Skip to content

Systems and records

Utility field visit exception data: when the technician cannot get in

By SourceX Editorial · Reviewed by Noah Loul ·

8 sections

Short answer

When a technician cannot access the site, record the visit as an exception, not a canceled job: keep the work order and asset, the attempt timestamps, one reason code from a fixed list, the technician's evidence, any safety boundary, the dated rule that governs the next step, and the outcome of the follow-up visit, reading or referral. That turns utility field visit exception data into gradable tasks.

System export

Utility field visit

This page covers utility field visit exception data: when the technician cannot get in.Illustrative

Key takeaways

  • A no-access visit is its own record, linked to the work order, not a reason to cancel the work order quietly.
  • Booking status timestamps are often the best evidence of what happened on the day.
  • The dated rule, the asset type and any recorded hazard decide which next step is permitted.
  • A recorded hazard turns a scheduling task into a safety routing task.
  • Prove outcomes with the follow-up work order, reading or referral, never with a final status label.

How do you represent a field visit when the technician cannot access the site?

Represent it as an exception record attached to the visit, not as a canceled job. The record holds the work order and asset, the attempt with its timestamps, a reason code from a fixed list, the evidence the technician left, any safety boundary that applied, the rule that governs the next step and the outcome of the rescheduled or alternative visit.

Most systems store the opposite. The work order closes as canceled or rolls back to unscheduled, the booking is marked canceled, and the reason survives only as a line of free text. That loses the decision a dispatcher, billing analyst or customer service agent made next, which is the part an AI model would need to learn or be tested on.

Utility field visit exception data, in this article, means the records created when a planned visit to a meter, service line or customer premises does not happen as planned: no access, unsafe conditions, wrong or missing asset, customer refusal or a visit cut short. The annotation guide below turns each of those records into one labeled task with a verifiable result.

The guide is written for administrative and operational records. It is not safety guidance, and nothing here replaces a utility's own safety rules, operating procedures or regulator requirements.

What a failed visit leaves behind in a field service system

A failed visit usually leaves a booking whose status history tells more than the work order does. Reading the booking first is how an annotator reconstructs what happened on the day.

Microsoft's documentation for Dynamics 365 Field Service is a useful reference model because it separates the two. It lists work order system statuses (Unscheduled, Scheduled, In Progress, Completed, Posted and Canceled) and default booking statuses (Scheduled, Traveling, In Progress, On Break, Completed and Canceled), and says work orders and bookings have independent lifecycles in which a booking's state influences the work order's status.

Timing comes from status changes. The same documentation says every booking status change creates a booking timestamp, that moving a booking to In Progress fills the actual arrival time, and that moving it to Traveling from the mobile app sets the start time. A no-access attempt therefore often shows Traveling with no In Progress, or a short In Progress followed by Canceled, and the gap between those timestamps is evidence in its own right.

Organizations add their own detail on top. Microsoft describes custom work order substatuses for more granular states, custom booking statuses mapped to a parent Field Service status, incident types that codify common issues and procedures, and resolution records that report how work orders are resolved. Utilities running other platforms, such as SAP, Oracle or IFS, have comparable objects under different names; the annotation guide maps to the concept, not the product.

What a failed visit leaves behind in a field service system
RecordWhat it shows about the failed visitCommon gap
Work order and typeWhat was supposed to be done and for which asset or premisesClosed as canceled with no link to the follow-up order
Booking status historyTraveling, arrival, cancellation and their timestampsStatuses changed later from the office, so times reflect data entry
Substatus or custom booking statusThe organization's own label, such as no access or customer not homeSeveral labels used for the same situation
Technician notes and photosLocked gate, hazard, missing meter, card left at doorGate codes or personal details mixed into the note
Customer contact logCalls, texts or letters before and after the attemptHeld in a separate customer system with no visit ID
Follow-up work order or billRescheduled visit, estimated bill, customer-supplied readingNot linked, so the outcome cannot be traced

Which rule, meter fact or hazard decides the next step after a no-access visit?

Three kinds of fact decide what is allowed next: the regulation or tariff in force on the visit date, the asset and its service conditions, and any safety boundary the technician recorded. The annotation must cite the version in force on the day, because the next step can be correct under one and wrong under another.

Meter reading is the clearest example of a dated rule. Pennsylvania's residential billing standards, 52 Pa. Code section 56.12, allow a utility to estimate a bill when its personnel are unable to gain access to obtain an actual meter reading, subject to conditions: the utility must take reasonable alternative measures such as postcards or telephone reporting of the reading, obtain an actual or customer-supplied reading at least every 6 months (or every four billing periods where longer billing periods are allowed), and obtain an actual reading at least once every 12 months. Other states and other utility services have their own rules, and rules are amended over time, so the packet stores the citation and the version date it used.

Asset facts matter as much. The same no-access event means one thing for a remotely read meter, where the reading can come from the system, and another for a manually read meter whose last actual reading is approaching the limit. Contract terms add a third layer: commercial customers may have agreed access windows, escort requirements or notice periods that a residential customer does not.

