Yes. AI can help find commitments discussed in sales calls before the project team starts work. Use it to prepare a referenced commitment register covering dates, deliverables, exclusions and dependencies. Sales and delivery should review that register before using it to assign work or confirm a schedule.
The key distinction is between something mentioned and something agreed. “We could launch next month” must retain its tentative meaning. A promise conditional on receiving customer content needs both the proposed delivery date and its condition.
This is a practical use case for CRM automation: carrying relevant sales evidence into delivery without asking the project manager to reconstruct every conversation.
Extract commitments as separate items
A call summary might say everyone discussed a website launch. A useful handover identifies what sales offered, what the customer requested, what remains undecided and who must act next.
Ask the proposed workflow to extract individual items across these categories:
- Deliverables: what will be supplied, including quantity, format and acceptance conditions where stated.
- Dates: promised deadlines, suggested targets and dates requested by the customer.
- Exclusions: work explicitly left outside the offer.
- Dependencies: customer inputs, third-party access or decisions needed before delivery.
Keep separate promises separate. Design approval, launch and staff training can have different owners and conditions, even when discussed together.
Preserve the words that limit a promise. Removing “subject to approval” changes its meaning. An exclusion should retain its boundary: excluding ongoing maintenance does not automatically exclude launch support.
Capture promises that were later withdrawn or changed, too. Otherwise an early offer may survive in the handover after the conversation has moved on. Link the change to the original item so the reviewer can follow the sequence.
Separate a request from an agreement
Use a small set of proposed review states. These are operating rules for your workflow, rather than product features or legal conclusions.
| State | Meaning | Expected human handling |
|---|---|---|
| Customer request | The customer asked for something | Sales checks whether it was accepted |
| Tentative proposal | An option or target was suggested | Keep outside the committed schedule |
| Conditional commitment | A promise depends on a stated event | Confirm the condition and responsible owner |
| Apparent agreement | The conversation appears to contain acceptance | Sales verifies the exchange and written scope |
| Unclear or conflicting | Evidence is incomplete or inconsistent | Assign clarification before scheduling |
An enthusiastic response does not settle every detail. “That sounds good” might acknowledge an approach without accepting a date or quantity. Include the preceding statement and relevant response so the reviewer can judge the exchange.
Compare the call evidence with the accepted quotation and later correspondence. Where they disagree, show both versions. The workflow prepares evidence for the responsible person to resolve; it should not decide which statement creates a binding obligation.
Keep the extracted state separate from the human decision. A reviewer can confirm a tentative proposal after obtaining clarification, but that later confirmation needs its own reference.
Make every finding traceable to the call
Start with the correct opportunity, recording and participants. A similar customer name is insufficient to attach promises to a project. If the match is uncertain, hold the extraction for sales to identify the correct record.
OpenAI’s supplied documentation describes transcription of completed recordings and a specialised option that returns speaker-labelled segments with start and end metadata. These provide possible building blocks for a referenced transcript. Source: OpenAI speech-to-text
Your proposed handover still needs a usable reference: recording identifier, call date, speaker and timestamp or transcript location. If timestamps are unavailable, retain a stable paragraph reference and label it accordingly. Never invent a precise audio location.
Check consequential words against the recording, especially negations, quantities and dates. A transcript saying “included” when someone said “not included” could reverse the scope. Where audio is unclear, record the uncertainty and request confirmation.
File search can retrieve information from uploaded files using semantic and keyword search, and its output can include file references. A file reference alone does not establish the exact spoken passage. Source: OpenAI file search
A search result should therefore lead the reviewer back to surrounding evidence. Finding one passage about training does not establish that every training discussion has been checked.
Use this sales commitment handover brief
Copy this brief for each opportunity. Add an item for every distinct promise, request or exclusion; leave unknown information explicitly unknown.
Sales commitment handover brief
- Opportunity: customer name, verified opportunity reference and project name.
- Evidence reviewed: call identifiers and dates, quotation version and relevant follow-up messages.
- Coverage gaps: missing recordings, incomplete transcripts or unreviewed conversations.
Repeat for each extracted item
- Item reference: a stable identifier within this handover.
- Subject: deliverable, date, exclusion or dependency.
- Candidate meaning: a plain-language summary preserving qualifications.
- Evidence: short exact excerpt, speaker and recording timestamp or transcript location.
- Candidate state: customer request, tentative proposal, conditional commitment, apparent agreement, or unclear/conflicting.
- Conditions and boundaries: required inputs, acceptance conditions and stated exclusions; write “unknown” where missing.
- Related evidence: supporting, conflicting or repeated statements, with references.
- Sales decision: confirmed, corrected or clarification required; reviewer and reason.
- Delivery decision: feasible, feasibility check required or blocked; responsible owner.
- Next action: specific question or task, assigned person and agreed follow-up date.
Before releasing the handover
- The customer and opportunity match the evidence.
- Every retained item has a source reference or is marked unsupported.
- Tentative wording and dependencies remain visible.
- Repeated items are linked; conflicting versions remain distinguishable.
- Sales has reviewed the meaning of each apparent commitment.
- Delivery has reviewed dates, capacity and prerequisites.
- Unresolved items have owners and are excluded from committed scheduling.
Work through ordinary and difficult cases
The following business scenarios, dates, quantities and recording references are entirely hypothetical.
Ordinary agreement: a supplier discusses a customer portal. At call reference C-A, 12:40, the salesperson offers three onboarding sessions after launch. At 13:05, the customer accepts. The extractor prepares one apparent-agreement item with both references. Sales checks the accepted quotation; delivery confirms who will run the sessions before creating tasks.
Conditional date: the salesperson says, “We can target 30 November if your approved product list arrives by 6 November.” Preserve the target and dependency together. The project manager checks whether the input has arrived and whether the target is feasible. Keep this as conditional planning until those checks are resolved.
Ambiguous scope: the customer asks for data migration and someone answers, “We should be able to help.” Record a tentative proposal with unknown scope. Sales asks which data, how much and whether migration is included in the offer. Delivery receives a scoping question rather than an assumed migration task.
Missing evidence: an account manager recalls a promise from an unrecorded call. Label it as a reported statement awaiting confirmation. Ask the account manager and customer to confirm the details in writing; do not manufacture a transcript citation. Absence from available recordings does not establish that no promise was made.
Duplicate or conflicting evidence: the same training promise appears in a call and its follow-up email. Link both sources to one item. If the email instead mentions two sessions, preserve the conflict. Sales must establish whether the quantity was revised or recorded incorrectly before delivery allocates time.
Review the register before changing project records
A consistent record structure makes omissions visible. OpenAI’s structured-output documentation describes responses constrained to a supplied schema. This can organise fields, but a completed field is not proof that its contents match the conversation. Source: OpenAI structured outputs
Sales reviews what was said and accepted. Delivery reviews dates, capacity and prerequisites. A promise can be confirmed by sales yet still require escalation because delivery cannot meet it. Preserve that disagreement and assign a commercial decision rather than quietly adjusting the promised date.
Start with a proposed workflow that prepares the brief for review. Before deployment, check current account, plan and regional eligibility for selected services, alongside recording access and approved handling of customer information.
For the wider design, AI agents versus automation helps frame the choice between fixed routing rules and interpretation. A custom AI agent could assist with varied language within a limited custom agent workflow.
FAQ: sales call commitment handovers
Should the project team receive the full transcript?
Give them the reviewed register and authorised access to relevant evidence. Include enough surrounding conversation to check disputed items. The handover should explain outstanding decisions without requiring everyone to read every call.
What if nobody knows who made the promise?
Keep the speaker marked unknown. Ask the call owner to identify them and check their role. Until resolved, treat the statement as unconfirmed and keep it outside the committed delivery schedule.
Can AI send the customer a confirmation automatically?
For this proposed workflow, prepare a draft for sales review. Check dates, exclusions and unresolved points before sending. The message should explicitly ask for clarification where needed, without turning tentative suggestions into new promises.
If your business loses context between sales and delivery, try this brief on a completed opportunity. For help shaping an AI automation workflow around your handover, get in touch.