AI uses for records
Fixed asset exception dataset: what transfers and disposals reveal
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Transfers and disposals reveal the asset-register decisions a finance AI can be tested on: classifying the event, checking that a permitted, independent approver authorized it, recomputing gain or loss from net proceeds and carrying amount, tying the period roll-forward and catching postings that cleared dimensions or kept depreciating a disposed asset. Each needs the event linked to its request, approval, vouchers and outcome.
Record timeline
Fixed asset exception
Key takeaways
- Routine depreciation repeats a system rule; transfers, reclassifications and disposals record decisions worth learning from.
- Many exceptions come from system behavior, such as a transfer clearing dimensions nobody meant to change.
- A usable record has two layers: the event with before and after states, and links to request, approval, policy and outcome.
- Gain or loss, the period roll-forward and approval authority are the three checks that reconcile a record.
- Keep failed and corrected events with their reasons and outcome labels; they are the training signal.
- Replace names with role codes and decide how much location and counterparty detail the task really needs.
What useful asset-register tasks can transfers and disposals reveal?
Transfers and disposals are where an asset register stops being a static list and records decisions: someone moved an asset to another cost center or legal entity, reclassified it, sold it or scrapped it, and someone else approved it. Linked to the request, the approval and the resulting postings, those events support tasks a finance AI can be checked on: classifying the event, judging whether the approval was sufficient, recomputing the gain or loss, reconciling the period roll-forward and spotting postings that broke the register.
The useful part is the exceptions. Routine depreciation runs teach almost nothing, because the system calculates them. A disposal posted before it was approved, a transfer that silently blanked a cost center, or a reclassification that left a transfer unfinished are the cases where a reviewer has to reason from documents, authority and the final financial state, which is exactly what a model needs to learn.
Where do exceptions hide in the register's transaction history?
They hide in system behavior that is correct by design but easy to misuse. Microsoft's Dynamics 365 Finance documentation is a good illustration because it spells the mechanics out. A transfer moves a fixed asset book to a new financial dimension set, and every dimension value changes when the transfer occurs, whether a value is entered or left blank. A transfer meant to change only the business unit can therefore clear the cost center and department as well.
Reclassification behaves differently again. Moving an asset to a new group or number creates new books, sets the original asset's books to Closed, and cancels and regenerates the existing transactions for the new asset. If the asset had a transfer transaction, the system flags in the Action center that a transfer was not completed and has to be finished separately. A history that only shows the end state hides that sequence.
Disposals change status fields as well as balances. Disposing of an asset as scrap updates the book's status to Scrapped and sets the disposal date, while the acquisition and accumulated depreciation are reversed through the posting profile. The same documentation notes that a voucher can hold only one fixed asset transaction, and that splitting an asset or calculating depreciation on disposal produces separate vouchers. Every one of those details decides how records must be linked before they form a usable packet.
Other systems name these steps differently, so map your own transaction types to the event classes below rather than assuming the labels carry over.
Which lifecycle events count as exceptions worth keeping?
An event counts as an exception when it departed from the normal pattern and someone had to decide what to do about it. Keep those, together with enough ordinary events of the same type for contrast.
| Event | Normal pattern | Exception signal | Evidence that resolves it |
|---|---|---|---|
| Transfer between cost centers or departments | One dimension changes; the rest are carried over | Dimensions cleared that the request never mentioned | Transfer request, before and after dimension values, correcting transfer |
| Reclassification to a new group | Original books closed and transactions regenerated on the new asset | Transfer flagged as incomplete after reclassification | Action center message, follow-up transfer voucher |
| Intercompany split or transfer | Clearing accounts balance in both entities | Service life or exchange rate option differs from policy | Transfer wizard choices, intercompany journals, policy version |
| Disposal by sale | Approval precedes posting; proceeds recorded against the asset | Posting dated before approval, or sale with no buyer document | Sale approval, invoice, disposal voucher |
| Disposal by scrap | Status set to Scrapped with a disposal date and a reason | Scrap with no physical evidence or by the asset's custodian alone | Scrap request, photo or certificate, approver independent of custodian |
| Depreciation after disposal | No depreciation dated after the disposal date | Depreciation posted for periods after disposal | Reversal voucher and corrected roll-forward |
| Asset missing at physical count | Count matches the register | Asset in the register but not found | Count sheet, investigation note, write-off or disposal approval |
What should the lifecycle exception and approval linkage schema include?
The schema has two layers. The event layer describes what happened to the asset in the system, with the state before and after. The linkage layer ties that event to the documents, the people who acted, the policy in force and the outcome that proved whether the event was right. A record is complete only when both layers are filled.
