
创业过程中,马克・洛尔向来信奉数字的力量。在他最新创办的Wonder公司,这套理念进一步延伸到晋升环节,他开始借助AI算法决定晋升人选。
在最近一期《Term Sheet》播客节目中,这位亿万富翁企业家告诉《财富》的艾莉・加芬克尔,这家食品科技公司正在使用一套由AI驱动的绩效管理系统来决定员工能否得到晋升,同事提交的评分也是其中重要的参考依据。
洛尔解释说,该公司每半年会安排至少12名同事对每名员工的绩效、行为表现和领导力进行评分,而后将这些评分连同书面反馈一并输入AI,由AI生成绩效报告。此外,该公司还设置了一项名为“超出替代者的价值”(value above replacement,简称 VAR)的指标,用于衡量以同级员工替代某名员工的难度。
“AI会结合VAR评分与绩效管理评分综合判断一名员工是否应该获得晋升,”洛尔说,“所以这套系统非常客观。”
这套系统还能算出员工升职前应在某个岗位上工作多长时间。洛尔解释说,如果管理者能提出有说服力的理由,证明算法遗漏了某些要素,也可以推翻模型的建议,不过这种情形越来越少,而且与模型的分歧本身也会变成数据,用来改进模型。
“每个考核周期,总有那么几次我们和AI模型的意见不一致,”洛尔说,“但模型现在越来越聪明,这样的例外也一次比一次少。”
洛尔认为,把更多决策权交给标准化系统,能削弱个人偏见的影响,甚至减少对女性和少数族裔员工的潜在歧视,从而让晋升变得更公平。
“该系统在某种程度上纠正了这些问题,”他说。
这套AI系统很符合洛尔长期以来“用数字处理复杂商业决策”的习惯。在去年《财富》杂志的一篇人物特写中,他的叔叔乔・洛尔回忆道,洛尔十几岁时就喜欢在赛马中玩“对冲下注”,他从不把全部筹码都押在自己最看好的某匹马上,而是小额分散投注好几匹来提高胜算。几十年过去,这种本能仍是他经营公司的核心策略。
乔・洛尔在向《财富》杂志谈起洛尔的行事风格时说:“他到最后总会回到百分比和概率上来,想办法让自己赢面更大。不管是卖螺丝、卖小商品还是卖汉堡,他都是这样。”
在Wonder,就连组织架构图都按颜色分级。洛尔告诉《财富》,该公司的不同职位对应不同颜色,仿照跆拳道腰带的等级,从白色、黄色一路升到棕色、黑色。他只要扫一眼这张彩色架构图,就能看出Wonder把高级人才安排在了哪些位置,以及有多少低层员工向高层员工汇报。
Wonder还实行透明的薪酬制度,员工都能看到其他同事的薪资水平。
Wonder的自动化历程
洛尔以AI辅助晋升决策,恰逢Wonder规模急剧扩大、日益倚重自动化之际。
今年夏天,Wonder以90亿美元估值完成逾6.5亿美元融资,至此,其自2018年创立以来的累计融资额已达约30亿美元。
洛尔在接受《财富》独家采访时表示,这家在美国东部10州经营135家美食广场的公司将“为明年初上市做好充分准备”。
洛尔还在推动Wonder厨房的自动化。他在今年6月《财富》科技头脑风暴大会(Fortune Brainstorm Tech)上表示,一套自动碗餐制作系统每小时最多可做500份碗餐,而人工每小时最多只能做约45份。(财富中文网)
译者:梁宇
审校:夏林
创业过程中,马克・洛尔向来信奉数字的力量。在他最新创办的Wonder公司,这套理念进一步延伸到晋升环节,他开始借助AI算法决定晋升人选。
在最近一期《Term Sheet》播客节目中,这位亿万富翁企业家告诉《财富》的艾莉・加芬克尔,这家食品科技公司正在使用一套由AI驱动的绩效管理系统来决定员工能否得到晋升,同事提交的评分也是其中重要的参考依据。
洛尔解释说,该公司每半年会安排至少12名同事对每名员工的绩效、行为表现和领导力进行评分,而后将这些评分连同书面反馈一并输入AI,由AI生成绩效报告。此外,该公司还设置了一项名为“超出替代者的价值”(value above replacement,简称 VAR)的指标,用于衡量以同级员工替代某名员工的难度。
“AI会结合VAR评分与绩效管理评分综合判断一名员工是否应该获得晋升,”洛尔说,“所以这套系统非常客观。”
这套系统还能算出员工升职前应在某个岗位上工作多长时间。洛尔解释说,如果管理者能提出有说服力的理由,证明算法遗漏了某些要素,也可以推翻模型的建议,不过这种情形越来越少,而且与模型的分歧本身也会变成数据,用来改进模型。
“每个考核周期,总有那么几次我们和AI模型的意见不一致,”洛尔说,“但模型现在越来越聪明,这样的例外也一次比一次少。”
洛尔认为,把更多决策权交给标准化系统,能削弱个人偏见的影响,甚至减少对女性和少数族裔员工的潜在歧视,从而让晋升变得更公平。
“该系统在某种程度上纠正了这些问题,”他说。
这套AI系统很符合洛尔长期以来“用数字处理复杂商业决策”的习惯。在去年《财富》杂志的一篇人物特写中,他的叔叔乔・洛尔回忆道,洛尔十几岁时就喜欢在赛马中玩“对冲下注”,他从不把全部筹码都押在自己最看好的某匹马上,而是小额分散投注好几匹来提高胜算。几十年过去,这种本能仍是他经营公司的核心策略。
乔・洛尔在向《财富》杂志谈起洛尔的行事风格时说:“他到最后总会回到百分比和概率上来,想办法让自己赢面更大。不管是卖螺丝、卖小商品还是卖汉堡,他都是这样。”
在Wonder,就连组织架构图都按颜色分级。洛尔告诉《财富》,该公司的不同职位对应不同颜色,仿照跆拳道腰带的等级,从白色、黄色一路升到棕色、黑色。他只要扫一眼这张彩色架构图,就能看出Wonder把高级人才安排在了哪些位置,以及有多少低层员工向高层员工汇报。
Wonder还实行透明的薪酬制度,员工都能看到其他同事的薪资水平。
Wonder的自动化历程
洛尔以AI辅助晋升决策,恰逢Wonder规模急剧扩大、日益倚重自动化之际。
今年夏天,Wonder以90亿美元估值完成逾6.5亿美元融资,至此,其自2018年创立以来的累计融资额已达约30亿美元。
洛尔在接受《财富》独家采访时表示,这家在美国东部10州经营135家美食广场的公司将“为明年初上市做好充分准备”。
洛尔还在推动Wonder厨房的自动化。他在今年6月《财富》科技头脑风暴大会(Fortune Brainstorm Tech)上表示,一套自动碗餐制作系统每小时最多可做500份碗餐,而人工每小时最多只能做约45份。(财富中文网)
译者:梁宇
审校:夏林
Marc Lore has built companies by trusting the numbers. At his latest startup Wonder, that philosophy now extends to deciding who gets promoted, with the help of AI.
The billionaire entrepreneur told Fortune’s Allie Garfinkle in a recent Term Sheet podcast episode the food-tech company uses an AI-powered performance management system that determines whether employees should advance through the organization based partly on scores submitted by their colleagues.