Safety boundaries override everything else. If a technician records a hazard, such as an aggressive animal, a suspected leak, damaged equipment or an unsafe structure, the next step is set by the utility's safety procedure and the task is labeled as a safety stop. A model should learn to route those cases, not to schedule around them.

Which rule, meter fact or hazard decides the next step after a no-access visit?
Fact checkedExample conditionHow it changes the next stepEvidence to cite
Reading historyLast actual reading near the regulatory limitEscalate to a priority access attempt rather than another estimateReading history with dates and the rule citation
Meter technologyRemotely read meter versus manual meterRemote read may resolve billing without a revisitAsset record with meter type
Repeat attemptsSecond consecutive no-access visitDifferent customer notice and scheduling rules may applyPrior attempt records and notices sent
Commercial access termsEscort or advance notice required by the accountBook with notice instead of same-week revisitAccount terms or site access instructions
Recorded hazardSuspected leak or aggressive animalSafety procedure governs; no routine reschedule labelTechnician note, photo and safety ticket reference
Customer-supplied readingCustomer submits a reading by phone or cardMay satisfy the billing step while the physical visit stays openContact log entry and reading record

The exception and rescheduling outcome annotation guide

The guide below is the asset. Each row is one field an annotator fills for every failed visit, with the values allowed, the evidence required and the rule for deciding. Fields with no supporting record are left blank and flagged rather than guessed.

Keep labels from a closed list. Free-text reasons are where most exception data loses its value, because three teams describing the same locked gate in three ways make the records impossible to compare.

The exception and rescheduling outcome annotation guide
FieldAllowed valuesEvidence requiredAnnotation rule
Visit identityWork order ID, booking ID, asset or premises pseudonym, visit dateWork order and booking recordsOne task per attempted visit, never per work order
Visit purposeMeter read, meter exchange, service connect or disconnect, inspection, repairWork order type or incident typeUse the organization's type mapped to this list
Attempt evidenceTraveling time, arrival time, departure time, time on siteBooking timestampsFlag if times were entered from the office after the fact
Exception reasonLocked or blocked access, customer absent, customer refused, animal, hazard, asset not found, wrong address, weather, otherTechnician note, photo or substatusPick one primary reason; record secondary reasons separately
Safety boundaryNone, hazard recorded, safety stop invokedNote, photo, safety ticket referenceAny hazard makes the task a routing case, not a scheduling case
Governing ruleRegulation or tariff citation with version date, account term, internal procedure IDRule text or procedure document dated on or before the visitCite the version in force on the visit date
Next-step decisionReschedule, contact customer, estimate or alternative reading, escalate, refer to safety, closeDispatcher or office action logRecord who decided and when, not just what happened
Customer communicationCard left, call, text, letter, noneContact logLink by visit ID; remove phone numbers and addresses
OutcomeCompleted on revisit, resolved without visit, repeated failure, escalated, closed unresolvedFollow-up work order, reading or bill recordOutcome must come from a later record, not the exception record itself
Label confidenceConfirmed, inferred, unknownAnnotator noteInferred labels stay out of grading sets

Worked example: a locked gate before a scheduled meter exchange (illustrative)

This example is illustrative. The utility, premises and counts are invented to show how the guide is applied.

Starting state: a work order to exchange a manually read residential meter, booked for a morning window. The booking history shows Traveling, then In Progress for 6 minutes, then Canceled. The technician's note says the side gate was padlocked and nobody answered; a photo shows the locked gate and a door card. The account's reading history shows the last actual reading was taken 11 months earlier, with estimates since.

Rule applied: the packet cites the state rule on actual readings in force on the visit date, which in this illustrative case requires an actual reading at least once every 12 months. That makes a routine reschedule in four weeks the wrong choice, because it would push the account past the limit.

Decision: the office calls the customer the same day, books a confirmed appointment within the week and records the customer's preferred access instructions in the account notes, with the gate code itself left out of the packet.

Verified outcome: a completed meter exchange 4 days later, with an actual reading on the removed meter. The outcome field cites the completed follow-up work order and the reading record, not the original booking.

Exception: the same locked gate, but the technician's note also reports a strong gas odor near the meter set. The task is no longer a scheduling task. It is labeled hazard recorded, the next-step decision is refer to safety, and the outcome cites the safety ticket. A model that proposes rebooking the meter exchange for later in the week has failed the task, however reasonable its schedule looks.

  • Visit purpose: meter exchange; exception reason: locked or blocked access
  • Attempt evidence: Traveling, 6 minutes In Progress, Canceled
  • Governing rule: actual-reading interval, cited with version date
  • Decision: contact customer and book within the week; outcome: completed exchange with actual reading
  • Exception case: hazard recorded, routed to safety, no reschedule label

Grading a model's next-step decision against the outcome

Grade the decision and the outcome separately. The decision is correct when it matches what the governing rule and the evidence allow; the outcome shows whether that decision actually resolved the visit.

