Research · Dataset
Enterprise AI Workflow Index
93 published accounts of enterprise work done with AI, classified by function, industry and evidence class, with the human checkpoint each source describes.
Every row links to a source that the organization or creator published. These are third-party examples, not Nuvepro customer cases. Nuvepro adds the classification and the review date. Sources reviewed through 1 November 2026. Licence: Creative Commons Attribution 4.0 International (CC BY 4.0). Nuvepro's classification and summaries are licensed under CC BY 4.0. Linked sources remain under their owners' terms. People are named only as the sources they link to name them. If you reuse it, credit “Nuvepro, Enterprise AI Workflow Index” and link to this page.
Licensed by Nuvepro Technologies Pvt Ltd. Nuvepro's own columns are short_work, function, industry, evidence, human_checkpoint, curation_date, nuvepro_url and license. The source-derived columns are company, source_url and source_date. The index is provided as is. It reflects each source as published on its review date, has not been independently audited and may change. Listing an organization or person does not imply a partnership or endorsement. To correct or remove an entry, contact us.
The index
| Work | Source | Function | Industry | Evidence | Human checkpoint |
|---|---|---|---|---|---|
| M&A deal sourcing diligence | OffDeal | Banking and Insurance | Financial Services | Reported operating use | Not stated in the source |
| service investigation dashboards | Grafana Labs | IT Operations | Technology | Limited published evidence | No review step stated; users get the assistant inside Grafana's own interface with access to their existing observability context |
| regulatory document drafting | Bluenote | Research and Regulated Documentation | Healthcare and Life Sciences | Reported operating use | Human expert review of draft documents, which arrive with review flags |
| scientific literature review | FutureHouse | Research and Regulated Documentation | Healthcare and Life Sciences | Limited published evidence | Not stated in the source |
| security alert triage | Trellix | Security | Technology | Reported operating use | Analysts receive Wise's investigations and recommendations; customer authority boundaries are not quantified in the source |
| security alert investigation | Panther | Security | Technology | Reported operating use | Not stated in the source |
| month-end accruals | Campfire | Finance and Accounting | Technology | Reported operating use | Not stated in the source |
| ad campaign management | Advolve | Sales and Marketing | Technology | Reported operating use | Not stated in the source |
| support request routing | Assembled | Customer Service | Technology | Reported operating use | Exceptions are routed between AI and human agents, and human edits to responses are used in evaluations |
| fraud risk review | Inscribe | Banking and Insurance | Financial Services | Limited published evidence | Not stated in the source |
| refund eligibility | Decagon | Customer Service | Technology | Reported operating use | Not stated in the source |
| insurance claims intake | DXC Technology | Banking and Insurance | Financial Services | Reported operating use | Humans keep checkpoints on material legal or financial decisions, and flagged work is routed to people for authorized review |
| car buyer qualification | Cars24 | Sales and Marketing | Retail and Consumer | Reported operating use | Not stated in the source |
| security investigation | Block | Security | Financial Services | Reported operating use | Humans validate results and guide deployment |
| threat investigation response | Sophos | Security | Technology | Reported operating use | Humans oversee uncomfortable or potentially destructive work, and analysts calibrate the authority boundaries within the Notify, Collaborate and Authorise modes |
| talent market research | Oracle | HR and Recruiting | Technology | Reported operating use | Humans stay responsible for design and output; one business user checked results against the previous manual process |
| agreement review | Sullivan & Cromwell | Legal | Legal and Professional Services | Limited published evidence | Lawyers inspect the sources and accept, adapt or reject the suggested changes |
| customer service actions | Intercom | Customer Service | Technology | Reported operating use | Not stated in the source |
| delivery exception handling | Nash | Field Service and Logistics | Logistics and Transport | Limited published evidence | Not stated in the source |
| expense policy review | Ramp | Finance and Accounting | Financial Services | Reported operating use | Reviewers retain authority, and ambiguous or missing context escalates to them |
| claims escalation investigation | Oscar Health | Banking and Insurance | Healthcare and Life Sciences | Limited published evidence | Not stated in the source |
| sales deal briefs | Rippling | Sales and Marketing | Technology | Limited published evidence | No review step stated; an early tester reports building the agent end to end without an engineering team |
