What Makes a Growth Champion in Agentic AI B2B Sales
- Heidi Schwende
- Jul 28
- 7 min read
Summary:
McKinsey's 2026 B2B Pulse Survey found that fewer than 10 percent of organizations have scaled AI in any function, even though almost every B2B company is already using some form of it. The companies pulling ahead redesigned the workflow the tool sits inside, backed by a data structure where someone actually owns the customer, product, and pricing information an agent needs to act on, rather than simply adding more tools. That's the same gap behind Salesforce's Agentforce adoption stall, just showing up at industry scale instead of one vendor's numbers. This piece lays out what the research found, why it lines up with what's already visible in CRM data, and what a mid-market company can build without an enterprise budget.
Key highlights
High-growth B2B companies are three times more likely than their peers to have increased AI investment by double digits in 2026, 71 percent versus 25 percent, but fewer than 10 percent of organizations have scaled AI in any function at all.
Growth leaders are nearly three times more likely to have fundamentally redesigned individual workflows around AI, rather than layering AI onto their existing process.
For every $1 spent deploying AI, companies need roughly $3 in change management. Most organizations spend the ratio in reverse.
Companies that rewired prospecting and account management with agentic AI saw 3 to 15 percent higher revenue per relationship manager and 20 to 40 percent lower cost-to-serve.
The research names data ownership, one accountable owner per domain such as customer, product, or pricing, as the harder problem to solve than the AI itself.
The Redesign Gap Behind Agentic AI Adoption Numbers
Most B2B companies have already added AI to their sales motion in some form, usually a chatbot, a summarization tool, or a copilot bolted onto the CRM. McKinsey's 2026 B2B Pulse Survey confirms what that pattern usually produces, real but shallow gains. Fewer than 10 percent of organizations have scaled AI in any function, despite adoption being close to universal at the tool level.
The companies growing faster than their peers didn't get there by rolling out more tools. High-growth companies are three times more likely to have increased AI investment by double digits year over year, 71 percent versus 25 percent of laggards, and growth leaders are nearly three times more likely to have redesigned individual workflows around AI rather than adding it to the workflow that already existed. The tool itself rarely changes much. What changes is whether the workflow around it gets rebuilt to use it well.
This tracks with a pattern I laid out earlier this year in Why Cleaning Your CRM Data Won't Fix Your AI Problem. When Salesforce's Agentforce adoption stalled at 34 percent, the read from most coverage was that the product hadn't landed. A closer look at what KeyBanc and other analysts pointed to told a different story.
The CRM data underneath the agent wasn't structured well enough for it to act on reliably. That piece also cited Fivetran's 2026 Agentic AI Readiness Index, which found only 15 percent of organizations fully prepared to run agentic AI in production even as most were already investing heavily in it. McKinsey's numbers here are the same finding at industry scale, across sales specifically, rather than one vendor's install base.
Five Workflows, One Underlying Requirement
McKinsey frames the opportunity as five commercial workflows worth rewiring end to end. Each one follows the same shape. Data gets pulled from multiple sources, an agent turns it into a recommendation or a drafted next step, and a person makes the judgment call the agent can't.
Opportunity mapping
Agents scan external signals, hiring trends, filings, procurement activity, to identify warm leads and route them to the right seller with a recommended angle.
Go-to-market coverage
Agents match account value to the right coverage model, acting as a copilot on high-value accounts and running more of the outreach directly on lower-tier ones.
Account growth
Agents build a live view of each account from outside-in intelligence and internal data, then surface the cross-sell, pricing, or win-back opportunity worth acting on next.
Pricing
Agents pull real-time market and deal signals to recommend price, bundle, and terms at the moment of quote, instead of relying on a static list.
Seller enablement
Agents analyze pipeline and conversion data to surface coaching opportunities, closing the gap between how the best reps sell and how everyone else does.
None of the five works if the underlying data can't support it. An agent scoring a lead needs firmographic and account data that's actually current. An agent recommending a price needs a clean history of what similar deals actually closed at. An agent drafting account intelligence needs a CRM where "stage," "opportunity," and "win" mean the same thing across every seller who touches them.
Financial services companies that rewired prospecting and relationship management this way saw 3 to 15 percent higher revenue per relationship manager and 20 to 40 percent lower cost-to-serve. That range is wide because the starting data quality varies just as widely.
The Harder Problem Was Never the Model
McKinsey's own language on this is direct. For most established companies, data is the harder problem, and agents can't produce trusted recommendations if customer, product, pricing, or interaction data stays fragmented or poorly governed. The fix the research points to is ownership, not a data cleanup project. One team accountable for each domain, customer, product, pricing, or interaction, with a named owner and stewards responsible for quality standards and resolving issues when the data doesn't hold up.