Agent benchmarks make the same separation. tau-bench, for instance, scores a task by comparing the database state an agent produced with the state produced by replaying the ground-truth actions, and checks any required outputs on their own. For field exceptions, the equivalent end state is the follow-up work order, reading or safety referral that exists after the decision, not the message the dispatcher sent.

A useful rubric has four checks: the model chose a next step permitted by the cited rule, it respected any safety boundary, it used the evidence in the record rather than assumptions, and its proposed step matches the reviewer's decision. A fifth, optional check compares the step with the recorded outcome to see whether similar decisions tended to succeed.

Do not grade on the work order's final status. A work order can read Completed because a different crew finished it weeks later after several more failures, and a rescheduled booking can read Scheduled because nobody closed it.

Records to leave out or send to a safety and confidentiality reviewer

Some exception records should not enter a dataset at all, and others need a specialist before they do. Decide this per record, using the safety boundary and confidentiality fields of the guide.

Leave out records that describe security arrangements at a site: gate and lockbox codes, alarm details, key locations and escort contacts. Send anything involving critical facilities, substations, regulated pipeline assets or law enforcement involvement to the utility's security or compliance owner before it goes further.

Customer premises data is personal data. Names, addresses, phone numbers and the pattern of when someone is home are all sensitive, and removing names does not on its own give the utility the right to license the records. Access to the field service system, privacy permission and the contractual right to license are three separate questions, and counsel should review the tariff, customer terms and any regulator rules that apply to customer information.

Finally, narrow the scope when the governing rule cannot be reconstructed. If the utility cannot say which tariff or procedure version applied on the visit date, the record can still be used to study exception reasons, but it should not be used to grade next-step decisions.

  • Hazard or safety stop recorded: label as routing case and have a safety reviewer confirm
  • Security details in notes or photos: remove the detail or drop the record
  • Critical facility or regulated asset: compliance owner approves first
  • Rule version unknown: keep for reason-code analysis only
  • Customer identifiers: pseudonymize and review customer terms with counsel

Licensing field exception histories for AI

Exception histories linked to rules and outcomes are scarce, because most systems record only the work that was finished. A utility or contractor that can produce packets in the shape above holds the kind of decision record AI developers look for when they build dispatch, customer service and billing agents.

When a utility or field contractor licenses exception histories, SourceX is responsible for sourcing, the rights review, contracting, preparation, delivery and payment. Exception records face the same volume test as any dataset: as a general guide, buyers look for multi-year history, typically two years or more, with tens of thousands of visits logged in one system, while unusual work may qualify with fewer, decided case by case.

The utility approves scope and terms and signs before any visit record is handed over, and the fit check that comes first is a short assessment with no records attached.

Frequently asked questions

What is utility field visit exception data?

It is the set of records created when a planned utility field visit does not happen as planned, such as no access, a recorded hazard, a missing asset or a customer refusal. Useful exception data links the attempt, the reason, the rule that governed the next step and the outcome of the follow-up.

Should a no-access visit be recorded as a canceled work order?

Not on its own. Canceling the work order hides the attempt and the decision that followed. Keep the attempt as its own record with a reason code and timestamps, and link it to the follow-up work order, reading or referral that resolved it.

What reason codes should a no-access exception use?

Use a short closed list, such as locked or blocked access, customer absent, customer refused, animal, hazard, asset not found, wrong address and weather. Record one primary reason and keep secondary reasons separately, so records from different teams stay comparable.

Can a utility estimate a bill when the meter cannot be reached?

Many state rules allow estimated bills under conditions. Pennsylvania's residential rules, for example, allow estimates when personnel cannot gain access, but require alternative measures and periodic actual or customer-supplied readings. The applicable rule depends on the state, the service and the date, so the utility's own regulatory team should confirm it.

How should hazards in exception records be handled?

Treat any recorded hazard as a safety case, not a scheduling case. Label it so that the expected next step is a referral under the utility's safety procedure, and have a safety reviewer confirm the label before the record is used to train or test anything.

Sources

  • Microsoft documents Field Service work order system statuses and default booking statuses, says work orders and bookings have independent lifecycles, and lists the fields each booking status updates. Source
  • Microsoft says booking timestamps record the date and time a booking status changes and are created by default when the booking status changes. Source
  • Microsoft describes creating custom booking statuses and mapping each to a parent Field Service status. Source
  • Microsoft describes resolution records used to report how work orders are resolved and to suggest resolutions by incident type. Source
  • Microsoft describes incident types as templates that codify common issues, procedures and resolutions for work orders. Source
  • Pennsylvania's 52 Pa. Code section 56.12 allows estimated residential bills when utility personnel cannot gain access to read the meter, subject to alternative measures and periodic actual or customer-supplied readings. Source
  • tau-bench scores a task by comparing the database state the agent produced with the state from replaying the ground-truth actions, and checks required outputs separately. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies