Reported operating use · Technology and sales operations
AI-assisted executive dinner planning: Merge example
OpenAI reports Merge’s dinner bot preparing guest and event materials from business-system context. Merge says a person reviews and curates the list; its time comparison is not independently audited.
Prepare executive dinner guest lists, outreach and event tracking
Merge
OpenAI reports internal use at Merge. The company describes people reviewing guest selection and outreach; reported time comparisons are company-reported and not independently audited.
A published third-party example, not a Nuvepro customer case or evidence of a partnership.
The workflow at a glance
Preparing an executive dinner with human curation
Illustrative summary of the published account
Reported work components. Their arrangement does not imply a fixed execution order or simultaneous work.
- Event planning input
City and intended guest mix
Merge describes starting a Dallas dinner with a city and the kinds of people the team wanted to invite.
- Dinner bot
Research candidates and contact history
The bot pulled prospects from CRM and Apollo and checked previous invitations and responses.
- Dinner bot
Prepare planning materials
It helped prepare outreach, maintained the attendee tracker, proposed seating and created calendar notes for sales and customer-success context.
- CEO and event team
Curate relationships and host
Shensi Ding reviews, rewrites and selects using relationship context. People retain hosting, relationship-building and event logistics.
What the source reports
OpenAI reports that for a Dallas executive dinner, Merge’s bot used CRM and Apollo to identify potential guests, checked prior invitations and responses, prepared outreach, maintained the attendee tracker, proposed a seating chart and created calendar notes for sales and customer-success context. CEO and co-founder Shensi Ding reviewed and rewrote output and applied relationship context the bot lacked. Merge reported that preparing guest lists, outreach and the tracker had fallen from dozens of company hours to about 30 minutes per event, measured over the preceding three months. Hosting, relationship-building and event logistics remained with people. OpenAI is the publisher; the time comparison is not independently audited.
Reported technology: OpenAI ChatGPT, GPT-6 Astra and Codex with Merge’s business-system connectors. Scope and outcomes are source-reported.
What the enterprise environment would need
The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.
Proposed practice environment
Plan a dinner from fictional relationship records
Illustrative project design to scope with your enterprise.
- 1 · Set up the environment
Mock event and attendee records
Fictional invitees, account owners, previous invitations, response/opt-out state and event constraints.
- 2 · Preparation and agent contributions
Prepare the event packet
Draft a candidate list, outreach, attendee tracker, seating suggestions and calendar notes with source references.
- 3 · Independent checks
Check contact and event constraints
Independently reconcile identity, duplicates, prior replies, opt-outs, owner context, seating and calendar conflicts.
- 4 · People decide acceptance
Host accepts the exact draft
Relationship owners confirm context and the event host decides the final guest list and communications.
A reviewable evidence package
A versioned event plan with candidates, reasons, draft outreach, tracker, seating proposal and held exceptions.
If a check fails → withhold acceptance.
A duplicate contact has an opt-out or an old invitation response. Withhold outreach, reconcile the contact state, rerun checks and return the corrected packet for host review.
Failure or owner rejection returns to review
- Hold
- → 2 · Revise work
- → 3 · Rerun independent checks
- → 4 · Owner review
Operating boundary. Fictional records and mock email/calendar only. No real invitations, contact edits or event commitments.
- Systems and tools
- Published: the dinner workflow drew candidate and invitation context from CRM and Apollo; the account also names Salesforce, Gong, Google Sheets and Gmail. Proposed: use mock CRM, invitation history, calendar and email tools with fictional contacts.
- Data and state
- Published: the event begins with a city and a target guest mix, and previous invitations/responses are checked. Proposed: version attendee, consent, relationship-owner, event and seating records.
- People and handoffs
- Published: CEO Shensi Ding reviews, edits and curates the guest list; people host and build relationships. Proposed: relationship owners confirm context and an event host accepts the final practice packet.
- Rules and authority
- Proposed: generate drafts and tracker changes in a sandbox only. Require human approval for invite selection and sending; no real messages or calendar updates.
- Normal and exception cases
- Published: prior invitation and response history informs candidate selection. Proposed: include opt-outs, prior declines, duplicate guests, conflicting account ownership and incomplete relationship notes.
- Reset and replay
- Not reported by the source. Proposed: reset fictional contact and invite state between runs; preserve each suggestion, human edit and rejection against the same fixture version.
