Two McKinsey studies on B2B sales, published in 2025 and 2026 respectively, are more revealing when read together than in isolation. The 2025 report, Unlocking Profitable B2B Growth Through Gen AI, asks which sales tasks are worth augmenting with generative AI. One year later, The Future of B2B Sales advances the discussion to a much more consequential question: how should the commercial organization itself be redesigned around agentic AI? The earlier report still uses seven discrete use cases as its unit of analysis, while the later report recombines those capabilities into five end-to-end commercial impact journeys. That shift may matter more to business leaders than any individual AI use case discussed in either report.
What Is Really Changing Is Not the Sales Tool, but the Unit by Which AI Value Is Measured
The 2025 report still reflects a conventional enterprise-software mindset: identify tasks capable of generating measurable ROI. Its seven use cases—Next-best opportunity, Next-best action, Meeting support, Proposal responder, Smart pricing, Smart research assistant, and Smart coach—roughly span the full deal cycle, from prospecting and opportunity progression to closing and seller enablement. At the time, the report found that 19 percent of B2B decision-makers had already implemented relevant GenAI use cases, while another 23 percent were in the process of doing so. The central management question remained: where should we start?
The 2026 report makes a different observation. Most enterprises are already “doing something with AI,” yet relatively few are capturing material business value. Fragmented data, weak insights, manual processes, organizational silos, and inadequate change management remain unresolved. Layering AI on top of these constraints often does little more than automate existing complexity. What distinguishes growth leaders is that they are redesigning entire workflows rather than simply adding more AI tools. Among high-growth companies, 71 percent increased AI investment by double digits year over year in 2026, compared with 25 percent among other companies.
This suggests that the fundamental unit for measuring AI value is changing. The old question was: “How many hours can this Copilot save each salesperson per year?” The more relevant question now is: “Across the entire process from detecting market signals to closing the deal, can we materially change conversion rates, sales-cycle duration, gross margin, and cost-to-serve?” AI truly enters the enterprise operating system only when the object of measurement shifts from an individual task, employee, or software seat to an end-to-end business process and its associated P&L.
The Evolution from 2025 to 2026 Is Fundamentally a Shift from Isolated Capabilities to Continuous Decision Chains
The structural difference between the two reports is striking. The exhibit on page 2 of the 2025 report separates AI into seven individual use cases. By page 4 of the 2026 report, those capabilities have been reorganized into five end-to-end journeys: reaching the right opportunities, using the right go-to-market model, delivering the right offer and pitch, getting the price right, and enabling sellers to perform consistently every time. Beneath these journeys sit four horizontal foundations: a business-led roadmap, talent and operating model, data and technology, and adoption and scaling.
| 2025: Use Case | 2026: Impact Journey | Fundamental Shift |
|---|---|---|
| Next-best opportunity + Next-best action | Reach the right opportunities | From lead identification to continuous market sensing, prioritization, and execution |
| Smart research assistant | Absorbed across multiple journeys | Information retrieval becomes a foundational capability |
| Meeting support | Right offer and pitch | From meeting preparation to an Account 360 → meeting → feedback loop |
| Proposal responder + Smart pricing | At the right price | From content generation to quoting, approvals, negotiation, and value-leakage management |
| Smart coach | Every single time | From a training tool to a continuous performance-management system |
| — | Right GTM model | Adds agent-driven account coverage and resource allocation |
The most interesting development is the apparent “disappearance” of the Smart Research Assistant. Its value has not diminished. Rather, searching, reading, synthesizing, and interpreting information have become foundational capabilities embedded across virtually every agentic workflow, making them less meaningful as a standalone application.
At the same time, the emergence of the “Right GTM model” as a first-order journey signals a broader shift. Agentic AI is moving beyond helping individual sellers and beginning to influence account coverage, resource allocation, and the design of the sales organization itself.
The Critical Capability Added by Agentic AI Is the Move from Recommendation to Execution
The 2025 report already anticipated this transition. In a next-best-action scenario, conventional AI might recommend that a salesperson send a prospect two warm-up emails. Agentic AI can go further: conduct the outreach, assess the prospect’s interest, continue the interaction based on responses, and bring a human seller into the process at the appropriate moment. In other words, the system evolves from a recommendation engine into an execution engine.
The distinction between an Agent and a traditional sales Copilot therefore goes beyond intelligence. A Copilot generally operates within a loop of:
Information → Recommendation → Human Execution
An agentic workflow increasingly operates through a continuous state cycle:
Information → Judgment → Action → Feedback Observation → Reassessment → Next Action
Only when this loop exists can the gains from multiple AI capabilities compound rather than remain isolated productivity improvements measured in minutes saved. This is why the 2026 report emphasizes integrating data, decision logic, human judgment, and AI agents within the same impact journey.
It also explains why meeting summaries, email generation, and sales-script drafting—although easy to deploy—rarely create durable competitive advantage on their own. They optimize a single action. Revenue performance, by contrast, is determined by a chain of decisions involving opportunity selection, resource allocation, pricing, customer response, and follow-up timing. If an Agent can only generate content but cannot participate meaningfully in these decision points, the enterprise has acquired a productivity tool, not a new commercial capability.
One of the Most Valuable Functions of Sales Agents Is Reallocating the Scarcity of Human Expertise
One of the fundamental constraints in B2B sales has never been that every customer lacks coverage. The deeper constraint is that the time of highly capable salespeople is scarce. Traditional organizations often allocate similar levels of human coverage to high-value and low-value accounts, causing substantial selling capacity to be consumed by low-potential opportunities.
The Agentic GTM model described in the 2026 report makes a more dynamic allocation possible. Resources can be assigned according to customer value, margin potential, and cost-to-serve. High-value accounts may receive a “seller + Agent Copilot” model, while lower-value or more transactional accounts can increasingly be handled by Virtual Account Representatives that conduct outreach, nurture responses, and identify needs autonomously.
This changes the basic economic model of the sales organization. Historically, expanding market coverage generally required hiring more SDRs, BDRs, inside-sales representatives, or account executives. In the future, additional coverage capacity may increasingly come from Agent capacity. The traditionally linear relationship between headcount and market coverage may therefore begin to weaken.
McKinsey cites implementations in which AI-powered next-best-experience programs generated revenue uplift of 5 to 8 percent while reducing cost-to-serve by 20 to 30 percent. In financial services, redesigned prospecting and relationship-management workflows supported by agentic AI produced 3 to 15 percent higher revenue per relationship manager and 20 to 40 percent lower cost-to-serve.
For enterprises, the business case for Agentic AI therefore should not be limited to the number of seller hours saved. More important questions include: How many accounts can the same sales organization cover? How many qualified opportunities can each seller manage effectively? Can the unit cost of serving lower-value accounts decline? And does this allow high-value customers to receive more human attention where judgment and relationships matter most?
AI’s Most Important Sales Capability May Be Turning Top-Seller Expertise from an Individual Asset into an Organizational Asset
B2B sales has long faced a persistent problem: the capabilities of star sellers are extraordinarily difficult to replicate. High performers synthesize customer context, product knowledge, historical relationships, market signals, and real-time feedback into effective judgment. Much of this capability remains tacit and embedded in individual experience.
One global enterprise technology company cited in the report deployed an AI sales adviser to more than 8,000 sellers. The system reduced onboarding time for new sellers from approximately three months to six weeks while making knowledge, guidance, and selling practices that had previously depended heavily on managers and specialists continuously available across the organization.
The real significance here goes far beyond knowledge Q&A. AI begins to perform a capability enterprises have historically struggled to scale: turning best practices into behavioral systems that are computable, callable, observable, and continuously improvable.
Customer insight, effective messaging, opportunity judgment, product combinations, negotiation strategies, objection handling, and follow-up cadence can gradually be extracted from the tacit expertise of top performers and converted into organizational data, rules, models, and Agent policies.
This structurally changes the meaning of sales enablement. Historically, enablement depended on courses, playbooks, training programs, and manager coaching. In the future, it may increasingly take the form of an embedded performance system that provides recommendations, simulations, feedback, and corrective guidance inside the flow of actual work. Sales development therefore shifts from “learn first, then work” toward “continuously learn while working.”
The Role of the Sales Manager Changes Fundamentally as Well
One case in the 2026 report is particularly revealing. A large construction company found significant performance variation across its sales team. Top performers demonstrated more disciplined discovery, follow-up, and closing behaviors, but these habits were not consistently replicated by the broader organization.
After deploying an AI coaching platform to analyze customer interactions, the company discovered that when discounting was discussed, sellers—not buyers—introduced the subject 70 percent of the time. It also found that conversion rates were 40 percent lower when salespeople opened by selling the brand rather than first discovering customer needs. Sellers using the system ultimately achieved a 4.5-percentage-point improvement in conversion.
Systems of this kind give sales management, for the first time, an opportunity to move from sampled supervision toward near-comprehensive observation. Historically, managers could listen to only a limited number of calls, inspect CRM records, review pipeline data, and then rely heavily on experience to diagnose problems. Agents can analyze large volumes of real interactions and identify which behaviors correlate with conversion, margin, or customer satisfaction.
The manager consequently moves from performance inspector toward performance coach, spending less time collecting and diagnosing information and more time explaining causes, handling complex situations, and driving behavioral change.
A deeper transformation follows: previously unobservable portions of the sales process become data. CRM systems typically capture outcomes and a limited number of structured fields. AI can convert meetings, emails, calls, customer objections, competitor references, and commitments into structured signals. Sales management can therefore progress from managing outcomes to managing the underlying process that produces those outcomes.
The Importance of the Data Foundation Increases Rather Than Diminishes in the Age of Large Models
The more autonomous an Agent becomes, the greater the requirement for trustworthy data. The 2026 report is explicit on this point: for established enterprises, data is often the harder problem. If customer, product, pricing, transaction, and interaction data remain fragmented or poorly governed, Agents cannot generate trusted recommendations.
The report therefore recommends placing ownership of critical data domains with the business and establishing clearly accountable data owners and stewards for domains such as customer, product, pricing, and interaction data. These domains can then be developed into reusable data products.
This has an important architectural implication. An enterprise building an Agentic Sales Platform should not design the system around a particular LLM. It should instead build around:
commercial data products + decision services + agent orchestration + workflow applications
Market-opportunity mapping, Customer 360, dynamic pricing, and sales-performance analytics may appear to be separate products, but they reuse much of the same customer, product, transaction, interaction, and external market data.
The enterprise assets with the greatest reuse value are therefore neither individual prompts nor individual Agents. They are governed data domains, business semantics, behavioral history, decision rules, successful cases, and evaluation systems. Models can be replaced. Agent frameworks can be upgraded. But these enterprise-specific context and decision assets are what accumulate into durable organizational capability.
“Buy or Build” Also Needs to Be Redefined from an Enterprise-Software Perspective
The 2025 report offers a pragmatic principle: standardized, low-complexity capabilities should generally be bought—for example, meeting transcription and summarization—while high-value capabilities with the potential to create competitive advantage should follow a buy-plus-build approach.
The report also recognizes that very few enterprises build their entire AI stack from scratch. Even what companies call “build” normally relies on existing models and infrastructure components, with customization focused on specific business scenarios.
This distinction becomes even more important in the Agentic AI era. What truly needs to be built is usually no longer the foundation model itself. It is the organization-specific workflow, business logic, data contracts, tool integrations, permissions, evaluation mechanisms, human approvals, and exception-handling logic.
Enterprises have little reason to reinvent a foundation model. But simply purchasing a generic Sales Copilot is equally unlikely to reproduce the end-to-end operating results described in these reports.
The boundary between what should be bought and what should be built can therefore be summarized more precisely:
Generic cognitive capabilities can be bought; enterprise-specific decision capabilities must be built. Standard tasks can be bought; competitively differentiating workflows require continuous development.
What the 2026 Report Is Really Describing Is a New Revenue Operating System
Viewed from a technology-architecture perspective, the 2026 report implicitly describes a relatively complete Revenue AI Stack.
At the foundation are customer, product, pricing, transaction, and interaction data products. Above that sits a layer combining analytics, automation, LLMs, and agentic capabilities. The upper layer connects these capabilities to CRM, GTM, pricing, customer-success, and seller-interaction applications. Business and technology teams jointly define the impact journeys, while operational data continuously feeds back into the system.
This differs materially from the traditional logic of CRM. The primary function of CRM is to record customers and sales activities. The emerging function of an Agentic Revenue System is to sense commercial states and drive the next action.
CRM may therefore remain the system of record, while the Agent layer increasingly becomes the system of intelligence and, in some areas, the system of action.
For this reason, whether Salesforce, Microsoft Dynamics, or another SalesTech platform contains AI functionality is not the decisive question. What ultimately determines enterprise value is whether the company can build and continuously improve its own closed loop across:
Commercial State → Judgment → Action → Outcome → Feedback
The Greatest Risk in AI Sales Transformation Is Not Model Hallucination, but Local Success with No Enterprise-Level Impact
The 2025 report already cautions that, for low-tolerance processes whose foundations remain heavily manual, conventional automation may sometimes be more appropriate than GenAI. Where necessary, direct links to source information may provide greater reliability than introducing generative capabilities. Hallucination is therefore treated as one factor in technology selection, rather than as an argument for applying GenAI everywhere.
Taken together, however, the two reports point to an even larger organizational risk: every individual tool may work, while the enterprise as a whole fails to change.
A meeting summarizer may save 15 minutes. Email generation may save another 10. A knowledge assistant may improve information retrieval. All of these benefits may be real. But if market selection, account prioritization, pricing strategy, sales process, and customer experience remain unchanged, the ultimate impact on P&L may still be negligible.
AI evaluation should therefore establish a three-level chain of metrics.
First come adoption and behavior metrics, such as usage frequency and acceptance of AI recommendations.
Second come leading business indicators, such as pipeline velocity and conversion speed.
Only then should enterprises evaluate lagging commercial outcomes, including closed-won pipeline, margin growth, revenue growth, and cost-to-serve.
The 2026 report explicitly advocates this type of performance cockpit, linking behavioral metrics to leading indicators and ultimately to financial outcomes.
