Systems and records
Sales forecast correction history data: explaining each revision
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
A sales forecast revision is explained by the dated evidence around it, not by the final number: the forecast submitted at each cutoff, who adjusted it and why, the deal changes to amount, stage, close date and category underneath, the signed contracts and approvals that confirmed or contradicted them, and what customers later bought, received and paid, all rebuilt as of each cutoff.
Record timeline
Sales forecast correction
Key takeaways
- Keep submissions, manager adjustments and deal changes as separate dated records; the rollup hides which one moved the number.
- Current CRM records overwrite the past, so history tables or cutoff snapshots are required, and both have retention limits.
- Signed contracts, amendments and approvals outrank CRM fields when they disagree; record the conflict instead of fixing the field.
- Capture a reason code at each adjustment and label target-setting moves as decisions, not forecasts.
- Judge both the period number and each revision's stated reasons, because a forecast can land for the wrong reason.
- Build features only from data recorded before each cutoff, and index values before anything leaves the company.
What evidence explains a sales forecast revision beyond the final number?
A revision is explained by the dated records around it: the forecast as submitted at each cutoff, the adjustment and who made it, the deal-level changes to amount, stage, close date and forecast category underneath, the signed contract, amendment or approval that supported or contradicted those changes, and what the customer later bought, received and paid. The final number is the last line of that history, not its explanation.
CRM forecasting tools already separate several of these layers. In Salesforce, a forecast manager's change to a subordinate's or child territory's forecast is stored as a ForecastingAdjustment record, distinct from an owner adjusting their own forecast, and the ForecastingItem object exposes amounts with and without adjustments, including AmountWithoutAdjustments, AmountWithoutManagerAdjustment, OwnerOnlyAmount and ForecastAmount. Salesforce's help adds that when a manager and a subordinate adjust the same forecast, only the manager's adjustment rolls up the hierarchy.
A sales forecast correction history dataset keeps those layers apart, dates each one, and joins them to contract and outcome evidence a revenue team can defend.
Why can't today's CRM tell you what last quarter's forecast knew?
The current opportunity record shows where each deal ended up, not what was believed when the forecast was called. Amounts, close dates and categories are overwritten as deals move, so the history has to come from history tables or from snapshots taken at each cutoff.
History tables have limits worth checking before you rely on them. Salesforce field history tracking covers up to 20 standard and custom fields per object, and without Field Audit Trail the data is retained for up to 18 months in the org and up to 24 months through the API. Stage history lives in its own OpportunityHistory object. HubSpot returns past values of deal properties when the request includes the propertiesWithHistory parameter, but only for the properties you name.
Categories add a second trap. Salesforce maps each opportunity stage to a forecast category, with defaults of Pipeline, Best Case, Commit, Closed and Omitted, and users can edit the Forecast Category field on opportunities they own, overriding the stage mapping. A deal can therefore sit in Commit at an early stage, and the dataset should keep both the stage and the category at each cutoff.
Which dated records make up the forecast decision and outcome dataset?
The dataset is a set of dated records keyed to a forecast period and a cutoff. Every revision must be traceable from the rolled-up number down to the deals and documents behind it. Field names are suggestions; the checks decide whether a period is usable.
- Values in the dataset can be kept in the original currency inside the company and converted to an index or share before anything leaves; the reasoning works on relative changes.
- Store roles rather than names for submitters and adjusters, and keep the role map internal.
| Record | Required fields | Source | Check |
|---|---|---|---|
| Period and cutoff | Forecast period, cutoff timestamp, forecast type, hierarchy level | Forecast settings and calendar | Every other record carries a cutoff it belongs to |
| Submission | Submitter role, submitted value, category column, note, submitted at | Forecast tool submissions | One submission per role per cutoff, or a recorded gap |
| Adjustment | Adjuster role, level adjusted, value before and after, adjusted at, reason code, note | Forecast adjustment records | Before and after values reconcile to the rollup |
| Deal change | Deal ID, field, old value, new value, changed at, changed by role | Field and stage history, or cutoff snapshots | Every material change is dated, not inferred from the final record |
| Contract evidence | Agreement or order form ID, signature date, effective date, term, amendments | Contract repository or e-signature records | Signed documents link to the deal they close |
| Approval | Discount, term or credit approval with approver role, decision and date | Quote or approval workflow | Approval precedes the commitment it supports |
| Outcome | Booked, delivered, invoiced, paid, reduced, cancelled or renewed, each with a date | Billing, ERP and customer success records | Read only after the outcome window closes |
| Entity map | Account, parent, subsidiary and billing entity IDs across systems | CRM account hierarchy and ERP customer master | One customer resolves to one entity group |
| Attribution note | Which outcome evidence can and cannot be tied to the revision | Reviewer | Present on every period, including when nothing can be attributed |
When a signed contract and the CRM disagree, which record wins?
