Systems and records
Sales handoff implementation dataset: tracing promises into delivery
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Carry sales promises into implementation constraints by giving each promise its own record with source, date and the role that made it, linking it to the signed term that adopted, narrowed or excluded it, and tracing that term down to requirements, tasks and acceptance tests. Validate the links in both directions so orphan promises and orphan constraints surface as findings.
License terms
Sales handoff implementation
Key takeaways
- CRM opportunity fields record a deal's state, not the content of the commitments made during the sale.
- Rank sources by authority: signed agreements and statements of work outrank quotes, notes and CRM fields, and counsel settles conflicts.
- Trace in both directions: every constraint up to a signed term or a derived label, every adopted promise down to tasks and tests.
- Keep a failing record beside a valid one so reviewers can test the schema and the validator.
- Delivery outcomes are observations with an attribution limit, not proof that a kept promise caused a renewal.
- Replace people with roles and check customer confidentiality terms before records leave the company.
How do you carry a sales promise into an implementation constraint?
Give every promise an identity and trace it in both directions. Record each commitment made during the sale as its own row with its source, date and the role of the person who made it; link it to the signed term that adopted, narrowed or excluded it; and link that term down to the implementation tasks and acceptance tests that deliver it. A constraint with no upward link and a promise with no downward link are both defects.
That is the core of a sales handoff implementation dataset: a set of traced records that shows how what was said in the deal became, or failed to become, what the delivery team was asked to build. The CRM alone cannot provide it, because opportunity fields record the deal's state, not the content of the commitments.
This guide gives the minimum fields, the relationships between them, a valid and a failing worked record, a validator and the limits on what delivery outcomes can prove. It is written for revenue operations and implementation leaders who want handoff history that a reviewer can audit.
Where does a promise live before the contract is signed?
Promises are scattered across systems with very different authority. Map them before designing fields, so each row can say where its evidence came from and how much weight it carries.
Extraction works best backwards from the signed documents. Start with the executed agreement, order form and statement of work, list every deliverable, assumption and exclusion, then search the deal's emails, notes and quotes for the commitments that led to each one. Finally sweep the same sources for commitments that match nothing signed. That last pass finds the promises that were made and never written down, which are often the most useful rows in the dataset.
| Source | What it holds | Authority | Trap |
|---|---|---|---|
| CRM opportunity | Stage, close date, products, owner; Salesforce defines an opportunity as a sale or pending deal | Low: internal state, not a commitment | Fields are overwritten; history depends on field tracking |
| Call notes and emails | Specific commitments in the rep's or customer's words | Low to medium: evidence of what was said | Personal data and unstated context |
| Quote or order form | Products, quantities, terms offered | Medium: offer, sometimes signed | Superseded versions left attached to the deal |
| Statement of work | Scope, deliverables, assumptions, exclusions | High once signed | Assumptions that silently contradict a promise |
| Signed agreement and amendments | Binding terms and their precedence | Highest, subject to counsel's reading | Side letters and later amendments not linked to the deal |
| Signature record | Who signed and when | Proof of execution, not of scope | Treated as approval of every earlier promise |
Which signed term, approval or later outcome contradicts the CRM's apparent answer?
When sources disagree, the trace follows the executed documents, and counsel decides how they rank. A CRM field may show a module as sold while the order form lists it as a future option, or a call note may promise a go-live date that the statement of work makes conditional on the customer's data being ready. Record the conflict as a finding instead of choosing the more convenient answer.
Check status before reading content. Salesforce's ContractStatus values include Draft, In Approval, Activated, Terminated and Expired, so a contract record that looks complete may not be the version in force. Docusign's Certificate of Completion records recipient and summary events with timestamps, which proves execution and timing but not what was agreed outside the envelope.
Approvals can also contradict the CRM. A discount, a custom integration or a non-standard service level often needs a deal desk or executive approval, and the approval record can carry conditions that never reached the opportunity. Where an approval exists, trace the promise through it; where a promise needed approval and none exists, flag the row.
The sales to delivery requirement tracing schema
Version 1.0 of the schema below has one row per promise. Field names are suggestions; the relationships and checks are the substance. References point to records in your systems and stay there until a privacy and contract review clears them.
At a glance
The sales to delivery requirement tracing schema
| Field | Purpose | Example value (illustrative) | Required check |
|---|---|---|---|
| promise_id | Stable key for one commitment | PR-0412 | Unique within the dataset |
| account_key and deal_ref | Pseudonymous customer key and the opportunity or deal ID | ACC-77, OPP-9031 | Both resolve in the crosswalk |
| source_ref, source_type and made_on | Where the promise was found and its date | Email, message ref, day 12 of the sales cycle | Date precedes contract execution |
| promise_text_coded | Normalized statement of the commitment | Nightly sync of open orders to the customer's ERP | Written from the source, not from the delivered result |
| made_by_role and approval_ref | Role that made it and any approval it needed | Account executive; deal desk approval DD-118 | Approval present when policy requires one |
| contract_ref and adoption | Clause or SOW line, and adopted, narrowed, excluded or absent | SOW section 3.2, narrowed to daily | Contract status is activated or executed |
| constraint_ref | The implementation requirement derived from the adopted term | REQ-55 daily batch sync | Traces up to a contract_ref or is flagged as derived |
| task_refs and link_type | Delivery work items and how they relate | IMPL-201 relates to REQ-55 | At least one task for every adopted promise |
| acceptance_test and result | How delivery was verified and the result | Sync completes before 06:00 for 10 consecutive days: passed | Result dated after the task closed |
| outcome_note and attribution_limit | What happened after delivery and what the record cannot show | Customer reported fewer manual order entries; cause not isolated | Limit stated whenever an outcome is recorded |
Tracing rules that keep the schema honest
Traceability has a long history in engineering, and the rules transfer well. NASA's requirements management guidance describes bidirectional traceability: each requirement traces up to its parent and down to its allocated children, and a requirement with no parent is either traced to a source or marked as self-derived. Apply the same rule to delivery constraints. Every constraint traces up to a signed term; any constraint with no such term is a derived requirement and is labeled as one.
Use the link types your systems already offer and record which one was used. Jira's standard work item links include blocks, is blocked by, clones, duplicates and relates to, and administrators can configure others. HubSpot's associations API lets records be associated with or without labels, such as marking a contact as a company's decision maker. A link type is part of the evidence: relates to says less than implements, so prefer a custom link or label that states the relationship.
Watch for orphans in both directions. An adopted promise with no task means a commitment may have been dropped at the handoff. A task with no promise or contract term may be scope creep, a derived requirement, or a promise that was never written down; the annotator records which, with evidence.
Change orders and amendments need their own rows rather than edits to existing ones. When a customer and the delivery team agree mid-project to drop a promised feature or add a new one, record the amendment as a new contract_ref with its execution date, link the affected promise to it with a superseded or added status, and keep the original adoption visible. The history of how scope moved after signature is part of what the dataset teaches, and overwriting it leaves a trace that looks cleaner than the project was.
A valid record and a failing record, side by side
Both records below are invented to show the schema and do not describe any company. In the valid record, a promise made in an email on day 12 is adopted in narrowed form in the statement of work, traced to a requirement, delivered through two linked tasks and verified by an acceptance test. In the failing record, the same kind of promise exists only in a CRM note and a kickoff slide.
| Field | Valid record | Failing record | Why it fails |
|---|---|---|---|
| source_ref | Email from account executive, day 12 | CRM note added after go-live | Note postdates delivery, so it cannot show what was promised |
| adoption | Narrowed: real-time sync became daily batch in SOW 3.2 | Absent from SOW and order form | Promise never reached a signed term |
| approval_ref | Deal desk approval for custom integration | None | Custom integration required approval under the policy |
| constraint_ref | REQ-55 traces up to SOW 3.2 | REQ-61 with no parent | Orphan constraint not labeled as derived |
| task_refs | IMPL-201 and IMPL-204 implement REQ-55 | IMPL-310 relates to an unrelated epic | Link type does not show implementation |
| acceptance_test | Passed on dated test record | Customer email saying it works | Not a defined test with a date and criteria |
Validator: checks that should fail a trace
Run the validator on every row and store the result beside it. A row that fails is either repaired from a source record or excluded with a reason; it is never completed from memory.
- Every promise has a source_ref and a made_on date earlier than the contract execution date.
- Every promise has an adoption value, and adopted or narrowed promises cite a clause or SOW line in an activated or executed document.
- Every adopted promise traces down to at least one constraint and one task.
- Every constraint traces up to a contract_ref or carries the derived label with a named reason.
- Promises that policy says need approval cite an approval record dated before signature.
- Acceptance results are dated after the related task closed and name the criteria tested.
- Field history for the deal covers the full sales cycle; if retention ends earlier, the row records the gap.
- Any outcome note carries an attribution_limit.
What can the delivery outcome prove, and what can it not?
A trace can show that a promise was adopted, built and tested. It usually cannot show that keeping the promise caused a renewal, expansion or reference, because many other things change in the same period. Record outcomes as observations with dates and keep the attribution limit explicit.
Forecast rationale belongs in the same caution. Opportunity stage and forecast fields reflect the team's view at the time. Salesforce records changes in OpportunityFieldHistory only for the fields an administrator chose to track, and keeps standard field history for up to 18 months, or 24 months through the API, unless Field Audit Trail is in use. Export it before it ages out, and use it to date when a belief changed, not as evidence that the belief was right.
Removing people, prices and confidential terms before records travel
Sales records name people and expose commercial terms. Replace names and email addresses with role codes, drop pricing and discount values or convert them to coarse bands only if counsel approves, and remove unique product or deal details that would identify the customer to anyone who knows the market.
The customer's agreement may also limit reuse of its order forms, statements of work and communications. Check the confidentiality and data use terms for each account before its records enter the dataset. De-identifying a statement of work does not by itself create a right to license it.
Licensing sales-to-delivery histories through SourceX
Traced histories from promise to delivered result are a kind of operational record AI developers look for. For a business that licenses them, SourceX takes on sourcing, rights review, contracting, data preparation, delivery and payment, and the business approves scope and terms and signs before delivery.
Eligibility is a quick, upload-free assessment, and the contract and privacy review described earlier still governs which customer accounts may enter the set.
Where do the records show how far sold scope moved after the handoff?
On the sales order line and the time booked against it, where sold, delivered and invoiced quantities sit side by side. The trace above carries each promise to a signed term, tasks and an acceptance test, but it never measures how much work a sold line took or who paid for any difference, and that is where post-handoff corrections show up.
Some ERP systems keep this link natively. In Odoo 18.0, a service product can be set to create a project for the order with a task for each sales order line, and each task names the sales order item its time is billed to. For lines delivered by timesheets, the delivered quantity is computed from the time booked, and prepaid time lines show the time remaining on the order. When a confirmed prepaid or fixed-price service line that is not fully invoiced logs more delivered time than was ordered, times a per-product threshold that defaults to 1, Odoo schedules an upsell activity for the salesperson and marks the line so the warning appears only once.
ERPNext records the gap from the project side: a project links to its sales order and totals actual hours, costing and billable amounts from timesheets, and each timesheet row keeps hours worked beside billing hours. These links can be edited later (Odoo's help text says the sales order item can be changed or removed on each timesheet entry), so read a link's change history, not only its current value. Add five fields to each sold line in the schema:
- sale_line_ref, joined to promise_id: the order line that priced the promise.
- sold_qty, with its unit and billing policy: prepaid or fixed price, time and materials, or milestone.