At least a dozen coworkers rate an employee on their performance, behaviors, and leadership qualities every six months, Lore explained, and these ratings, combined with written feedback, are fed into AI that generates a performance report. The company also adds another metric it calls “value above replacement,” or VAR, which measures how difficult it would be to replace an employee with someone at the same organizational level.
“Between your VAR score and your performance management score, AI basically calculates whether or not you should be promoted,” Lore said. “So it’s very objective.”
The system also calculates how long employees should remain in a position before advancing. Human management can override the model’s recommendation with a convincing argument that the algorithm missed something, Lore explained, but those cases are increasingly rare and disagreements with the model can also turn into data used to change it.
“There’s a handful of exceptions every period where we disagree with the AI model,” Lore said, but “the model is getting smarter, and there’s less exceptions every time.”
Lore argues putting more of the decision into a standardized system can make promotions fairer by reducing the role of personal bias and even potential discrimination against women and minority employees.
“This kind of corrects for that,” he said.
The AI system fits with Lore’s longstanding habit of approaching complicated business decisions with numbers. In a Fortune profile last year, his uncle Joe Lore, recalled that even as a teenager he would try to arbitrage horse-racing rather than simply cheer on one favorite, betting on multiple in smaller amounts to win. Decades later, that instinct remained central to how he approached his companies.
“It will always revert to a percent or percentage, and putting the odds in your favor,” Joe Lore told Fortune about how Lore operates. “It doesn’t matter if he’s selling screws, widgets, hamburgers.”
At Wonder, even the organizational chart has a formal ranking system. Lore told Fortune positions are represented by colors modeled after taekwondo belts, progressing from white and yellow through brown and black. Looking at the color-coded organizational chart lets him quickly assess where Wonder is putting senior talent and how many lower-level employees report to higher-level workers.
Wonder also has a transparent compensation system so employees can see what others at the company make.
Wonder’s automation track record
Lore’s use of AI in helping decide promotions is landing right as Wonder is becoming considerably larger—and increasingly leaning on automation.
Wonder raised more than $650 million at a $9 billion valuation this summer, bringing its total funding since its 2018 founding to roughly $3 billion.
Lore told Fortune exclusively the company, which operates 135 food halls across 10 East Coast states, will be “ready and prepared to go public early next year.”
Lore is also pushing automation into Wonder’s kitchens. He said at Fortune Brainstorm Tech in June an automated bowl-making system can produce as many as 500 bowls an hour, compared with up to 45 for a human worker.