That structure is worth pulling out on its own, because it's the part a mid-market company can actually build without enterprise resourcing. Rather than a data platform overhaul, it takes someone in the business, not just IT, who's accountable for what "customer" means in the CRM, what "qualified" means on a lead, and what a closed-won deal actually includes. Once those definitions hold across the business, the agent has something reliable to work from instead of a guess dressed up as automation.
The financial case for getting the sequencing right is bigger than most budgets currently reflect. McKinsey's research suggests companies need to spend roughly $3 on change management for every $1 spent deploying AI, and most organizations currently invert that ratio, weighting the spend toward the tool instead of the adoption work around it.
It's the same infrastructure-before-capability pattern I wrote about in The Agentic Web's Two Layers, just showing up inside a company instead of across the open web. An agent's capability, what it can draft, score, or recommend, only means something once the identity layer underneath it, who owns the data, what the fields mean, who's accountable when they're wrong, is actually built. Skip that layer and the agent has nothing reliable to stand on, whether it's negotiating a commerce transaction on the open web or scoring a lead in a CRM.
The challenge is getting the right message in front of the right buyer, solving the right problem at the right moment. Shane Paladin, chief customer and revenue officer, Equinix
What This Means for a Mid-Market Sales Team
The research behind this is built largely on enterprise examples, companies like Equinix, Covestro, Dechra, and global financial services firms. None of those companies are running a 40-person sales team on a single CRM instance. What carries over is the sequence of steps, not the scale. A mid-market business doesn't need five rewired workflows at once. It needs one, built on a data foundation that can actually support it. The sequence that holds up across the research looks like this.
Name an owner for each core data domain
Customer, product, and pricing data each need one accountable person, not a shared responsibility that quietly belongs to no one.
Pick the workflow with the clearest data trail already in place
Account growth and pricing tend to have more usable history than opportunity mapping for a company just starting out.
Budget for adoption at the same weight as the tool
If the technology costs a dollar, plan for the training, coaching, and workflow redesign to cost close to three.
Keep a person in the loop on judgment calls
The research is consistent on this across every workflow. Agents handle the synthesis and drafting, people handle the relationship and the decision.
None of this depends on predicting exactly how agentic AI evolves over the next year. Treat the data and the workflow as the actual project, and the agent becomes what runs on top of it once that work is done.
Frequently asked questions
Does agentic AI replace the sales rep?
No. Across every workflow McKinsey studied, agents take on research, synthesis, and drafting, while the relationship-building, negotiation, and judgment calls stay with the person. The research frames this as agents freeing up seller time rather than replacing the seller.
What's the difference between a gen AI tool and agentic AI in sales?
A gen AI tool typically drafts something on request, an email, a summary, a first-pass proposal. Agentic AI is built to take a sequence of steps on its own within approval rules a team sets, like scoring a lead, checking it against account history, and queuing a recommended next action for a rep to approve.
Why do most AI sales tools fail to move the needle?
The pattern across the research is consistent. The tool gets added to an unchanged workflow running on fragmented data. The gain shows up as individual productivity, faster email, a cleaner summary, but it doesn't change the underlying economics of the sales process.
Do we need to fix our CRM before using AI agents?
Not a full cleanup. What matters more is agreement on what the core fields actually mean, what counts as a qualified lead, a stage, a closed deal, so the agent is working from a consistent definition rather than reconciling different sellers' interpretations.
Who owns data quality at a company too small for a dedicated data team?
It doesn't require a new hire, just naming an existing person, often a sales operations lead, an office manager, or a founder, as the accountable owner for each domain, with the authority to enforce a definition once it's set. The role is about accountability, not headcount.
How much should a mid-market company budget for change management versus the AI tool itself?
The research suggests roughly three dollars in change management for every dollar spent on deployment. Most companies currently spend closer to the reverse ratio, which helps explain why tool adoption and scaled value capture often move at different speeds.
Sources:
Alexander Dierks, Isabel Huber, Maria Valdivieso, and Richelle Deveau, with Neha Singh, "The future of B2B sales: How growth champions rewire their playbooks with AI," McKinsey & Company, 2026 B2B Pulse Survey (mckinsey.com).
Heidi Schwende, "Why Cleaning Your CRM Data Won't Fix Your AI Problem" (2026), citing Fivetran's 2026 Agentic AI Readiness Index.
Heidi Schwende, "The Agentic Web's Two Layers: What's Really Underneath Identity and Capability" (2026).
Comments