The Hardest Part Ultimately Returns to Organizational Transformation
This is perhaps the strongest point of agreement between the two reports.
The 2025 report already emphasizes seller-centric design, frequent test-and-learn cycles, seller champions, training, management role-modeling, and AI Centers of Excellence. The 2026 report elevates Adoption and Scaling into one of the four foundational pillars of transformation.
McKinsey cites research suggesting that for every $1 spent deploying AI, companies may need to spend approximately $3 on change management—while many organizations do the reverse.
This should not be surprising. Once Agents genuinely enter the workflow, responsibilities must be redistributed, approval authority reconfigured, KPIs changed, roles redesigned, and some intermediate coordination activities eliminated.
The 2026 report even anticipates the emergence of new roles such as AI workflow designer and agent manager. At the same time, Account Executives are expected to spend less time on administrative coordination and more time on strategic account ownership, executive engagement, complex deal shaping, and relationship development.
From an enterprise-services perspective, the center of gravity in Agentic AI implementation is therefore likely to move from “developing an AI feature” toward “redesigning how an organization gets work done.”
Necessary Caution When Interpreting the Two Studies
The data in both reports is highly useful, but it should not be interpreted as universally reproducible AI ROI.
Many of the cited results come from specific, and sometimes anonymous, enterprises. Outcomes such as a 10 percent earnings uplift, a 4.5-percentage-point improvement in conversion, or a 20 to 40 percent reduction in cost-to-serve are influenced by industry structure, data quality, existing levels of digital maturity, customer mix, and implementation design.
The reports therefore provide evidence of a validated range of possibilities, not a universal benchmark.
A second caution concerns causality. The findings that growth leaders invest more aggressively in AI and that high-performing companies are more likely to redesign workflows around AI are important correlations, but they do not by themselves prove that AI is the sole cause of superior growth.
Stronger management capability, better data foundations, and more effective execution cultures may simultaneously contribute to growth and make AI transformations more likely to succeed. Enterprises should therefore avoid copying investment levels and instead focus on building the underlying organizational capabilities.
One further point deserves particular emphasis. The 2025 report explicitly states that many of the practices it describes combine GenAI with analytical AI, machine learning, and other AI techniques. The resulting business impact therefore should not automatically be attributed to LLMs.
This distinction matters for enterprise architecture. In areas such as sales forecasting, propensity scoring, and dynamic pricing, conventional machine learning may remain better suited than large language models. LLMs are particularly strong in understanding unstructured information, semantic reasoning, human interaction, and workflow orchestration.
Competition in Agentic AI Will Ultimately Be Competition Between Enterprise Work Systems
If one reads only the 2025 report, the obvious conclusion is that B2B sales contains seven highly promising GenAI use cases.
Reading the 2026 report alongside it leads to a more important conclusion:
Those use cases are merely components. The real objective is to redesign the Revenue Engine.
McKinsey’s most consequential argument, therefore, is not that salespeople should use more AI. It is that enterprises need to reconnect opportunity identification, customer coverage, account management, quoting, negotiation, seller enablement, and performance management into commercial workflows that are observable, computable, executable, and continuously fed by feedback.
High-performing companies are nearly three times more likely than their peers to have fundamentally redesigned workflows around AI. That empirical finding provides the most important context for understanding why the analytical unit has shifted from individual use cases to end-to-end impact journeys.
Seen from this perspective, the end state of AI in B2B sales is unlikely to be merely a smarter CRM or an omnipresent Copilot.
What is more likely to emerge is a new AI-native commercial operating system: Agents continuously observe market and customer states; conventional algorithms perform prediction and optimization; large language models interpret complex semantics and unstructured information; business rules constrain decision boundaries; humans remain responsible for trust, judgment, negotiation, and value co-creation; and a unified set of business metrics determines whether the overall system is actually generating greater revenue, margin, and customer value.
This may be the most important lesson for enterprise leaders across both reports:
The maturity of an AI transformation should not be measured by how much AI an enterprise has deployed. It should be measured by how many critical business processes can consistently produce better operating outcomes through effective collaboration between humans and Agents.
Related Topic
From 'Using AI' to 'Rebuilding Organizational Capability'
A Full-Spectrum Analysis of AI Use Cases: From Incremental Efficiency to Compounding Growth
From Tool to Teammate: The Organizational Reconstruction of an AI-Native Enterprise
The Truth About Enterprise AI Deployment: Why 90% of Projects Never Make It Past the Demo Stage
How to Train Teams to Master Artificial Intelligence
The Strategic Shift of Generative AI in the Enterprise: From Adoption Surge to Systemic Evolution
IBM Enterprise AI Transformation Best Practices and Scalable Pathways
OpenAI’s Seven Key Lessons and Case Studies in Enterprise AI Adoption