With the same access to AI, what do top talents rely on to reap the maximum dividends?
Many people assume that AI will close performance gaps, but surveys show that the biggest dividends brought by AI usually flow to top talents. They not only use AI to save time, but also actively explore tools and treat AI as a thinking partner. Enterprises can gradually adjust talent screening, post setting, tool selection and permission arrangement to adapt to such new changes.
Generative AI has spawned two diametrically opposed views centered on which practitioners can benefit the most from it. One view holds that AI will narrow the performance gap: it can be directly embedded into workflows to provide templates, ready-made answers and guidance for average-performing employees. The other view argues that AI will widen the gap — after routine but necessary work is automated, top employees can free up more time to engage in high-value work, where they already have inherent advantages.
When looking at a single task alone, both high-performing and average-performing employees may benefit. But the more critical question is, when placed in real job positions (where employees have to handle a full set of different tasks at the same time), which of the two effects will prevail in combination. The answer will directly affect corporate recruitment, training, team building, and various management systems designed for the existing talent structure.
Two of the authors of this article are technology-focused economists, and the third is the CEO of Terret, a company dedicated to helping organizations implement AI in their revenue teams. To study the impact of AI in leading enterprises, we focused our research on sales positions. Sales roles are ideal for understanding how AI affects performance: they include both repetitive work and work that relies heavily on judgment. The results of this study have not been released through other channels yet. We interviewed 8 revenue-focused executives from the CRO Advisory Board of Bain Capital Ventures, held more than a dozen advisory seminars with sales leaders, and conducted a questionnaire survey on 236 sales practitioners from Stripe and Checkr.
One of our core conclusions: AI is helpful to all employees on the whole, but at least in cutting-edge enterprises that actively embrace AI, the biggest dividends usually flow to top employees. This article will break down the reasons behind this. In short: while AI can certainly save time, more importantly, it amplifies the inherent advantages that top employees already have. To achieve this, managers need to address four key issues: how to recruit and interview talents suitable for the AI era; how to redesign entry-level positions to keep them as a valid talent pipeline; which AI capabilities should be developed in-house and which ones should be directly procured; and how to strike a balance — centrally manage background information while delegating the authority for exploration and innovation. Although our research focuses on sales, this core contradiction (whether AI narrows or widens the skill gap) is very likely to recur in a large number of positions. As long as a job mixes routine tasks and judgment-intensive work, whether it is consulting, customer management, or a broader range of knowledge-based work, the same problem will arise.
The Rise of Self-Driven Explorers
Why can top employees gain more benefits from AI?
The logic we initially envisioned was simple: AI takes over repetitive work, and top employees use the saved time to connect with more leads, maintain more customer accounts, and carry out more customer communications. The survey data also confirms this to a certain extent. In the questionnaire, sales representatives with better performance do save more time than their underperforming colleagues; managers reported that after the introduction of AI, 44% of top salespeople in the top 25% were assigned higher-value tasks, while among the bottom 25% of employees, this proportion was only 12%.
But there is another more core mechanism at play: top employees are more familiar with the usage of AI tools, and can even build tools by themselves; they use AI throughout the entire transaction cycle, not just to solve scattered tactical problems. In addition to saving them time on research, content creation, and document sorting, they treat AI as a thinking partner to refine strategies, review past communications, predict customer objections, and practice negotiation scripts. This more comprehensive application approach helps them improve transaction closing rates. Managers reported that the proportion of top employees who achieved improved work quality is 16 percentage points higher than that of the bottom tier, which means that performance growth does not come at the expense of customer experience.
Many people may think that such auxiliary tools for pre-preparation should be more helpful to underperforming employees. For example, some studies show that if AI suggestions pop up automatically on the screen when customer service agents are on calls, average-performing agents will benefit significantly. But the sales scenario is completely different: sales representatives need to actively select tools, design workflows, and adapt to technology suites with extremely fast iteration speeds and continuous changes.
AI literacy is exactly the core that determines the above performance differences. In the questionnaire, respondents rated their own AI literacy and the benefits they get from AI on a scale of 1 to 5. For every 1-point increase in literacy, the benefits perceived by respondents increased by an average of 0.71 points. The AI literacy mentioned here is not simply trusting AI (the degree of trust has almost no correlation with benefits). It refers to: being able to use existing AI tools, quickly get started with new tools, identify AI output content, troubleshoot errors, and know when humans must control the final decision. This distinction is crucial — AI tools are updated so fast that static training and traditional learning methods cannot cover all the emerging new usages.
