[{"data":1,"prerenderedAt":1865},["ShallowReactive",2],{"blog-\u002Fblog\u002Fai-agent\u002Fai-fangchan":3,"blog-related-\u002Fblog\u002Fai-agent\u002Fai-fangchan":409},{"id":4,"title":5,"author":6,"body":7,"category":377,"cover":378,"date":379,"description":380,"draft":381,"extension":382,"faq":383,"featured":381,"image":378,"keywords":393,"meta":398,"navigation":399,"path":400,"seo":401,"sitemap":402,"stem":403,"tags":404,"updated":379,"__hash__":408},"blog\u002Fblog\u002Fai-agent\u002Fai-fangchan.md","房产中介怎么用AI提效","HNREIS",{"type":8,"value":9,"toc":359},"minimark",[10,19,23,42,46,51,62,66,77,81,92,96,107,111,119,122,211,216,220,246,249,281,284,332,335,353],[11,12,13,14,18],"p",{},"房产中介每天要写大量房源描述、记带看、回客户、跟市场，经纪人精力被重复工作占据。",[15,16,17],"strong",{},"AI能把这些\"围着数据转\"的活儿接过来，让经纪人专注带看和成交。"," 这篇讲清房产中介AI的落地场景。",[20,21,22],"h2",{"id":22},"房产中介的重复痛点",[24,25,26,30,33,36,39],"ul",{},[27,28,29],"li",{},"房源描述写不完，千篇一律。",[27,31,32],{},"客户需求匹配靠经纪人记忆和翻盘源。",[27,34,35],{},"带看后纪要整理耗时。",[27,37,38],{},"市场行情、小区数据查询慢。",[27,40,41],{},"客户跟进记录乱，复访靠提醒。",[20,43,45],{"id":44},"ai能落地的场景","AI能落地的场景",[47,48,50],"h3",{"id":49},"_1-房源描述生成","1. 房源描述生成",[24,52,53,56,59],{},[27,54,55],{},"基于房源参数（户型、面积、楼层、朝向、配套）自动生成描述文案。",[27,57,58],{},"多平台风格适配（朋友圈、贝壳、自有App）。",[27,60,61],{},"批量处理，经纪人微调即可。",[47,63,65],{"id":64},"_2-客户需求智能匹配","2. 客户需求智能匹配",[24,67,68,71,74],{},[27,69,70],{},"客户需求标签化（预算、区域、户型、通勤、学区）。",[27,72,73],{},"和房源库自动匹配、排序推荐。",[27,75,76],{},"新房源上架自动推给匹配客户。",[47,78,80],{"id":79},"_3-带看纪要","3. 带看纪要",[24,82,83,86,89],{},[27,84,85],{},"带看录音\u002F记录自动整理成结构化纪要。",[27,87,88],{},"提取客户反馈、关注点、意向度。",[27,90,91],{},"辅助跟进决策。",[47,93,95],{"id":94},"_4-市场数据问答","4. 市场数据问答",[24,97,98,101,104],{},[27,99,100],{},"\"这个小区近半年成交均价多少\"。",[27,102,103],{},"\"同区域类似户型挂牌情况\"。",[27,105,106],{},"自然语言查内部和市场数据。",[47,108,110],{"id":109},"_5-客户跟进","5. 客户跟进",[24,112,113,116],{},[27,114,115],{},"跟进记录整理、关键节点提醒。",[27,117,118],{},"复访时机建议。",[20,120,121],{"id":121},"落地的现实考量",[123,124,125,144],"table",{},[126,127,128],"thead",{},[129,130,131,135,138,141],"tr",{},[132,133,134],"th",{},"场景",[132,136,137],{},"数据依赖",[132,139,140],{},"难度",[132,142,143],{},"见效",[145,146,147,161,173,186,197],"tbody",{},[129,148,149,153,156,158],{},[150,151,152],"td",{},"房源文案生成",[150,154,155],{},"低",[150,157,155],{},[150,159,160],{},"快",[129,162,163,166,169,171],{},[150,164,165],{},"市场数据问答",[150,167,168],{},"中",[150,170,155],{},[150,172,160],{},[129,174,175,178,181,183],{},[150,176,177],{},"客户需求匹配",[150,179,180],{},"中（房源+需求数据）",[150,182,168],{},[150,184,185],{},"中期",[129,187,188,191,193,195],{},[150,189,190],{},"带看纪要",[150,192,168],{},[150,194,168],{},[150,196,185],{},[129,198,199,202,205,208],{},[150,200,201],{},"客户跟进智能",[150,203,204],{},"高",[150,206,207],{},"中高",[150,209,210],{},"中长期",[11,212,213],{},[15,214,215],{},"先从文案和数据查询切入，再做匹配和纪要。",[20,217,219],{"id":218},"ai不能替代的","AI不能替代的",[24,221,222,228,234,240],{},[27,223,224,227],{},[15,225,226],{},"带看与谈判","：核心成交环节，AI帮不上。",[27,229,230,233],{},[15,231,232],{},"客户信任建立","：买房是重决策，靠人和人。",[27,235,236,239],{},[15,237,238],{},"房源真实性核验","：AI生成描述，房源信息要人工确保真实。",[27,241,242,245],{},[15,243,244],{},"合规","：房源发布、价格、宣传要符合监管。",[20,247,248],{"id":248},"别踩的坑",[24,250,251,257,263,269,275],{},[27,252,253,256],{},[15,254,255],{},"AI生成的房源描述不核对就发","：参数错、夸大宣传有风险。",[27,258,259,262],{},[15,260,261],{},"数据不全做匹配","：房源或需求数据不全，匹配不准。",[27,264,265,268],{},[15,266,267],{},"忽视房源真实性","：AI只管写，真假仍要核验。",[27,270,271,274],{},[15,272,273],{},"过度依赖推荐","：客户偏好复杂，AI推荐只是线索。",[27,276,277,280],{},[15,278,279],{},"忽视合规","：房产宣传有监管要求。",[20,282,283],{"id":283},"成本参考",[123,285,286,299],{},[126,287,288],{},[129,289,290,293,296],{},[132,291,292],{},"方案",[132,294,295],{},"说明",[132,297,298],{},"成本量级",[145,300,301,311,321],{},[129,302,303,306,309],{},[150,304,305],{},"房源文案\u002F数据问答",[150,307,308],{},"SaaS 或 API",[150,310,155],{},[129,312,313,316,319],{},[150,314,315],{},"匹配+纪要模块",[150,317,318],{},"和内部系统打通，定制",[150,320,168],{},[129,322,323,326,329],{},[150,324,325],{},"中介智能运营平台",[150,327,328],{},"多模块 + 客户管理 + 数据",[150,330,331],{},"中高，定制",[20,333,334],{"id":334},"怎么开始",[336,337,338,341,344,347,350],"ol",{},[27,339,340],{},"找最耗时的环节（文案？