| appointment booking | Practo | Customer Service | Healthcare and Life Sciences | Reported operating use | Not stated in the source |
| legal research and drafting | LexisNexis Legal & Professional | Legal | Legal and Professional Services | Reported operating use | Human evaluators, organized by workflow, jurisdiction and language, review output quality |
| lead follow-up dashboard | Zapier | Sales and Marketing | Technology | Limited published evidence | Not stated in the source |
| launch readiness checks | RingCentral | Sales and Marketing | Technology | Limited published evidence | Not stated in the source |
| freight invoice review | Dow | Finance and Accounting | Manufacturing and Energy | Reported operating use | Not stated in the source |
| recruiting shortlists | The Adecco Group | HR and Recruiting | IT Services and Staffing | Reported operating use | Recruiters review the shortlists the workflow produces |
| student service cases | Wiley | Customer Service | Telecom, Media and Education | Limited published evidence | Complex issues go to service representatives |
| cruise booking assistant | Holland America Line | Customer Service | Retail and Consumer | Limited published evidence | No review step stated; contact center staff were the first group in the phased rollout before employees and website customers |
| personal budgeting | Starling Bank | Banking and Insurance | Financial Services | Reported operating use | Not stated in the source |
| HR service desk | IBM | HR and Recruiting | Technology | Reported operating use | Not stated in the source |
| purchase order processing | StarLink (Infinigate Group) | Finance and Accounting | IT Services and Staffing | Reported operating use | Not stated in the source |
| clinical policy digitization | Cohere Health | Research and Regulated Documentation | Healthcare and Life Sciences | Reported operating use | Experts author skills, data science approval gates production release, and analysts annotate outputs after release |
| mortgage self-service | Rocket Companies | Banking and Insurance | Financial Services | Reported operating use | Uncertainty and keyword triggers escalate the customer to a human banker |
| network diagnosis | One New Zealand | IT Operations | Telecom, Media and Education | Reported operating use | Most VoLTE action paths retain human review, and One NZ describes human accountability |
| property due diligence | Orbital Witness | Legal | Legal and Professional Services | Limited published evidence | Legal engineers develop prompts and assess expected outputs in test environments |
| sales email CRM | Anthropic | Sales and Marketing | Technology | Reported operating use | No review step stated; sales employees describe email drafting and CRM automations, and one representative built a dashboard and pipeline |
| legal intake review | Anthropic | Legal | Technology | Reported operating use | No review step stated; an Anthropic lawyer built the automations that scan risks, triage legal questions and prepare initial reviews of public-facing copy |
| candidate sourcing | Anthropic | HR and Recruiting | Technology | Reported operating use | No review step stated; a recruiter uses Claude for sourcing, application tracking and analysis, and spends more attention on interview design and applicant conversations |
| purchase request review | Anthropic | Finance and Accounting | Technology | Reported operating use | No review step stated; a finance manager built the agent that reviews purchase requests |
| well report extraction | Petoro and SLB | Research and Regulated Documentation | Manufacturing and Energy | Limited published evidence | Not stated in the source |
| merchandising analysis | Walmart | Sales and Marketing | Retail and Consumer | Reported operating use | Merchants use the assistant as decision support; autonomous purchasing, pricing or assortment changes are not established by the source |
| model porting and optimization | Deepgram and AWS Forward Deployed Engineering | Software Engineering | Technology | Reported operating use | Engineers check layer outputs against a CPU reference and end to end word error rate against Deepgram's targets |
| investment memo review | Terminal X | Banking and Insurance | Financial Services | Reported operating use | Analyst corrections feed later instructions, and evaluator judgments are compared with senior human review |
| cloud incident investigation | Eon | IT Operations | Technology | Reported operating use | Engineers investigate alongside the SRE agent; production access is read only, approved just in time and audited |
| complaint triage | Octave | Research and Regulated Documentation | Healthcare and Life Sciences | Reported operating use | The quality handler makes the final triage decision after the assistant prepares the brief and recommended priority |
| contract obligation tracking | Birlasoft | Legal | IT Services and Staffing | Reported operating use | Delivery staff retain oversight of the agents' work |
