer · Suppliers being onboarded

3.5 → 3.6

Query answered correctly on the first response

customer · Suppliers querying an invoice

68.2 → 81.4

Complaints per 10,000 transactions

customer · All service recipients

9.8 → 14.2

Net promoter score

customer · All service recipients

28.0 → 22.0

Is policy and strategy coming to fruition

The stated position, tested against the telemetry

The program committed that automation would not be paid for out of the experience of the people around it. Here is what the record says.

Position

No experience measure degrades while its paired throughput measure improves.

Not holding — 7 measures of 18 moved the wrong way. 39% of the telemetry.

Position

A person can always reach a person.

Employees raising an HR case: 71.2 → 58.4 · Customers who asked to speak to someone: 69.0 → 44.2

Position

Experience is asked, not inferred.

Partly — 9 of 18 measures come from asking somebody. The rest are derived from system events, which tell you what happened but not how it felt.

What this page is, and what it is not

A security certification and a penetration test are organizational attestations. The product can house them, track their currency and surface their scope; it cannot manufacture them. Present the trust center as the place they live, never as evidence that they exist.

Experience telemetry has the same limit. These 18 readings are generated from the modeled estate; no survey on this page was sent to a real person and no customer was actually asked. What the page does establish is a discipline rather than a fact: experience is reported on the same surface and at the same cadence as throughput, so a cycle-time gain cannot be presented without the 7 measures that moved the other way appearing next to it.