匹配？纪要？）。",[27,342,343],{},"从低门槛场景试点（文案、数据问答）。",[27,345,346],{},"完善房源和客户数据。",[27,348,349],{},"逐步做匹配、纪要。",[27,351,352],{},"始终让经纪人把关关键输出。",[354,355,356],"blockquote",{},[11,357,358],{},"广州市汉诺雷斯（HNREIS）帮房产中介落地AI应用，从房源文案、客户匹配到带看纪要和市场数据问答，和内部系统打通。把你的中介业务痛点告诉我们，我们给出务实的AI方案。",{"title":360,"searchDepth":361,"depth":361,"links":362},"",2,[363,364,372,373,374,375,376],{"id":22,"depth":361,"text":22},{"id":44,"depth":361,"text":45,"children":365},[366,368,369,370,371],{"id":49,"depth":367,"text":50},3,{"id":64,"depth":367,"text":65},{"id":79,"depth":367,"text":80},{"id":94,"depth":367,"text":95},{"id":109,"depth":367,"text":110},{"id":121,"depth":361,"text":121},{"id":218,"depth":361,"text":219},{"id":248,"depth":361,"text":248},{"id":283,"depth":361,"text":283},{"id":334,"depth":361,"text":334},"ai-agent",null,"2024-07-28","房产中介房源描述难写、客户匹配靠经验、带看记录乱。AI能帮做房源文案生成、需求匹配、带看纪要和市场分析，本文讲清落地场景与边界。",false,"md",[384,387,390],{"q":385,"a":386},"AI能帮房产中介做什么？","几个方向：批量生成房源描述文案、根据客户需求智能匹配房源、带看后自动整理纪要、用自然语言查询市场数据。核心是把经纪人从重复的文字和整理工作中解放出来，把精力放在带看、谈判这类核心环节。AI不替代经纪人，是辅助工具。",{"q":388,"a":389},"AI能自动给客户推荐房子吗？","能做辅助匹配。基于客户需求（预算、区域、户型、通勤）和房源标签做筛选和排序，比纯人工翻盘源快。但买房是重决策，客户偏好常说不清楚还会变，AI推荐只是线索，最终仍需经纪人沟通判断。推荐质量依赖房源和需求数据的完整性。",{"q":391,"a":392},"房产中介上AI要多少钱？","轻量场景（房源文案、市场数据问答）成本较低；要做客户匹配、带看管理、和内部系统打通的，属于定制开发，几万到十几万看功能范围。建议先从最耗时的文案和数据查询切入，验证价值再扩展。",[394,395,396,397],"房产中介AI","房源描述AI","AI房产匹配","房产数字化",{},true,"\u002Fblog\u002Fai-agent\u002Fai-fangchan",{"title":5,"description":380},{"loc":400},"blog\u002Fai-agent\u002Fai-fangchan",[405,406,407],"房产","AI","提效","4eVWvzdqEjtQ7A2XdG-YXZLQi4BBxqNtJyMpObABTyU",[410,799,1148,1513],{"id":411,"title":412,"author":6,"body":413,"category":377,"cover":378,"date":774,"description":775,"draft":381,"extension":382,"faq":776,"featured":381,"image":378,"keywords":786,"meta":790,"navigation":399,"path":791,"seo":792,"sitemap":793,"stem":794,"tags":795,"updated":774,"__hash__":798},"blog\u002Fblog\u002Fai-agent\u002Fagent-kuangjia.md","主流Agent框架怎么选",{"type":8,"value":414,"toc":761},[415,422,426,429,449,456,459,596,601,604,608,620,624,631,635,638,642,645,647,679,681,737,739,756],[11,416,417,418,421],{},"企业要落地一个AI智能体，市面框架多得让人发懵：LangChain、LlamaIndex、LangGraph、AutoGen、CrewAI、Dify、Coze……选错了要么过度复杂，要么撑不住业务。",[15,419,420],{},"框架不是越新越热门越好，而是要匹配你的场景和团队能力。"," 这篇讲清怎么选。",[20,423,425],{"id":424},"先想清楚你到底需不需要框架","先想清楚：你到底需不需要框架",[11,427,428],{},"很多人一上来就要\"上框架\"，其实很多场景不需要：",[24,430,431,437,443],{},[27,432,433,436],{},[15,434,435],{},"简单问答","：直接调大模型 API，几十行代码搞定，不用框架。",[27,438,439,442],{},[15,440,441],{},"RAG 知识库问答","：用轻量库或框架的检索能力，中等复杂度。",[27,444,445,448],{},[15,446,447],{},"复杂智能体","：多步推理、调用多个工具、多轮对话有记忆、多个智能体协作——这才是框架真正发挥作用的地方。",[11,450,451,452,455],{},"需求复杂度决定框架价值。",[15,453,454],{},"简单需求硬上重框架，是典型的过度设计","。",[20,457,458],{"id":458},"主流框架对比",[123,460,461,480],{},[126,462,463],{},[129,464,465,468,471,474,477],{},[132,466,467],{},"框架",[132,469,470],{},"类型",[132,472,473],{},"核心定位",[132,475,476],{},"适合谁",[132,478,479],{},"主要局限",[145,481,482,499,515,531,547,563,580],{},[129,483,484,487,490,493,496],{},[150,485,486],{},"LangChain",[150,488,489],{},"代码框架",[150,491,492],{},"通用 LLM 应用编排，生态最全",[150,494,495],{},"有研发团队、要深度定制",[150,497,498],{},"抽象较重，早期 API 频繁变动",[129,500,501,504,506,509,512],{},[150,502,503],{},"LangGraph",[150,505,489],{},[150,507,508],{},"基于图的状态机，做可控多步 Agent",[150,510,511],{},"复杂流程、需要稳定状态控制",[150,513,514],{},"学习曲线陡",[129,516,517,520,522,525,528],{},[150,518,519],{},"LlamaIndex",[150,521,489],{},[150,523,524],{},"数据连接与检索（RAG 见长）",[150,526,527],{},"以知识库\u002F文档检索为核心",[150,529,530],{},"Agent 编排不如 LangChain 全",[129,532,533,536,538,541,544],{},[150,534,535],{},"AutoGen",[150,537,489],{},[150,539,540],{},"多智能体对话协作",[150,542,543],{},"多 Agent 角色分工场景",[150,545,546],{},"偏研究，工程化要自己补",[129,548,549,552,554,557,560],{},[150,550,551],{},"CrewAI",[150,553,489],{},[150,555,556],{},"角色化多 Agent 协作，上手快",[150,558,559],{},"快速搭多角色协作原型",[150,561,562],{},"复杂控制力有限",[129,564,565,568,571,574,577],{},[150,566,567],{},"Dify",[150,569,570],{},"低代码平台",[150,572,573],{},"可视化编排 + 知识库 + 应用托管",[150,575,576],{},"快速落地、业务人员参与",[150,578,579],{},"深度定制受平台能力限制",[129,581,582,585,587,590,593],{},[150,583,584],{},"Coze",[150,586,570],{},[150,588,589],{},"字节出品，插件生态丰富",[150,591,592],{},"快速搭建对话型应用",[150,594,595],{},"偏消费侧，企业私有化受限",[11,597,598],{},[15,599,600],{},"没有\"最好\"的框架，只有\"最匹配\"的框架。",[20,602,603],{"id":603},"选型要看的维度",[47,605,607],{"id":606},"_1-编程式还是低代码","1. 编程式还是低代码",[24,609,610,615],{},[27,611,612,614],{},[15,613,489],{},"（LangChain\u002FLlamaIndex 等）：灵活、可控、能深度集成业务系统，但要有研发投入。",[27,616,617,619],{},[15,618,570],{},"（Dify\u002FCoze）：拖拽配置、上手快、适合原型和非纯技术团队，但定制边界明显。",[47,621,623],{"id":622},"_2-单-agent-还是多-agent","2. 单 Agent 还是多 Agent",[11,625,626,627,630],{},"大部分企业场景，",[15,628,629],{},"单个能调用工具的 Agent 就够用","。多智能体协作（AutoGen\u002FCrewAI）适合确实需要多角色分工的复杂流程，比如\"研究员+执行者+审核者\"。别为了炫技上多 Agent。",[47,632,634],{"id":633},"_3-能不能私有化部署","3. 能不能私有化部署",[11,636,637],{},"涉及企业内部数据，通常要求私有化。Dify 支持私有部署，Coze 偏公有云。代码框架本身可私有化，但要自己搭运维。",[47,639,641],{"id":640},"_4-社区活跃度与稳定性","4. 社区活跃度与稳定性",[11,643,644],{},"框架迭代快是双刃剑。优先选社区活跃、有企业背书、更新稳定的，避免踩\"作者弃坑\"的雷。",[20,646,248],{"id":248},[24,648,649,655,661,667,673],{},[27,650,651,654],{},[15,652,653],{},"过度设计","：简单需求上重框架，开发和维护成本翻倍。",[27,656,657,660],{},[15,658,659],{},"被框架锁死","：把提示词、工具、业务逻辑和框架深度耦合，迁移不动。",[27,662,663,666],{},[15,664,665],{},"忽视核心","：框架只是壳，真正决定效果的是提示词、数据质量和知识库。",[27,668,669,672],{},[15,670,671],{},"盲目追新","：新框架未必成熟，企业项目要稳。",[27,674,675,678],{},[15,676,677],{},"忽视可观测","：Agent 行为不确定性高，没有日志和追踪很难调试。",[20,680,283],{"id":283},[123,682,683,693],{},[126,684,685],{},[129,686,687,689,691],{},[132,688,292],{},[132,690,295],{},[132,692,298],{},[145,694,695,705,715,726],{},[129,696,697,700,703],{},[150,698,699],{},"自研（直接调 API）",[150,701,702],{},"简单场景，研发 1-2 周",[150,704,155],{},[129,706,707,710,713],{},[150,708,709],{},"基于代码框架开发",[150,711,712],{},"中等复杂度，研发 3-8 周",[150,714,168],{},[129,716,717,720,723],{},[150,718,719],{},"低代码平台搭建",[150,721,722],{},"快速原型到中等应用",[150,724,725],{},"平台费 + 少量定制",[129,727,728,731,734],{},[150,729,730],{},"复杂多 Agent 系统",[150,732,733],{},"多角色协作 + 工具集成 + 私有化",[150,735,736],{},"较高，定制开发",[20,738,334],{"id":334},[336,740,741,744,747,750,753],{},[27,742,743],{},"明确场景和复杂度（是否真需要框架）。",[27,745,746],{},"评估团队能力（能否驾驭代码框架）。",[27,748,749],{},"对比 2-3 个候选框架，做小范围验证。",[27,751,752],{},"关注私有化、可观测、可迁移。",[27,754,755],{},"把核心资产与框架解耦。",[354,757,758],{},[11,759,760],{},"广州市汉诺雷斯（HNREIS）帮企业做AI智能体落地，从场景评估、框架选型到私有化部署和工具集成。把你的业务场景告诉我们，我们给出务实的选型建议与方案。",{"title":360,"searchDepth":361,"depth":361,"links":762},[763,764,765,771,772,773],{"id":424,"depth":361,"text":425},{"id":458,"depth":361,"text":458},{"id":603,"depth":361,"text":603,"children":766},[767,768,769,770],{"id":606,"depth":367,"text":607},{"id":622,"depth":367,"text":623},{"id":633,"depth":367,"text":634},{"id":640,"depth":367,"text":641},{"id":248,"depth":361,"text":248},{"id":283,"depth":361,"text":283},{"id":334,"depth":361,"text":334},"2024-05-12","LangChain、LlamaIndex、LangGraph、AutoGen、CrewAI、Dify、Coze 等Agent框架各有侧重。