A proposed simulation
An illustrative practice design derived from this work, not an executed assessment.
Proposed: prepare a fictional executive dinner plan from approved attendee and invitation records while preserving contact preferences and human relationship judgment.
Review gate: A separate checker confirms identity, prior contact, opt-out status, account-owner context and event constraints; the named relationship owner and host accept the exact draft.
Challenge the result: A high-fit contact has opted out and a duplicate record has a different invitation history. The checker must suppress the contact, reconcile identity and withhold the draft until the event owner reviews it.
What needs checking
- Source-reported: candidate sourcing, prior invitation checks, outreach preparation, attendee tracking, seating proposals and calendar notes are described for an executive dinner.
- Source-reported: Merge reported about 30 minutes per event compared with dozens of staff hours for selected planning tasks over the prior three months; methodology is not independently validated.
- Proposed: independently check opt-out and prior-contact status, duplicate identities, relationship-owner notes and seating constraints.
- Proposed: event and account owners accept the exact draft; these checks are designed, not executed.
Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.
How an enterprise could build similar work
A proposed build order for an enterprise doing similar work, written by Nuvepro from the published account and our interpretation of the environment and checks above. It is not how Merge built or operates this work.
- Define the fictional eventSet a city, target guest mix, schedule and explicit event constraints. Keep all candidate identities and messages fictional throughout the exercise. Use only the agreed fictional audience and event purpose.
- Prepare contact historyVersion fictional account ownership, prior invitations, responses, preferences and opt-out state. Include consent and contact history in the versioned test records. Keep invitation status and consent independent for each record.
- Draft the event packetPrepare candidate reasons, outreach drafts, tracker updates, seating suggestions and calendar notes. Label every suggestion with its supporting record and uncertainty. Do not infer personal affinity from unsupported data.
- Check identity and contact rulesIndependently verify duplicates, previous replies, opt-outs, account-owner context and calendar conflicts. Keep the checking logic separate from the drafting agent. Reject stale contact preferences before drafting outreach.
- Repair held itemsWithhold mismatched or incomplete contacts, correct the records and rerun independent checks. Compare changes and rejected candidates after each replay. Test that rejected candidates remain excluded after replay.
- Have the host acceptThe relationship owner and event host review the exact packet before any action. Practice approval is not permission to contact real people. Record the relationship owner and host decision for review. Hosting and relationship work remain with people.
Your systems, data, rules and accountable owners decide the real project. Agree acceptance criteria, escalation owners and operating responsibility with the people accountable for the work before anything runs on live records. Treat the first run as a rehearsal on test records: compare the draft output with an independently prepared answer, and record what the people reviewing it change.
Questions about this example
- What work did the Merge dinner bot help with?
- OpenAI reports that it searched CRM and Apollo, checked invitation history, helped prepare outreach, maintained a tracker, proposed seating and created calendar notes for an executive dinner.
- Did the system replace the event host?
- No. Merge’s CEO reviewed and edited the output and applied relationship context. People retained hosting, relationship-building and event logistics.
- How should the reported time comparison be read?
- Merge reported about 30 minutes per event for guest lists, outreach and attendee tracking compared with dozens of company hours over the prior three months. OpenAI publishes the account; no independent time audit is supplied.
- What would a similar enterprise practice need?
- A proposed exercise uses fictional contacts and invitation history, checks opt-outs and duplicates independently, and requires a human host to approve the exact event packet. It sends no real invitations.
Shape this around your enterprise
Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.
Sources and limitations
Curation review: . This is the date this source example was reviewed, separate from source publication dates.
- The case is an OpenAI-published customer account, not an independent time study; the reported comparison is not independently audited.
- The comparison covers guest lists, outreach and attendee tracking; it should not be generalized to event hosting or relationship work.
- The proposed environment is fictional and unexecuted. It does not reproduce Merge’s systems, contact data or business outcomes.
- Merge builds product and business tools with OpenAI technologySource publication: 2026-11-05
More sales and marketing examples
- AI ad campaign management: Advolve exampleAdvolve
- AI launch readiness checks: RingCentral exampleRingCentral
- AI lead follow-up dashboard: Zapier exampleZapier
- AI RFP contract agent experiments: Box exampleBox internal experiments