| vulnerability patch verification | PHP Foundation and PHP security team | Security | Technology | Reported operating use | People make every decision and work on every patch |
| test failure triage | Rescale | Software Engineering | Technology | Reported operating use | Rescale engineers validate the automated first pass classifications |
| candidate sourcing | Blue Castle Agency | HR and Recruiting | IT Services and Staffing | Reported operating use | No review step stated; recruiters handle candidate conversations |
| welcome email workflow | Vetano | Sales and Marketing | Technology | Reported operating use | The founder reviews a test email before workflow changes go live |
| chip verification workflows | Unnamed semiconductor company | Software Engineering | Manufacturing and Energy | Reported operating use | No review step stated; customer verification engineers helped define evaluations alongside the embedded OpenAI engineers |
| agent containment testing | OpenAI | Security | Technology | Reported operating use | Security and infrastructure teams reproduce findings and verify configuration repairs |
| commercial loan process | Customers Bank | Banking and Insurance | Financial Services | Limited published evidence | No review step stated; an April announcement proposed bankers focus on relationships and deal structure |
| analytics failure investigation | Bayer Crop Science | Research and Regulated Documentation | Healthcare and Life Sciences | Reported operating use | No review step stated; researchers ask about workflow failures and results through the natural-language layer |
| vulnerability repair | Security | Technology | Reported operating use | Engineers review and merge changes, SRE teams monitor canary deployment, and product teams control precision and recall priorities | |
| vulnerability review | Cloudflare | Security | Technology | Limited published evidence | Cloudflare team review precedes sharing proposals with customers, who retain control over implementation |
| restaurant cash reconciliation | Ted Taylor, independent restaurant franchisee | Finance and Accounting | Retail and Consumer | Reported operating use | The owner reviews the reconciliation results before implementing anything |
| merchandise recommendations | Fanatics | Sales and Marketing | Retail and Consumer | Reported operating use | Fanatics reviewed results, supplied feedback and approved the initial product before production rollout |
| database migration | The New York Times | Software Engineering | Telecom, Media and Education | Reported operating use | NYT engineers own the migration design, quality assurance, rollback plan and final production sign off |
| cross-repo security review | CodeRabbit | Security | Technology | Reported operating use | Users select which findings get exploit verification, and humans review and approve any drafted fix |
| batch record review | Enveda | Research and Regulated Documentation | Healthcare and Life Sciences | Reported operating use | People decide whether advisory findings require action, and the quality unit retains batch disposition |
| billing discrepancy review | Telstra | Finance and Accounting | Telecom, Media and Education | Reported operating use | Human validation remains part of the process, and cases are prepared for human review |
| network config review | Microsoft Digital | IT Operations | Technology | Limited published evidence | Engineers review recommendations and approve proposed actions |
| technician briefing | TK Elevator | Field Service and Logistics | Manufacturing and Energy | Reported operating use | Technicians inspect equipment and decide the physical work; analysts and technical specialists interpret monitoring patterns |
| energy model customization | Exowatt | Software Engineering | Manufacturing and Energy | Limited published evidence | Engineers curate decisions and preserve labels and numbers when expanding training examples |
| IT service requests | Unnamed enterprise customer reported by Redpanda | IT Operations | Technology | Reported operating use | People retain incident closure and existing approvals; employee identity is retained for writes |
| voice support | Anker | Customer Service | Retail and Consumer | Reported operating use | Retell's platform handoff briefs a representative and confirms readiness before connecting the caller; Anker's exact escalation setup is not detailed |
| retail team debriefs | Credit Saison | Sales and Marketing | Financial Services | Reported operating use | No review step stated; retail teams use a custom GPT to summarize daily meeting audio and keep their regular debrief |
| engineering agent coordination | Deel | Software Engineering | Technology | Reported operating use | Engineers initiate work and answer questions, and agent permissions follow the initiating person |
| incident investigation | InEight | IT Operations | Technology | Reported operating use | Engineers validate investigations in sandboxes and retain approval; disruptive operations are blocked |