本文从企业落地视角对比主流框架，讲清选型维度，帮你按场景选对工具而不踩坑。",[777,780,783],{"q":778,"a":779},"企业做AI智能体一定要用框架吗？","不一定。简单的单轮问答或单次工具调用，直接调大模型API就能实现。框架的价值在复杂场景：多步推理、工具编排、记忆管理、多智能体协作、可观测的流程控制。需求越复杂，框架越能省去重复造轮子。但框架也有学习成本和被锁定的风险，按需选择，不要为了用框架而用框架。",{"q":781,"a":782},"LangChain和Dify怎么选？","看团队和技术诉求。LangChain是代码框架，灵活、可控、可深度定制，适合有研发团队、要和业务系统深度集成的企业；Dify是低代码平台，拖拽编排、可视化，适合快速出原型、让业务人员参与配置。要做底层定制选代码框架，要快速落地和降低门槛选低代码平台。",{"q":784,"a":785},"用框架会被技术锁定吗？","有一定风险。深度依赖某框架的特有抽象，后续迁移成本会高。建议把核心资产——提示词、工具定义、业务数据、知识库——与框架解耦，框架只承担编排层。同时优先选开源、社区活跃、更新稳定的框架，降低被单一厂商锁定或项目停更的风险。",[787,486,567,788,789],"Agent框架","AI智能体开发","框架选型",{},"\u002Fblog\u002Fai-agent\u002Fagent-kuangjia",{"title":412,"description":775},{"loc":791},"blog\u002Fai-agent\u002Fagent-kuangjia",[796,789,797],"Agent","大模型","yYDRGbvFeaR227ajWv5qiaN844ebZL6NwOSs4iOxEcE",{"id":800,"title":801,"author":6,"body":802,"category":377,"cover":378,"date":1121,"description":1122,"draft":381,"extension":382,"faq":1123,"featured":381,"image":378,"keywords":1133,"meta":1139,"navigation":399,"path":1140,"seo":1141,"sitemap":1142,"stem":1143,"tags":1144,"updated":1121,"__hash__":1147},"blog\u002Fblog\u002Fai-agent\u002Fai-agent-vs-chatbot.md","AI Agent 和传统聊天机器人有什么区别？别再被忽悠",{"type":8,"value":803,"toc":1112},[804,811,814,828,834,837,940,943,948,961,964,967,970,984,987,991,994,1026,1029,1074,1078,1081,1107],[11,805,806,807,810],{},"很多企业几年前上过\"智能客服\"，结果发现它\"一点都不智能\"——答非所问、死板的流程树、动不动就\"转人工\"。于是对\"AI 客服\"有阴影。",[15,808,809],{},"但现在的 AI Agent 和当年的聊天机器人，是完全不同的两种东西。"," 这篇文章讲清它们的本质区别，帮你判断企业该上哪种、值不值得换。",[20,812,813],{"id":813},"一句话区分",[24,815,816,822],{},[27,817,818,821],{},[15,819,820],{},"传统聊天机器人","：基于关键词匹配和预设流程图。用户说\"退款\"，它跳到\"退款流程\"；用户换个说法\"我不要了\"，它就懵了。",[27,823,824,827],{},[15,825,826],{},"AI Agent","：基于大语言模型，能理解意图、规划任务、调用工具、记住上下文。用户说\"我上周买的那个订单想退\"，它能自己查订单、判断是否符合退款规则、发起退款流程。",[11,829,830,831],{},"本质区别：",[15,832,833],{},"前者是\"执行固定脚本的程序\"，后者是\"能理解并完成任务的助手\"。",[20,835,836],{"id":836},"六维对比表",[123,838,839,850],{},[126,840,841],{},[129,842,843,846,848],{},[132,844,845],{},"能力",[132,847,820],{},[132,849,826],{},[145,851,852,863,874,885,896,907,918,929],{},[129,853,854,857,860],{},[150,855,856],{},"对话方式",[150,858,859],{},"关键词 + 流程图",[150,861,862],{},"大模型理解自然语言",[129,864,865,868,871],{},[150,866,867],{},"理解能力",[150,869,870],{},"死板，换个说法就懵",[150,872,873],{},"灵活，懂口语、省略、上下文",[129,875,876,879,882],{},[150,877,878],{},"自主性",[150,880,881],{},"被动应答",[150,883,884],{},"主动规划、多步执行",[129,886,887,890,893],{},[150,888,889],{},"工具调用",[150,891,892],{},"几乎不能",[150,894,895],{},"能查库、调 API、发通知、生成文件",[129,897,898,901,904],{},[150,899,900],{},"记忆",[150,902,903],{},"无或当次会话",[150,905,906],{},"长期记忆（记住用户历史）",[129,908,909,912,915],{},[150,910,911],{},"复杂任务",[150,913,914],{},"只能引导转人工",[150,916,917],{},"可独立完成跨系统任务",[129,919,920,923,926],{},[150,921,922],{},"知识更新",[150,924,925],{},"改流程图\u002F规则",[150,927,928],{},"喂新文档即可",[129,930,931,934,937],{},[150,932,933],{},"用户体验",[150,935,936],{},"机械、易激怒用户",[150,938,939],{},"接近真人，体验好",[20,941,942],{"id":942},"真实场景对比",[11,944,945],{},[15,946,947],{},"场景：用户问\"我上周买的那个能退吗\"",[24,949,950,956],{},[27,951,952,955],{},[15,953,954],{},"传统机器人","：抓不到关键词，回\"请问您要咨询什么\"，或跳到通用退款说明页。用户烦躁，转人工。",[27,957,958,960],{},[15,959,826],{},"：理解意图（查订单+判断退款）→ 调用订单系统查到上周的订单 → 判断商品是否符合退款规则 → 回答\"您的订单 xxx 符合 7 天无理由退款，我帮您发起申请，确认吗？\" → 用户确认 → 调用退款 API → 完成。",[11,962,963],{},"整个流程 AI Agent 自己做完，用户感觉\"这个客服真懂\"。",[20,965,966],{"id":966},"为什么传统机器人体验差",[11,968,969],{},"传统机器人的工作机制是\"if 用户说 X，then 回 Y\"。