- delivered_qty and its source (timesheets, milestones or manual entry), with the date it was read.
- invoiced_qty and unbilled_effort, the hours worked minus the hours billed on that line.
- correction_refs: each change order, credit note, write-off or upsell activity, dated, with approval_ref and the approving role.
Who absorbed the gap between sold and delivered?
Whoever the dated documents say paid for it: the customer through a change order, the delivery firm through unbilled hours, or nobody when delivery stayed inside the sold quantity. Give every sold line a correction label on that evidence, and split the line's hours when more than one label applies, so the parts add up to what was delivered. A label says who carried the cost; on its own it does not say whose mistake the gap was.
| Correction label | Evidence that proves it | Who carried the cost | What it cannot show |
|---|---|---|---|
| Delivered as sold | Delivered quantity within the sold quantity, or within a tolerance fixed in advance; no correction records | Nobody beyond the agreed price | That the estimate was good, since a generous price can hide a small overrun |
| Customer-funded correction | Signed change order or upsell order citing the same promise_id, dated after kickoff | The customer | Whether the promise was under-scoped or the customer asked for more; the change order's stated reason decides |
| Vendor-absorbed overrun | Delivered beyond sold plus funded changes, or hours worked above hours billed, with no matching invoice | The delivery firm | That delivery was slow, since customer delays and third-party changes also add hours |
| Narrowed by agreement | Amendment or credit note that reduces the line, signed by both parties | The customer receives less, usually for less | Which side proposed the narrowing |
| Dropped without record | Adopted promise whose line shows no delivered quantity and no amendment | The customer, until someone notices | That anyone decided to drop it |
| Built outside the sold lines | Booked hours on tasks with no sales order item, or on a promise the contract never adopted | The delivery firm, unless billed later | Whether the work was goodwill, scope creep or internal |
Was the go-live smooth because the handoff was right, or because someone paid for it?
The go-live alone cannot say; the corrections and their dated causes can. Corrections can buy a smooth delivery, and a messy one can come from causes outside the handoff, such as a customer's data arriving late, so treat a clean go-live as something to explain rather than as proof that the handoff carried the right scope.
Take an illustrative extension of the valid worked record above, with every value invented. Promise PR-0412 was priced as 40 prepaid hours on line SOL-1, and its two tasks booked 52. Forty hours fell inside the sold quantity, 8 were paid through a signed change order for retry logic the statement of work never mentioned, and 4 were written off. The daily sync passed its acceptance test and went live nine days late, and a dated customer ticket shows the first data extract arrived eight days after the agreed date. By its outcome, the handoff worked. By its corrections, it under-scoped the integration by 12 hours, while most of the delay sat with the customer. A second promise on the account, PR-0415, was priced at 30 hours on fixed-price line SOL-2 and booked 27.
Before labeling any handoff accurate, apply these rules:
- Accurate needs a passed acceptance test, no customer-funded correction traced to the same promise, and vendor-absorbed hours under a tolerance fixed before anyone reads the data.
- A delay or overrun counts as external only when a dated record from outside the delivery team supports it, such as a customer ticket, a vendor notice or a late countersignature.
- Each correction is charged to the promise it cites, never spread evenly across the account.
- Deal outcome, renewal and satisfaction scores stay out of the handoff label.
Can the dataset show every promise end to end? A trace completeness check
Only if six dataset-level ratios clear thresholds fixed in advance, each published with its numerator and denominator. The validator above fails individual rows; these six checks decide whether a collection of traces is complete enough to train or evaluate on. Repair a failing set from source records or narrow it to the accounts that pass; never complete it by hand.