Top employees will not wait for the company to arrange training or demonstration courses. They are active self-driven explorers, which gives them more opportunities to reap AI dividends. Such people gradually build up AI literacy through trial and error, rapid iteration, peer comparison, feedback communication, and mutual experience sharing. From the feedback of managers, several common factors that contribute to success are: executives leading by example, encouraging trial and error, knowledge sharing among peers, and benchmark users providing practical and actionable cases.
There are many examples in the interviews: top employees will continue to find ways to improve their skills outside the company's official training. For example, a salesperson at verification platform Checkr built a negotiation intelligence assistant by himself before such auxiliary tools became popular. Lindsey Scrase, Chief Operating Officer of Checkr, said that this creative usage that fits customer scenarios is exactly where top salespeople stand out from others.
Scrase introduced that with the continuous evolution of AI tools, Checkr's sales staff can now generate highly personalized prospect messages, build workflows to manage a large number of customer resources, and capture customer signals; all these combined can effectively improve business conversion and customer experience. A similar case also appeared in fintech company Brex: an employee developed a practical AI tool over the weekend, which was later widely promoted internally by the company.
Rethinking Recruitment and Talent Pipeline
If people who actively explore on their own are more likely to stand out, enterprises must focus on screening and cultivating this trait during recruitment and post design, rather than just checking whether employees can "use AI". Kevan Yalowitz, Global Industry Lead for Software & Platform at Accenture, who is in charge of sales-related segments, mentioned that top talents can quickly adapt to the fast-changing AI environment, while others need structured training and continuous support.
Therefore, screening for learning adaptability is an indispensable part of recruitment. Managers should not only ask candidates "Can you use AI", but ask how they learn AI. High-quality interviews need to dig deep into these questions: Has the candidate ever built or optimized an AI workflow? Has he ever compared different AI outputs and identified AI errors? Has he ever used AI to prepare for communication with specific customers? Or has he ever chosen not to use AI because the customer scenario is highly sensitive and high-risk?
Such questions shift the focus of assessment from familiarity with tools to curiosity and autonomous learning ability, helping enterprises find candidates who are willing to continuously optimize their work methods and can clearly explain why new methods can improve results. There is an inspiring interview idea from an executive in the interview: he will ask candidates to "share the AI interaction records you used to prepare for this interview".
In addition to recruitment, managers also need to redesign entry-level positions — these positions were originally springboards for employees to move up to higher ranks. The position of Sales Development Representative (SDR) clearly exposes this problem. Most of the lead mining and lead nurturing work of SDR is repetitive and follows fixed rules, which is naturally suitable for AI automation. But it is also a typical starting point for a sales career and the most important talent source for account specialist positions.
Faced with this contradiction, Garrett Marker, Chief Revenue Officer of Brex, chose to restructure the position: automate 20%–40% of manual work, so that motivated SDRs can use the saved time to independently follow up on full-cycle transactions. If SDRs successfully close a certain number of full-cycle orders while completing their original targets, they can be directly promoted to account specialists. The enterprise did not fully automate this position, but retained space for entry-level employees to create value through their own efforts and creativity, so as to distinguish two types of people: one is the future top employees who are empowered by AI and keep growing; the other is people who completely rely on AI and hand over decision-making power to tools. After the transformation, this position has attracted higher-quality candidates, internal promotions have increased, and the onboarding speed of new employees has also accelerated significantly.
How to Choose Technical Solutions: In-house Development or Procurement
The advantages of "self-driven explorers" are not only personal traits, but also depend on what tools salespeople can get and whether they can flexibly combine tools, which puts forward high requirements for the enterprise's own technology stack. The value of AI is closely related to the tools selected. All the enterprises we interviewed are streamlining suppliers, reconstructing interactive interfaces, and carefully judging which work links need to rely on the enterprise's exclusive business data.
In the past, salespeople had to switch back and forth between a bunch of disconnected tools: Salesforce, Gong, Outreach, ZoomInfo, spreadsheets, Slack, sales resource centers, and various dashboards. Now managers are building a more coherent operation layer, connecting multi-system data (CRM records, call transcripts, product usage data, contract information, public information) to the same interface to form a complete business background information. For example, Checkr evaluated about 20 service providers, built an AI control center layer, and integrated multiple systems commonly used by sales staff into a unified window.
Should enterprises directly procure off-the-shelf tools or develop them in-house? Christopher specifically discussed this issue in another article on *Harvard Business Review*. For tools for general tasks that are mature in the market and do not require the support of the enterprise's exclusive business information, procurement is prioritized. Eileen O'Mara, former Chief Revenue Officer and current Vice Chairman of Stripe, mentioned that the conversational intelligence platform Gong does a great job in call recording and recommending next actions, and Stripe will not develop such functions from scratch on its own. But if the task involves sensitive data, complex product logic, and industry-exclusive business rules, Stripe chooses to develop it in-house.
Brex has a similar idea: off-the-shelf tools can handle general tasks, such as extracting notes and entering them into CRM fields; what really brings differentiated competitiveness comes from exclusive data and business logic. Jeanne Grosser, Chief Operating Officer of Vercel, told us that for agents to perform well, they need a well-structured data layer that contains the enterprise's own definitions of customers, customer segmentation, and potential opportunities.
The above cases all come from cutting-edge enterprises with strong AI capabilities and sufficient resources that can develop tools in-house as needed. Other companies need to make more careful considerations when making in-house development/procurement decisions: internal technical reserves, project complexity, implementation risks, and the time required to see returns. Yalowitz from Accenture mentioned that when evaluating in-house development solutions, it is necessary to calculate the total cost of ownership, including platform service fees, token costs, development manpower and subsequent maintenance costs, and each case needs to be discussed separately.
Centrally Manage Background Information, Delegate Permission for Exploration and Innovation
Top employees can tap into AI usages that the central team cannot foresee in advance, which leads to the second operational question: which permissions should be centralized and which should be delegated?
The survey results support a hybrid model: the background information layer is centrally managed, and full authority is granted to the innovation and exploration link.
Enterprises should centrally manage the background information layer: data, permissions, approved knowledge bases, evaluation standards, stably launched agents, tool lists, and CRM-related security constraints. The effectiveness and compliance of agents rely on clean and unified data, consistent business definitions, and stable risk control mechanisms. The risks fed back by salespeople also confirm the necessity of centralized management: contradictory conclusions from different tools, errors in AI output, outdated knowledge base information, and the need to retain manual review for important scenarios.
Exploration and innovation should be delegated to front-line staff. Top employees and heavy users often can find usages that fit specific customer scenarios, which are difficult for the headquarters team to quickly identify and implement. The questionnaire shows that high-performing teams will build lightweight tools by themselves, hold AI sharing sessions every week, exchange prompts with each other, and use natural language development tools like Lovable to quickly make prototypes; after the prototypes are verified to be effective, they will be handed over to the central team for polishing and promoted across the company.
This hybrid model allows front-line staff to discover valuable AI applications first; after verification, they are productized: connected to the system, security is strengthened, indicators are established, documents are retained, supporting training is provided, and duplicate tools are phased out. Leading enterprises are generally doing three things: centrally managing business background data, delegating exploration permissions, and promoting and productizing verified high-quality tools and solutions.
AI will not simply replace sales jobs, nor will it automatically flatten everyone's performance gaps; for individuals and enterprises that build supporting systems based on their own advantages, AI can bring significant gains far exceeding the average level. Whether productivity can be improved depends crucially on whether salespeople and managers can convert the saved time into a deeper understanding of customers, more mature transaction strategies, and more reasonable human judgment. Therefore, the core work of managers is to clarify the boundaries: which links are handed over to AI for automation, which are enhanced with AI assistance, and which work must retain human leadership.
Keywords: #AI
Christopher Stanton, Justin Shriber, Simon Rudat | Text
Christopher Stanton is a professor of Entrepreneurship Management at Harvard Business School. Justin Shriber is co-founder and CEO of Terret. Simon Rudat is a PhD candidate in Entrepreneurship and Innovation at the Otto Beisheim School of Management in Germany.
Zhou Qiang | Editor
This article is from the WeChat official account "Harvard Business Review" (ID: hbrchinese), author: HBR-China, published with authorization from 36Kr.