问题在于：",[24,971,972,975,978,981],{},[27,973,974],{},"用户的说法千变万化，关键词覆盖不全。",[27,976,977],{},"业务一复杂，流程图爆炸，维护噩梦。",[27,979,980],{},"没法理解上下文，每句都从头开始。",[27,982,983],{},"没法真正\"做事\"，只能\"指路\"。",[11,985,986],{},"所以传统机器人最后基本都退化成\"转人工的入口\"，用户也学会了\"上来就转人工\"。",[20,988,990],{"id":989},"ai-agent-为什么能真正解决问题","AI Agent 为什么能真正解决问题",[11,992,993],{},"AI Agent 的核心能力：",[24,995,996,1002,1008,1014,1020],{},[27,997,998,1001],{},[15,999,1000],{},"理解自然语言","：基于大模型，用户怎么说都能懂。",[27,1003,1004,1007],{},[15,1005,1006],{},"调用工具","：接业务系统（CRM、订单、知识库），能真正查数据、改数据。",[27,1009,1010,1013],{},[15,1011,1012],{},"规划任务","：把复杂目标拆成步骤，自己执行。",[27,1015,1016,1019],{},[15,1017,1018],{},"长期记忆","：记住用户历史，体验连贯。",[27,1021,1022,1025],{},[15,1023,1024],{},"知识可更新","：喂新文档即学习，不用改代码。",[20,1027,1028],{"id":1028},"企业该上哪种",[123,1030,1031,1040],{},[126,1032,1033],{},[129,1034,1035,1037],{},[132,1036,134],{},[132,1038,1039],{},"建议",[145,1041,1042,1050,1058,1066],{},[129,1043,1044,1047],{},[150,1045,1046],{},"客服量大、问题多样",[150,1048,1049],{},"AI Agent（体验好、省人工）",[129,1051,1052,1055],{},[150,1053,1054],{},"问题高度标准化、简单",[150,1056,1057],{},"传统机器人够用，或 AI Agent 低配版",[129,1059,1060,1063],{},[150,1061,1062],{},"需要跨系统办事（查单、改单、退款）",[150,1064,1065],{},"AI Agent（能调工具）",[129,1067,1068,1071],{},[150,1069,1070],{},"预算极有限、想先试水",[150,1072,1073],{},"AI Agent MVP（先接知识库做问答）",[20,1075,1077],{"id":1076},"怎么从传统机器人升级到-ai-agent","怎么从传统机器人升级到 AI Agent",[11,1079,1080],{},"不用一次性推翻：",[336,1082,1083,1089,1095,1101],{},[27,1084,1085,1088],{},[15,1086,1087],{},"先接知识库","：把产品手册、FAQ、SOP 喂给 AI，让它能准确回答咨询类问题。",[27,1090,1091,1094],{},[15,1092,1093],{},"再接业务系统","：让 AI 能查订单、查库存、发通知，真正\"办事\"。",[27,1096,1097,1100],{},[15,1098,1099],{},"加人工兜底","：AI 把握不大时转人工，持续优化。",[27,1102,1103,1106],{},[15,1104,1105],{},"逐步替换","：AI 能力稳定后，逐步下线老机器人的场景。",[354,1108,1109],{},[11,1110,1111],{},"广州市汉诺雷斯（HNREIS）提供 AI 智能客服定制，支持从传统机器人渐进式升级。告诉我们你现在的客服痛点和量级，我们帮你评估 AI Agent 能解决多少、ROI 多少。",{"title":360,"searchDepth":361,"depth":361,"links":1113},[1114,1115,1116,1117,1118,1119,1120],{"id":813,"depth":361,"text":813},{"id":836,"depth":361,"text":836},{"id":942,"depth":361,"text":942},{"id":966,"depth":361,"text":966},{"id":989,"depth":361,"text":990},{"id":1028,"depth":361,"text":1028},{"id":1076,"depth":361,"text":1077},"2024-05-26","传统聊天机器人基于关键词和流程图，只能被动应答；AI Agent 基于大模型，能理解意图、规划任务、调用工具、长期记忆。本文用对比表和真实场景讲清两者的本质区别，帮你判断企业该上哪种。",[1124,1127,1130],{"q":1125,"a":1126},"AI Agent 比传统聊天机器人贵很多吗？","一次性投入是的，但要看性价比。传统聊天机器人便宜（几千到一两万），但只能回答固定问题，用户体验差、转人工率高；AI Agent 投入更高（几万起），但能理解自然语言、调用业务系统、真正解决问题，长期看能省更多人工成本。如果客服量大，AI Agent 的回报更快。",{"q":1128,"a":1129},"我们已经有客服机器人了，要全部换掉吗？","不必全部换。可以分步：先用 AI Agent 接管高频、标准的问题（产品咨询、订单查询、售后流程），复杂问题继续走老系统或人工。我们支持渐进式升级，避免一次性推翻重来。",{"q":1131,"a":1132},"AI Agent 会不会答错或者说胡话？","会，但有办法控制。关键是 RAG（检索增强生成）——让 AI 先在企业知识库里查到准确信息再回答，而不是凭空生成。再配上人工兜底机制（AI 把握不大时转人工）和回答质量回流，准确率可以做到很高。",[1134,1135,1136,1137,1138],"AI Agent和聊天机器人区别","智能客服","AI客服机器人","传统客服机器人","AI智能体",{},"\u002Fblog\u002Fai-agent\u002Fai-agent-vs-chatbot",{"title":801,"description":1122},{"loc":1140},"blog\u002Fai-agent\u002Fai-agent-vs-chatbot",[1145,406,1146],"对比","概念","8qKIdcO08ytoPMy5eHwEUZP4TpzwR1mih2PoU1dQnm0",{"id":1149,"title":1150,"author":6,"body":1151,"category":377,"cover":378,"date":1487,"description":1488,"draft":381,"extension":382,"faq":1489,"featured":381,"image":378,"keywords":1499,"meta":1504,"navigation":399,"path":1505,"seo":1506,"sitemap":1507,"stem":1508,"tags":1509,"updated":1487,"__hash__":1512},"blog\u002Fblog\u002Fai-agent\u002Fai-caiwu-baobiao.md","AI辅助财务报表和数据分析怎么做",{"type":8,"value":1152,"toc":1468},[1153,1160,1164,1168,1179,1183,1194,1198,1209,1213,1221,1223,1249,1254,1258,1261,1307,1312,1315,1319,1330,1334,1342,1346,1354,1358,1366,1368,1415,1418,1444,1446,1463],[11,1154,1155,1156,1159],{},"财务部门每天和报表、数据打交道，重复分析多、对数据敏感度要求极高。",[15,1157,1158],{},"AI能帮财务提效，但财务数据的特殊性决定了\"怎么用\"比\"用不用\"更重要。"