Signed documents and recorded approvals outrank CRM fields, and outcomes outrank both when the question is whether the forecast was right. When they disagree, the record should show the conflict, not quietly correct the CRM value.
Revenue accounting shows why the CRM can lag. Under ASC 606, a contract modification is a change in the scope or price, or both, of a contract that is approved by the parties; it can be approved in writing, by oral agreement or by customary business practice, and can exist while the parties still dispute the price. A deal record that says nothing changed may sit beside a modification that already did.
| What the CRM shows | What the evidence shows | How the record should read |
|---|---|---|
| Closed Won in the quarter | Order form signed with a start date in the next quarter | Booked in period, delivery and billing start later; both dates kept |
| Amount unchanged after close | Signed amendment reduced scope two months later | Original booking plus a dated modification event |
| Commit category, early stage | Owner override of the stage mapping, no signed paper | Category and stage both stored; override flagged |
| Discount within standard | Approval record for a deeper discount the week before close | Approval linked; deal amount reflects the approved terms |
| Renewal forecast at full value | Customer success notes and billing show reduced usage and a partial renewal | Outcome recorded as partial; forecast error explained by usage, not slippage |
How should the reason for each adjustment be recorded so it can be tested?
Capture a reason code and a short note at the moment of each adjustment, because reasons reconstructed after the quarter closes are shaped by what happened. HubSpot's forecast tool, for example, takes an amount and a note with each submission and keeps a submission history that can be viewed for up to one year back, which is a starting point if the notes are used consistently.
Forecasting research is blunt about adjustments. Hyndman and Athanasopoulos write that practitioners adjust forecasts much more often than they should, that adjustments should not try to correct a systematic pattern the model supposedly missed, and that shifting a number to set targets or plan supply belongs to goal setting, not forecasting. A reason code list that separates those cases lets a later reviewer test which kinds of adjustment helped.
- Deal evidence: a signature, a loss, a slip or a scope change on named deals, with the deal IDs.
- Known event: information the deal data does not hold yet, such as a customer reorganization announced to the account team.
- Pattern correction: a belief that the rollup is systematically high or low; flag these, because they are the adjustments most likely to add noise.
- Target or plan: a number moved to match a goal or a supply plan; keep it, but label it as a decision rather than a forecast.
- Unknown: no reason recorded; keep the adjustment and the label, and do not invent a reason later.
Worked example: a commit cut in week 10, partly restored in week 12 (illustrative)
This example is illustrative. The figures are index values, with the week 1 commit set to 100, and describe no real company.
A regional team calls commit at 100 in week 1 of a 13-week quarter. In week 10 the regional manager cuts it to 82. In week 12 it is raised to 91. The quarter closes at 93.
- Answer to the plan question: the week 10 revision is explained by one dated close-date change and one loss, not by the final 93; the week 12 revision by two signed upsells.
- Limit: the record cannot say whether the slipped deal would have closed with a different discount; that is outside the evidence.
| Date | Event | Evidence | Effect on the record |
|---|---|---|---|
| Week 1 | Commit submitted at 100 | Submission with note citing six named deals | Baseline at first cutoff |
| Week 9 | Largest deal's close date moves to next quarter | Field history on close date; customer email about budget timing | Deal change event with dated source |
| Week 10 | Manager adjusts commit to 82, reason code deal evidence | Adjustment record with before and after values | Adjustment explained by the slipped deal and one smaller loss |
| Week 11 | Mid-size deal signed with a reduced scope | Signed order form; discount approval dated two days earlier | Contract evidence and approval linked |
| Week 12 | Commit raised to 91, reason code deal evidence | Adjustment record citing two upsells signed in week 12 | Partial restoration tied to signed paper |
| Quarter end plus 60 days | All signed deals invoiced; one paid late | Billing and payment records | Outcome recorded; late payment noted, not an error in the forecast |
What if the forecast landed for the wrong reason?
A forecast that lands close to the outcome can still rest on wrong reasons, and the dataset should say so. If the week 10 cut had been for a deal that then closed while a different, unforecast deal was lost, the final number would look accurate while both underlying calls were wrong.
Label correctness at two levels: the period number against the outcome, and each revision's stated reason against the deals it named. A period where the number was right and the reasons were wrong is a valuable case, but only if the dataset keeps the deal-level evidence that shows it.
How do you build as-of snapshots so later edits do not leak into the past?