Event layer fields are listed first. Every field name is a suggestion; keep the meaning and the validation rule even if your column names differ.
| Field | Type | Purpose | Validation rule |
|---|---|---|---|
| packet_id and schema_version | string | Identifies the packet and the schema it follows | Unique; version matches the published schema |
| asset_id and asset_book | string | The asset and the book the event touched | Exists in the register extract for the period |
| event_type | enum | transfer_dimension, transfer_intercompany, reclassification, split, disposal_sale, disposal_scrap, count_adjustment | One value from the list |
| event_date and posting_date | date | When the event took effect and when it was posted | Posting date not before event date unless a backdating reason is present |
| voucher_ids | list of strings | Every voucher the event produced, including separate ones | Each voucher references this asset only |
| state_before and state_after | object | Group, status, location, all financial dimensions, service life | Every dimension present in both; changes listed explicitly |
| carrying_amount_at_event | decimal with currency code | Book value used to judge a disposal | Equals acquisition less accumulated depreciation and impairment at the event date |
| net_proceeds and gain_loss_posted | decimal with currency code | What the sale brought in and what was booked | Gain or loss equals net proceeds less carrying amount |
Which linkage fields connect the event to authority and outcome?
Linkage fields carry the evidence a reviewer needs to say whether the event should have happened and whether it was booked correctly. They are the fields most often missing from register exports, because the approval sits in a workflow tool, an email or a capital request form outside the ledger.
| Field | Type | Purpose | Validation rule |
|---|---|---|---|
| request_ref | document reference | The request or form that started the event | Resolves to a document dated on or before the event |
| preparer_role and approver_role | role code | Who prepared and who approved, as roles not names | Approver differs from preparer and from the asset custodian |
| approval_date | date | When authority was given | On or before posting date, or exception_reason set |
| authority_policy_ref | policy id and version | The approval limits and disposal policy in force | Version effective on the event date |
| supporting_evidence_refs | list of document references | Invoice, scrap certificate, count sheet, photos | At least one per event type that requires it |
| exception_reason | enum plus note | Why the event departed from the normal pattern | Required when any validation rule fails |
| correction_refs | list of voucher ids | Reversals or follow-up postings that fixed the event | Each correction dated after the original |
| roll_forward_ref and outcome_label | reference and enum | The period report that shows the final state, and whether the event stood, was corrected or was reversed | Outcome consistent with correction_refs |
Can the reviewer reconcile the answer to documents, authority and the final financial state?
Yes, if three checks pass. The first is arithmetic. Under IAS 16, the gain or loss on derecognizing an item of property, plant and equipment is the difference between the net disposal proceeds, if any, and its carrying amount, and carrying amount means cost after accumulated depreciation and impairment. A packet that holds all three values lets anyone recompute the booked figure.
The second is the period roll-forward. Dynamics 365 Finance's roll-forward report, for example, gives opening and closing net book value per asset with the period's acquisitions, depreciation, write-ups, write-downs and disposals in between, plus the disposal date, net book value at disposal and the profit or loss calculated on disposal. Opening value plus the period's movements should equal closing value for every asset in the packet.
The third is authority. The approval must come from a role allowed to approve that event under the policy version in force, and from someone other than the person who prepared it. GAO's Green Book, the federal internal control standard, treats segregation of duties as part of designing control activities, and the same principle is how most private companies frame disposal approvals.
For U.S. taxpayers the packet can also note where a sale of depreciable business property is reported. The IRS instructions for Form 4797 use Part III to figure ordinary income recapture on property such as section 1245 property, with any gain beyond the recapture amount reported elsewhere. A tax reviewer, not the dataset, decides the treatment; the packet only records which form and part the company used.
A passing and a failing asset record, side by side (illustrative)
Two illustrative records show the schema at work. Neither describes a real company, and they use counts and dates rather than amounts.
Valid record (illustrative). A packaging machine moves from one plant's cost center to another's. The request is dated three days before the transfer, the approver is the receiving plant's controller, and the preparer is a fixed asset accountant. State_before and state_after list all five financial dimensions; only cost center and location change. One voucher is produced, the roll-forward for the month shows the asset under the new cost center with unchanged net book value, and outcome_label is stood.
Failing record (illustrative). A forklift is disposed of as scrap. The disposal is posted on the 4th of the month, the approval is dated the 11th, and the preparer and approver are the same role. Depreciation for the following month is posted after the disposal date, and a reversal voucher appears six weeks later. The validator flags three failures: approval after posting, no segregation of duties, and depreciation after disposal. The record still belongs in the dataset, with exception_reason set and outcome_label corrected, because detecting exactly this pattern is a task the dataset should teach.