," 这篇讲清AI辅助财务分析的边界、做法和安全要点。",[20,1161,1163],{"id":1162},"ai在财务分析里能做什么","AI在财务分析里能做什么",[47,1165,1167],{"id":1166},"_1-报表解读与可视化分析","1. 报表解读与可视化分析",[24,1169,1170,1173,1176],{},[27,1171,1172],{},"把利润表、资产负债表、现金流量表的关键指标提炼成通俗解读。",[27,1174,1175],{},"自动生成同比、环比、趋势的文字说明。",[27,1177,1178],{},"辅助管理层快速理解\"这份报表说明了什么\"。",[47,1180,1182],{"id":1181},"_2-异常检测","2. 异常检测",[24,1184,1185,1188,1191],{},[27,1186,1187],{},"识别费用、成本、应收应付的异常波动。",[27,1189,1190],{},"标记可能的问题（如某类支出突增、账期异常）。",[27,1192,1193],{},"帮财务聚焦风险点，而不是大海捞针。",[47,1195,1197],{"id":1196},"_3-辅助预测","3. 辅助预测",[24,1199,1200,1203,1206],{},[27,1201,1202],{},"基于历史数据做现金流、收入、成本的短期预测。",[27,1204,1205],{},"结合业务假设做情景分析。",[27,1207,1208],{},"预测是辅助参考，不是承诺。",[47,1210,1212],{"id":1211},"_4-数据问答","4. 数据问答",[24,1214,1215,1218],{},[27,1216,1217],{},"用自然语言问数据：\"上季度哪个产品线毛利率最高\"。",[27,1219,1220],{},"AI转为查询，返回结果——类似 BI 工具的自然语言版。",[20,1222,219],{"id":218},[24,1224,1225,1231,1237,1243],{},[27,1226,1227,1230],{},[15,1228,1229],{},"记账与出表","：强规则、强合规，应以专业财务软件为准。",[27,1232,1233,1236],{},[15,1234,1235],{},"对外财报","：必须人工编制和审计，AI不可直接生成。",[27,1238,1239,1242],{},[15,1240,1241],{},"专业判断","：税务筹划、投资决策、合规判断，需要财务专业人员。",[27,1244,1245,1248],{},[15,1246,1247],{},"责任承担","：AI出错无人担责，最终决策和签字在人。",[11,1250,1251],{},[15,1252,1253],{},"AI是财务的助手，不是替代者。",[20,1255,1257],{"id":1256},"数据安全财务ai的红线","数据安全：财务AI的红线",[11,1259,1260],{},"财务数据敏感度极高，这是用AI最大的顾虑：",[123,1262,1263,1273],{},[126,1264,1265],{},[129,1266,1267,1270],{},[132,1268,1269],{},"风险点",[132,1271,1272],{},"应对做法",[145,1274,1275,1283,1291,1299],{},[129,1276,1277,1280],{},[150,1278,1279],{},"数据泄露",[150,1281,1282],{},"敏感数据不上公有云，或先脱敏",[129,1284,1285,1288],{},[150,1286,1287],{},"合规要求",[150,1289,1290],{},"涉及上市公司\u002F监管要求，必须私有化",[129,1292,1293,1296],{},[150,1294,1295],{},"模型幻觉",[150,1297,1298],{},"计算和分析结果必须人工复核",[129,1300,1301,1304],{},[150,1302,1303],{},"权限控制",[150,1305,1306],{},"AI应用也要做数据权限隔离",[11,1308,1309],{},[15,1310,1311],{},"核心原则：敏感财务数据优先私有化部署，脱敏后再分析。",[20,1313,1314],{"id":1314},"实施路径",[47,1316,1318],{"id":1317},"_1-先从低风险场景切入","1. 先从低风险场景切入",[24,1320,1321,1324,1327],{},[27,1322,1323],{},"用 AI 解读已公开\u002F已汇总的报表数据。",[27,1325,1326],{},"做趋势分析、异常提示等辅助工作。",[27,1328,1329],{},"不碰原始敏感凭证。",[47,1331,1333],{"id":1332},"_2-数据脱敏与隔离","2. 数据脱敏与隔离",[24,1335,1336,1339],{},[27,1337,1338],{},"分析前对客户名、金额等敏感字段脱敏。",[27,1340,1341],{},"AI 应用与核心财务系统数据隔离。",[47,1343,1345],{"id":1344},"_3-私有化部署","3. 私有化部署",[24,1347,1348,1351],{},[27,1349,1350],{},"有条件的企业部署私有化模型。",[27,1352,1353],{},"数据不出内网。",[47,1355,1357],{"id":1356},"_4-人机协作流程","4. 人机协作流程",[24,1359,1360,1363],{},[27,1361,1362],{},"AI 出初稿和提示，财务复核确认。",[27,1364,1365],{},"关键决策必须人工签字。",[20,1367,283],{"id":283},[123,1369,1370,1380],{},[126,1371,1372],{},[129,1373,1374,1376,1378],{},[132,1375,292],{},[132,1377,295],{},[132,1379,298],{},[145,1381,1382,1393,1404],{},[129,1383,1384,1387,1390],{},[150,1385,1386],{},"轻量分析工具",[150,1388,1389],{},"接入脱敏数据做报表解读",[150,1391,1392],{},"较低",[129,1394,1395,1398,1401],{},[150,1396,1397],{},"私有化AI分析",[150,1399,1400],{},"本地部署模型 + 财务数据集成",[150,1402,1403],{},"中到高",[129,1405,1406,1409,1412],{},[150,1407,1408],{},"完整智能财务中台",[150,1410,1411],{},"多模块 + 权限 + 审计 + 私有化",[150,1413,1414],{},"高，定制开发",[20,1416,1417],{"id":1417},"常见误区",[24,1419,1420,1426,1432,1438],{},[27,1421,1422,1425],{},[15,1423,1424],{},"让AI直接做账出表","：合规风险高，不可取。",[27,1427,1428,1431],{},[15,1429,1430],{},"原始财务数据直接传公有云","：泄露和合规风险。",[27,1433,1434,1437],{},[15,1435,1436],{},"盲目相信AI分析结论","：不复核直接用，可能出错。",[27,1439,1440,1443],{},[15,1441,1442],{},"忽视权限","：AI能看的数据范围要和岗位职责匹配。",[20,1445,334],{"id":334},[336,1447,1448,1451,1454,1457,1460],{},[27,1449,1450],{},"梳理可被AI辅助的重复分析工作。",[27,1452,1453],{},"评估数据敏感度和合规要求。",[27,1455,1456],{},"选私有化或脱敏方案。",[27,1458,1459],{},"从低风险场景试点。",[27,1461,1462],{},"建立人机协作和复核流程。",[354,1464,1465],{},[11,1466,1467],{},"广州市汉诺雷斯（HNREIS）帮企业搭建私有化的AI数据分析应用，在数据安全和合规前提下，让财务和业务团队用自然语言分析数据。