| NDA playbook review | Legora | Legal | Legal and Professional Services | Reported operating use | Lawyers review and interpret agent findings during acquisition diligence; the NDA approval procedure is not disclosed |
| disaster recovery evidence | Nubank | IT Operations | Financial Services | Reported operating use | The agent generates a report for engineers; the source does not say who accepts the recovery conclusions |
| transactional diligence | Ropes & Gray | Legal | Legal and Professional Services | Limited published evidence | Lawyers inspect source documents and interim summaries before delivering advice to clients |
| AP exception resolution | Varick AP pattern (customer unnamed) | Finance and Accounting | Cross-Industry | Limited published evidence | Discrepancies are routed to people, who retain transaction approval and payment release |
| shared skill maintenance | Varick shared-skill patterns (customers unnamed) | Software Engineering | Cross-Industry | Limited published evidence | The process owner defines correct outcomes, reads failures and approves changes, and human diagnosis and domain judgment are retained |
| startup finance model dashboards | Unnamed startup (reported by Alex Lieberman) | Finance and Accounting | Technology | Limited published evidence | No review step stated; a finance leader prepares the model, dashboards, QBR materials and compensation scenarios |
| security questionnaire | Box product demonstration | Sales and Marketing | Technology | Limited published evidence | Questions the approved policy documents do not support are routed for human review |
| RFP contract agent experiments | Box internal experiments | Sales and Marketing | Technology | Limited published evidence | Not stated in the source |
| content strategy in Google Docs | Unnamed agency (reported by Mike Fishbein) | Sales and Marketing | Legal and Professional Services | Limited published evidence | Open questions are tagged for a strategist or client, and changes arrive as suggestions in the document |
| cash application | Varick cash-application pattern (customer unnamed) | Finance and Accounting | Cross-Industry | Limited published evidence | Not stated in the source |
| cold call practice | Unnamed client (reported by Mike Fishbein) | Sales and Marketing | Cross-Industry | Limited published evidence | No review step stated; sellers practice cold calls and the AI coach judges when a seller is ready for real leads |
| hiring outreach | Hiring-outreach workshop example (Alex Lieberman) | HR and Recruiting | Cross-Industry | Limited published evidence | People set criteria and permitted sources, review AI research, choose contacts, and explicitly approve outreach before AI sends it |
| freight billing exceptions | Worldwide Express | Finance and Accounting | Logistics and Transport | Reported operating use | Reviewers approve agent recommendations, moving from individual to bulk approval; exact approval thresholds are undisclosed |
| synthetic evaluation | Microsoft | Software Engineering | Technology | Reported operating use | Not stated in the source |
| content platform migration | All England Lawn Tennis and Croquet Club (Wimbledon) | Software Engineering | Telecom, Media and Education | Reported operating use | Not stated in the source |
| entity close reconciliation | Varick close and reconciliation pattern | Finance and Accounting | Cross-Industry | Limited published evidence | Not stated in the source |
| variance commentary | Varick variance-commentary illustration and prescription | Finance and Accounting | Cross-Industry | Limited published evidence | No review step stated; an FP&A analyst creates the commentary skill and a VP's preferred format guides the output |
| vendor onboarding | Varick vendor-onboarding prescription (customer unnamed) | Finance and Accounting | Cross-Industry | Limited published evidence | Not stated in the source |
| controls and audit evidence | Varick controls-and-audit prescription (customer unnamed) | Finance and Accounting | Cross-Industry | Limited published evidence | No review step stated; auditors are described as recipients of the collected evidence |
| auto-insurance claim recommendation | Jerry Liu and LlamaIndex instructional example | Banking and Insurance | Financial Services | Limited published evidence | Not stated in the source |
| after-hours insurance intake | Ken Reiners insurance-intake recipe (agency hypothetical) | Banking and Insurance | Financial Services | Limited published evidence | A licensed person reads, edits if needed and sends ordinary replies; claim or loss messages trigger an on-call alert |
CSV columns: short_work, company, function, industry, evidence, human_checkpoint, source_url, source_date, curation_date, nuvepro_url, license. The human checkpoint states where a person reviews or decides according to the source. “No review step stated” means the source names a person or role in the work but no review or approval step; “Not stated in the source” means it names neither.