Every feature in a training or evaluation case must come from what was recorded before the cutoff, and every label from what happened after it. Joining the current deal table to an old forecast is the most common way a correction history dataset quietly cheats.
Build the as-of view from history records or stored snapshots, then test it.
- Rebuild each cutoff's pipeline from history and compare its total with the forecast submitted at that cutoff; unexplained gaps mean missing history.
- Reject any feature whose changed-at timestamp is after the cutoff.
- Treat records edited after close, such as retroactive close dates, as their own events rather than overwriting the as-of value.
- Keep periods older than the history retention limit only where snapshots exist, and mark them as snapshot-based.
- Hold out whole later periods for evaluation rather than random deals, so seasonality and team changes are tested honestly.
Which customer, contract and seller-performance limits apply to revenue records?
Forecast records carry customer names, contract terms and pricing that agreements often treat as confidential, and they expose individual sellers' performance. Before any period leaves the revenue team, check customer contract confidentiality clauses, replace customer and seller identities with stable pseudonymous IDs, convert values to indexes or shares, and confirm employment and privacy policies allow the use.
Having export access to the CRM is not the same as having the right to license what is in it. Third-party enrichment fields, partner-sourced deals and data covered by a reseller agreement may carry their own restrictions, and removing names does not by itself create a right to share contract terms.
Narrow the scope or leave a period out when the history cannot be rebuilt as of the cutoff, when a single customer dominates the period so that indexing still reveals them, or when contracts in the period forbid disclosure of terms.
Can a company with a long forecast history license it?
Companies that have run a consistent forecast cadence for years, with history tracking and contract records kept, hold exactly the dated evidence this kind of dataset needs. SourceX works with such companies to license chosen slices of their records to AI developers, and it handles the rights review, contract, preparation and delivery along the way.
Nothing has to be uploaded to find out whether a company fits; a brief assessment of the business comes first. The company signs off on scope and terms, and signs the agreement, before a single forecast record moves, which leaves room to exclude sensitive regions, customers or quarters.
Frequently asked questions
What should a dated forecast decision and outcome dataset include?
Period and cutoff records, each submission and adjustment with role, reason and before and after values, dated deal changes, signed contract and approval evidence, outcomes read after the window closes, an entity map and an attribution note. Every record is tied to the cutoff it belongs to.
How far back can we rebuild forecast history from Salesforce?
It depends on what was tracked. Without Field Audit Trail, Salesforce retains field history for up to 18 months in the org and up to 24 months through the API, for up to 20 tracked fields per object. Older periods need snapshots or exports taken at the time.
Is the final closed amount a good label for a forecast?
Only for the period-level number. To explain revisions you also need each adjustment's reason checked against the deals it named, and outcomes such as delivery, billing and renewal that the closed amount does not show.
Should management target-setting adjustments be removed?
Keep them and label them. They are decisions, not forecasts, and separating them lets a reviewer measure forecast accuracy without punishing or rewarding a target that was never meant as a prediction.
Do we need actual revenue values in the dataset?
Usually not. Indexing each period to its first commit or expressing changes as shares keeps the reasoning intact while removing pricing that contracts may treat as confidential.
Sources
- Salesforce's object reference describes ForecastingAdjustment as an individual forecast manager's adjustment of a subordinate's or child territory's forecast, separate from ForecastingOwnerAdjustment. Source
- Salesforce's ForecastingItem object exposes forecast amounts including AmountWithoutAdjustments, AmountWithoutManagerAdjustment, ForecastAmount and OwnerOnlyAmount. Source
- Salesforce's help on pipeline forecast adjustments says that when a manager and a subordinate adjust the same forecast, only the manager's adjustment rolls up the hierarchy. Source
- Salesforce lists Pipeline, Best Case, Commit, Closed and Omitted as the standard forecast categories. Source
- Salesforce maps each opportunity stage to a forecast category, and users can edit the Forecast Category field on opportunities they own, overriding the default mapping. Source
- Salesforce field history tracking covers up to 20 fields per object and, without Field Audit Trail, retains history for up to 18 months in the org and 24 months through the API. Source
- HubSpot's deals API returns current and historical values of the deal properties named in the propertiesWithHistory parameter. Source
- HubSpot's forecast tool records submissions with an amount and a note, and shows submission history up to one year back. Source
- ASU 2014-09 (Topic 606) defines a contract modification as a change in the scope or price, or both, approved by the parties, which may be approved in writing, orally or by customary business practice. Source
- Forecasting: Principles and Practice (3rd edition) advises using judgmental adjustments sparingly, not to correct supposed systematic patterns, and treats target or supply shifts as goal setting rather than forecasting. Source
See if your company qualifies
A short company assessment. No data uploads are needed.