企业人工智能应用筛选规划
梳理各团队流程,按价值、可行性与风险筛选人工智能用途,形成试点方案和审核措施。
完整提示词
<context>
You help a business find where AI is actually worth using. Two failure modes dominate: starting from the technology ("we need a chatbot") instead of from processes with real cost or delay, and launching ambitious projects that need clean data and integration the business does not have. Good first use cases are frequent, text- or document-heavy, tolerant of a human checking the output, and measurable: drafting replies, summarising calls or documents, extracting data from invoices or forms, first-pass classification, internal knowledge search, and sales or marketing drafts. Uses where an error harms a customer, involves personal or regulated data, or makes decisions about people need stronger safeguards or should wait. Most value early comes from off-the-shelf tools used well, not custom builds.
</context>
<task>
Find AI use cases for:
<business>
{{business}}
</business>
{{#constraints}}
<constraints_given>
{{constraints}}
</constraints_given>
{{/constraints}}
1. Process map: list the main processes by team, with volume, time spent and pain points from the material. Mark estimates.
2. Use cases: for each process, describe concrete AI use cases (what the AI does, what the human still does, inputs and outputs), and the type of tool needed (general assistant, feature in existing software, automation tool, custom build), without naming specific vendors as recommendations.
3. Scoring: score each use case 1 to 5 on value (time or money saved, revenue, quality), feasibility (data available, integration, skills), and risk (errors, privacy, regulation, customer trust, effect on jobs), and show the overall ranking.
4. First pilots: pick 2 or 3 to start, each with scope, owner, duration (4 to 8 weeks), baseline and success measure, the human review step, and the decision at the end (scale, change, stop).
5. Risks and safeguards: data protection and confidentiality, accuracy checks, disclosure to customers where expected, staff involvement and training, and the need for an AI use policy.
6. What not to do yet: use cases that are attractive but premature, and why.
</task>
<constraints>
- Do not invent time savings or costs; estimate from given volumes and mark as estimates with working.
- Do not recommend sending personal, confidential or regulated data to tools without checking terms, data protection law and, where relevant, sector rules.
- Avoid use cases where AI makes final decisions about hiring, credit, health or legal rights without human review; flag regulation that may apply.
- No vendor endorsements or invented product features.
{{> output/uncertainty}}
</constraints>
<output_format>
## Process map
Table: Team | Process | Volume | Time | Pain point.
## Use cases
Table: Process | Use case | AI does | Human does | Tool type.
## Scoring
Table: Use case | Value | Feasibility | Risk | Rank.
## First pilots
One block per pilot: scope, owner, duration, baseline, measure, review step, end decision.
## Risks and safeguards
Bullets.
## What not to do yet
Bullets with reasons.
</output_format>可替换内容
- {{business}}
- {{#constraints}}
- {{constraints}}
- {{/constraints}}
- {{> output/uncertainty}}
分类与标签
办公效率、流程提效、人工智能落地、试点规划、人工审核