把你的数据分析需求告诉我们，我们给出安全合规的方案。",{"title":360,"searchDepth":361,"depth":361,"links":1469},[1470,1476,1477,1478,1484,1485,1486],{"id":1162,"depth":361,"text":1163,"children":1471},[1472,1473,1474,1475],{"id":1166,"depth":367,"text":1167},{"id":1181,"depth":367,"text":1182},{"id":1196,"depth":367,"text":1197},{"id":1211,"depth":367,"text":1212},{"id":218,"depth":361,"text":219},{"id":1256,"depth":361,"text":1257},{"id":1314,"depth":361,"text":1314,"children":1479},[1480,1481,1482,1483],{"id":1317,"depth":367,"text":1318},{"id":1332,"depth":367,"text":1333},{"id":1344,"depth":367,"text":1345},{"id":1356,"depth":367,"text":1357},{"id":283,"depth":361,"text":283},{"id":1417,"depth":361,"text":1417},{"id":334,"depth":361,"text":334},"2024-06-03","AI能帮财务做报表解读、异常检测、趋势预测，但财务数据高度敏感。本文讲清AI辅助财务分析能做什么、不能做什么，以及数据安全和私有化的关键考量。",[1490,1493,1496],{"q":1491,"a":1492},"AI能直接帮我做账出报表吗？","目前不适合让AI直接生成对外财报。AI擅长的是辅助：解读已有报表、做数据透视、发现异常、辅助预测分析。记账、出表这类强规则、强合规的工作，仍应以专业财务软件和会计为准，AI作为辅助分析工具。涉及对外披露的报表必须人工复核把关。",{"q":1494,"a":1495},"财务数据这么敏感，能用云端AI吗？","要谨慎。财务数据涉及商业机密和合规要求，一般不建议直接传公有云大模型。可行做法：敏感字段脱敏后再分析、用私有化部署的模型、或只让AI处理汇总后的非敏感数据。涉及上市公司或有严格保密要求的，必须私有化部署。",{"q":1497,"a":1498},"AI做财务分析的准确率可靠吗？","对结构化数据的计算和趋势识别较可靠，但对复杂业务判断和因果分析可能出错（幻觉）。正确用法是AI出初稿和提示，财务专业人员复核确认。AI不能替代财务判断，尤其涉及投资、税务、合规决策时，必须以人为准。",[1500,1501,1502,1503],"AI财务分析","财务报表AI","AI数据分析","智能财务",{},"\u002Fblog\u002Fai-agent\u002Fai-caiwu-baobiao",{"title":1150,"description":1488},{"loc":1505},"blog\u002Fai-agent\u002Fai-caiwu-baobiao",[1510,1511,244],"财务","数据分析","Pdri7PqftoKRgTfvW1VvvY-liWyZ8b5GUMVIU9UtDsk",{"id":1514,"title":1515,"author":6,"body":1516,"category":377,"cover":378,"date":1838,"description":1839,"draft":381,"extension":382,"faq":1840,"featured":381,"image":378,"keywords":1850,"meta":1855,"navigation":399,"path":1856,"seo":1857,"sitemap":1858,"stem":1859,"tags":1860,"updated":1838,"__hash__":1864},"blog\u002Fblog\u002Fai-agent\u002Fai-canyin.md","餐饮行业怎么用AI做营销和排班",{"type":8,"value":1517,"toc":1824},[1518,1525,1528,1545,1547,1551,1562,1569,1573,1584,1588,1599,1603,1614,1618,1629,1631,1704,1710,1712,1744,1746,1800,1802,1819],[11,1519,1520,1521,1524],{},"餐饮行业人力成本高、客流波动大、点评和营销又特别耗时。",[15,1522,1523],{},"AI帮餐饮解决的不是\"做饭\"，而是那些重复、依赖经验、占用店长精力的环节。"," 这篇讲清餐饮AI的落地场景。",[20,1526,1527],{"id":1527},"餐饮的痛点",[24,1529,1530,1533,1536,1539,1542],{},[27,1531,1532],{},"排班靠店长经验，忙时人不够、闲时人过剩。",[27,1534,1535],{},"客流预测不准，备料和排班都跟着错。",[27,1537,1538],{},"大众点评、外卖平台的评价回复耗时。",[27,1540,1541],{},"朋友圈、社群营销文案写不出来。",[27,1543,1544],{},"会员复购靠拍脑袋，缺少数据支撑。",[20,1546,45],{"id":44},[47,1548,1550],{"id":1549},"_1-客流预测与智能排班","1. 客流预测与智能排班",[24,1552,1553,1556,1559],{},[27,1554,1555],{},"基于历史客流、天气、节假日、周边活动预测各时段客流。",[27,1557,1558],{},"自动生成排班建议，匹配高峰人力需求。",[27,1560,1561],{},"连锁门店可统一调度。",[11,1563,1564,1565,1568],{},"这是餐饮AI",[15,1566,1567],{},"价值最直接","的场景——直接关联人力成本。",[47,1570,1572],{"id":1571},"_2-智能推荐与连带销售","2. 智能推荐与连带销售",[24,1574,1575,1578,1581],{},[27,1576,1577],{},"根据顾客点单、时段、季节推荐搭配。",[27,1579,1580],{},"收银\u002F小程序点餐页的智能推荐位。",[27,1582,1583],{},"提升客单价。",[47,1585,1587],{"id":1586},"_3-点评分析与自动回复","3. 点评分析与自动回复",[24,1589,1590,1593,1596],{},[27,1591,1592],{},"批量分析大众点评\u002F外卖评价，提取菜品、服务、卫生的口碑趋势。",[27,1594,1595],{},"辅助生成评价回复（人工把关后发送）。",[27,1597,1598],{},"发现差评共性问题，改进运营。",[47,1600,1602],{"id":1601},"_4-营销文案生成","4. 营销文案生成",[24,1604,1605,1608,1611],{},[27,1606,1607],{},"朋友圈、社群、公众号文案快速生成。",[27,1609,1610],{},"节日活动、新品上市的推广素材。",[27,1612,1613],{},"多门店内容统一又本地化。",[47,1615,1617],{"id":1616},"_5-会员与私域","5. 