Which reproducible checks should run before a packet is released?
Run these on every release and keep the failing records. Each check maps to a validation rule in the schema tables, so a failure points straight to the field to inspect.
At a glance
Which reproducible checks should run before a packet is released
- Completeness: every event has both layers filled, or an exception_reason explaining what is missing and why.
- Voucher linkage: every voucher id resolves in the ledger extract and references a single asset.
- Dimension integrity: compare state_before and state_after; flag any dimension cleared or changed that the request did not mention.
- Date order: request, approval, event and posting dates in a plausible order; backdated postings carry a reason.
- Gain or loss recomputation: booked gain or loss equals net proceeds less carrying amount within a stated rounding tolerance.
- Post-disposal activity: no depreciation or transfer dated after a disposal unless a correction reverses it.
- Roll-forward tie-out: opening value plus period movements equals closing value per asset and per group.
- Authority: approver role permitted for the event type under the policy version in force, and different from the preparer.
What must be removed or restricted before an asset packet leaves the company?
Replace people's names with role codes in every approval and preparer field, and remove free-text notes that name employees or describe disciplinary matters, which sometimes appear in investigations of missing assets. Counterparty names on sale invoices and bank details belong out of the packet unless the buyer's task needs them and the agreement allows it.
Location detail deserves a decision too. A plant code is usually enough; street addresses and security-related notes about where equipment is stored rarely help a model and add risk. Policy documents referenced by authority_policy_ref may be confidential; include the approval thresholds as ratios or tiers if the documents themselves cannot leave.
Turning asset lifecycle exceptions into a licensed finance dataset
Fixed asset histories with approvals and corrections are the kind of operational record finance AI developers ask about, and SourceX helps companies turn selected business records like these into licensed datasets, handling rights review, contracting, preparation and delivery. Buyers generally look for several years of consistent history and a substantial number of records in one system, though rarer work can qualify with less, and every case is assessed on its own facts. The qualification check is a brief assessment that asks for no uploaded files.
Frequently asked questions
What is a fixed asset exception dataset?
It is a set of asset-register events, such as transfers, reclassifications and disposals, that departed from the normal pattern, each linked to its request, approval, postings and final outcome. The links let a reviewer or a model judge whether the event was authorized and correctly booked.
Why are disposals more useful than depreciation runs for training finance AI?
Depreciation is calculated by the system from settings, so it repeats a rule. Disposals and transfers involve a request, an approval, a valuation and sometimes a correction, which are the judgments a finance model has to learn to check.
Can a fixed asset transfer change more than the field the user intended?
In some systems, yes. Dynamics 365 Finance documentation states that all financial dimension values change on a transfer whether a value is entered or left blank, so unmentioned dimensions can be cleared. Comparing before and after states catches it.
How is the gain or loss on an asset disposal checked?
Under IAS 16 it is the net disposal proceeds less the carrying amount, where carrying amount is cost after accumulated depreciation and impairment. A packet that stores all three values lets a reviewer recompute the booked figure and flag differences.
Should failed or corrected asset events be removed from the dataset?
No. Keep them with the exception reason, the correcting vouchers and an outcome label saying the event was corrected or reversed. Those records are what teach a model to spot the same failure, as long as they are labeled correctly.
Sources
- In Dynamics 365 Finance, a transfer moves a fixed asset book to a new financial dimension set, and all dimension values change whether a value is entered or left blank. Source
- Reclassifying a fixed asset creates new books, closes the original books, cancels and regenerates transactions, and flags an incomplete transfer in the Action center. Source
- Disposing of a fixed asset as scrap sets the book status to Scrapped and the disposal date, reversing acquisition and depreciation through the posting profile. Source
- Only one fixed asset transaction is allowed per voucher, and splits or depreciation on disposal use different voucher numbers. Source
- Intercompany split and transfer generates journals through clearing accounts and offers service life and exchange rate options for the destination asset. Source
- The fixed assets roll forward report gives opening and closing net book value, period movements, disposal date, net book value at disposal and profit or loss. Source
- IAS 16 sets the gain or loss on derecognition as net disposal proceeds less the carrying amount of the item. Source
- Form 4797 instructions use Part III to figure ordinary income recapture, with gain above the recapture amount reported elsewhere. Source
- GAO's 2025 Green Book places segregation of duties within the principle on designing control activities. Source
See if your company qualifies
A short company assessment. No data uploads are needed.