The illustrative case is a permitted one: it assumes counsel reviewed the customer agreement behind every row and cleared de-identified delivery records for reuse, with people replaced by roles as the privacy section above requires. A script applying the labels and checks to this invented packet is held for reviewers. An AI team can put the same table into a data request as its acceptance test; SourceX works from such requests and finds matching records at U.S. businesses, with no standing inventory to sell from. The checks, with the illustrative case's results:
| Check | Numerator / denominator | Threshold (illustrative) | Illustrative case |
|---|---|---|---|
| Line coverage | Sold service lines with at least one linked task / all sold service lines | 100% | 2 of 2 |
| Effort linkage | Hours on tasks linked to a sold line / all hours booked on the account's projects | 90% | 79 of 85 hours (92.9%), the 85 including the 27 booked to PR-0415; 6 hours built a dashboard the contract never adopted |
| Promise closure | Adopted promises with a dated acceptance result / adopted promises | 100% | 2 of 2 |
| Correction linkage | Change orders and write-offs citing a promise_id / all correction records | 100% | 2 of 2 |
| Approval coverage | Correction records with an approval reference / all correction records | 100% | 1 of 2: the 4-hour write-off has none, so the set fails |
| Reuse clearance | Rows whose customer agreement counsel cleared for reuse / all rows | 100% | 11 of 11 in this permitted case |
Sources for this section
- Odoo 18.0's sale_project module (commit 421ec75) lets a service product create a project for the order with a task for each sales order line, links each task to the sales order item its time is invoiced against, and notes that the sales order item can be changed or removed on each timesheet entry. Source
- Odoo 18.0's sale_timesheet module computes delivered quantity from timesheets, shows the time remaining on prepaid time lines, and schedules an upsell activity for the salesperson, marking the line so the warning shows once, when a confirmed, not fully invoiced prepaid or fixed-price service line delivers more than its ordered quantity times a threshold that defaults to 1. Source
- ERPNext's project doctypes link a project to its sales order, total actual time, costing and billable amounts from timesheets, and keep hours, a billable flag and billing hours on each timesheet row. Source
Frequently asked questions
What is a sales handoff implementation dataset?
It is a set of traced records that link each commitment made during a sale to the signed term that adopted or excluded it, the implementation tasks that delivered it and the test that verified it. It shows how sales promises became delivery constraints, or where they were lost.
Why is the CRM not enough to trace sales promises?
CRM opportunity fields record the deal's state, such as stage, products and close date, not the content of individual commitments. Promises live in emails, notes, quotes and statements of work, and only the signed documents show which ones bind.
What should happen when a promise is not in the contract?
Record it with adoption set to absent and keep the row. If a delivery task exists anyway, the trace shows work done outside the signed scope; if none exists, it shows a promise dropped at the handoff. Both are findings, and counsel decides what the contract obliges.
How do you show a valid and an invalid trace?
Keep at least one passing record and one failing record with the validator's result for each. The failing record should name the exact check it fails, such as an orphan constraint or a source dated after go-live, so reviewers can test the schema.
Can delivery results prove that a kept promise caused a renewal?
Usually not. The trace can show that a promise was adopted, built and tested, but renewals depend on many factors. Record outcomes with dates and an explicit attribution limit rather than a causal claim.
Sources
- Salesforce's object reference describes the Opportunity object as representing a sale or pending deal. Source
- Salesforce's object reference describes OpportunityFieldHistory as the history of changes to the values in an opportunity's tracked fields. Source
- Salesforce's ContractStatus reference lists contract status values including Draft, In Approval, Activated, Terminated and Expired. Source
- Docusign's Certificate of Completion lists envelope details, recipient events and summary events with timestamps. Source
- Salesforce's help documentation says that without Field Audit Trail, field history data is kept for up to 18 months, and up to 24 months through the API. Source
- NASA's requirements management guidance describes bidirectional traceability of each requirement to its parent and allocated children, with parentless requirements marked as self-derived. Source
- Atlassian's Jira documentation lists work item link types such as blocks, clones, duplicates and relates to, which administrators can configure. Source
- HubSpot's associations API documentation says records can be associated unlabeled or with labels, such as a contact marked as a company's decision maker. Source
See if your company qualifies
A short company assessment. No data uploads are needed.