会员与私域",[24,1619,1620,1623,1626],{},[27,1621,1622],{},"会员消费数据分析，识别高价值客户和流失风险。",[27,1624,1625],{},"个性化优惠券和召回。",[27,1627,1628],{},"复购预测。",[20,1630,121],{"id":121},[123,1632,1633,1645],{},[126,1634,1635],{},[129,1636,1637,1639,1641,1643],{},[132,1638,134],{},[132,1640,137],{},[132,1642,140],{},[132,1644,143],{},[145,1646,1647,1658,1669,1681,1693],{},[129,1648,1649,1652,1654,1656],{},[150,1650,1651],{},"营销文案生成",[150,1653,155],{},[150,1655,155],{},[150,1657,160],{},[129,1659,1660,1663,1665,1667],{},[150,1661,1662],{},"点评分析回复",[150,1664,168],{},[150,1666,155],{},[150,1668,160],{},[129,1670,1671,1674,1677,1679],{},[150,1672,1673],{},"客流预测排班",[150,1675,1676],{},"高（历史数据）",[150,1678,168],{},[150,1680,185],{},[129,1682,1683,1686,1689,1691],{},[150,1684,1685],{},"智能推荐",[150,1687,1688],{},"中（点单数据）",[150,1690,168],{},[150,1692,185],{},[129,1694,1695,1698,1700,1702],{},[150,1696,1697],{},"会员复购预测",[150,1699,204],{},[150,1701,207],{},[150,1703,210],{},[11,1705,1706,1709],{},[15,1707,1708],{},"先从低门槛、见效快的场景切入","（文案、点评），再逐步做需要数据的场景（排班、推荐）。",[20,1711,248],{"id":248},[24,1713,1714,1720,1726,1732,1738],{},[27,1715,1716,1719],{},[15,1717,1718],{},"数据质量差就上预测","：历史客流数据不准，预测也不准。",[27,1721,1722,1725],{},[15,1723,1724],{},"自动回复不经审核","：AI回复可能不当，引发公关问题。",[27,1727,1728,1731],{},[15,1729,1730],{},"忽视门店执行","：AI给排班建议，门店不执行等于零。",[27,1733,1734,1737],{},[15,1735,1736],{},"一上来搞大而全","：建议单点突破，验证后再扩展。",[27,1739,1740,1743],{},[15,1741,1742],{},"把AI当万能","：餐饮核心仍是产品和体验，AI是提效工具。",[20,1745,283],{"id":283},[123,1747,1748,1758],{},[126,1749,1750],{},[129,1751,1752,1754,1756],{},[132,1753,292],{},[132,1755,295],{},[132,1757,298],{},[145,1759,1760,1770,1780,1789],{},[129,1761,1762,1765,1768],{},[150,1763,1764],{},"文案\u002F点评辅助",[150,1766,1767],{},"SaaS 或 API，按量计费",[150,1769,155],{},[129,1771,1772,1775,1778],{},[150,1773,1774],{},"客流排班模块",[150,1776,1777],{},"和门店系统打通，定制",[150,1779,168],{},[129,1781,1782,1784,1787],{},[150,1783,1685],{},[150,1785,1786],{},"点餐\u002F收银集成",[150,1788,168],{},[129,1790,1791,1794,1797],{},[150,1792,1793],{},"连锁智能运营中台",[150,1795,1796],{},"多门店 + 预测 + 推荐 + 私域",[150,1798,1799],{},"高，定制",[20,1801,334],{"id":334},[336,1803,1804,1807,1810,1813,1816],{},[27,1805,1806],{},"找出最耗时的环节（排班？点评？文案？）。",[27,1808,1809],{},"从低成本场景试点（文案、点评辅助）。",[27,1811,1812],{},"积累门店数据，再做客流预测排班。",[27,1814,1815],{},"逐步扩展到推荐、会员。",[27,1817,1818],{},"始终让人把关关键输出。",[354,1820,1821],{},[11,1822,1823],{},"广州市汉诺雷斯（HNREIS）帮餐饮企业落地AI应用，从客流预测排班、点评分析到营销文案和会员私域，和小程序、收银系统打通。把你的门店运营痛点告诉我们，我们给出务实的AI落地方案。",{"title":360,"searchDepth":361,"depth":361,"links":1825},[1826,1827,1834,1835,1836,1837],{"id":1527,"depth":361,"text":1527},{"id":44,"depth":361,"text":45,"children":1828},[1829,1830,1831,1832,1833],{"id":1549,"depth":367,"text":1550},{"id":1571,"depth":367,"text":1572},{"id":1586,"depth":367,"text":1587},{"id":1601,"depth":367,"text":1602},{"id":1616,"depth":367,"text":1617},{"id":121,"depth":361,"text":121},{"id":248,"depth":361,"text":248},{"id":283,"depth":361,"text":283},{"id":334,"depth":361,"text":334},"2024-06-16","餐饮门店客流波动大、排班靠经验、点评回复耗时。AI能帮餐饮做客流预测排班、智能推荐、点评分析和营销文案，本文讲清落地场景与成本。",[1841,1844,1847],{"q":1842,"a":1843},"餐饮店有必要上AI吗？","看规模和痛点。单店小店，痛点不突出可以先不上；连锁或多门店，客流预测排班、点评批量回复、营销文案生成这类重复工作，AI能明显提效。建议从最耗时的环节切入，比如用AI生成朋友圈\u002F社群文案、辅助排班，先看到效果再扩展。",{"q":1845,"a":1846},"AI排班靠谱吗？","AI排班是基于历史客流、天气、节假日等数据预测各时段客流，再匹配人力需求，比纯靠店长经验更稳定。但它依赖历史数据质量和门店执行，适合作为排班参考，最终仍需店长结合实际情况调整。新店数据不足时预测效果有限。",{"q":1848,"a":1849},"餐饮用AI要花多少钱？","轻量场景（文案生成、点评回复辅助）用现有SaaS或大模型API，成本很低；客流预测排班、智能推荐这类要和门店系统打通的，属于定制开发，几万到十几万不等，看门店数量和集成深度。建议先从低成本场景验证价值。",[1851,1852,1853,1854],"餐饮AI","餐饮排班AI","AI营销文案","餐饮数字化",{},"\u002Fblog\u002Fai-agent\u002Fai-canyin",{"title":1515,"description":1839},{"loc":1856},"blog\u002Fai-agent\u002Fai-canyin",[1861,1862,1863],"餐饮","营销","排班","wwPJmcDR-PabsSXv_ZcpsIAe0cgabBUgBGYdFdSCdPc",1781688904748]