DataWorks

79% of B2B Teams Deploy Agents, but 94% Say Their Infrastructure Can't Handle Them

A crawl-walk-run maturity model for agentic GTM defined by decision authority, measurement readiness, and governance, not by how many agents you deploy.

DemGen Daily Editorial · November 2, 2026

A LeanData survey of 157 B2B revenue, marketing, and sales operations leaders found that 79% reported deploying agents while 94% said their GTM infrastructure isn't AI-ready. Read those together and the picture is clear: most teams shipped the agent before they built the road it drives on.

That gap is the reason a crawl-walk-run maturity model for agentic GTM matters right now. Not as a framework exercise. As a survival mechanism.

Why "How Many Agents" Is the Wrong Scorecard

Counting deployed agents tells you about procurement activity, not maturity. Microsoft's agentic AI adoption maturity model makes this point explicitly: maturity should be assessed across business process, governance and security, technology and data, strategy, and organizational readiness. Not by tallying bots.

An agent that drafts outbound sequences is worthless if the data feeding it is full of duplicates, the routing logic is opaque, and nobody can audit what it sent. A Basis survey of 184 marketing and advertising professionals found that 65.1% said their organization either does not measure AI ROI or they don't know whether it does. The agent ships, the team reports "efficiency gains," and finance has no way to validate the claim. That's a pilot with no exit criteria.

Three Tiers, Defined by Decision Authority

The model works when each tier is defined by who makes the decision and what evidence you need before graduating.

Crawl: agents assist, humans decide. The agent drafts outreach, flags accounts, or scores leads. A person reviews every recommendation before anything ships. Ops owns at this stage: clean signal data matched to accounts and contacts, a named owner for the shared context layer, and a record of every time a reviewer changes an agent's output. That last piece is your evidence base. Without it, "ready to move up" is a gut call, and gut calls create brittle automations.

Walk: agents run workflows, humans handle exceptions. An agent triggers a follow-up sequence when an account hits a defined buying stage. Frequency caps, audience filters, data access permissions, and escalation triggers decide what runs autonomously and what gets routed to a person. Ops owns guardrails, a shared context layer so every agent works from the same account picture, plus audit trails and rollback capability. Apollo's description of agentic GTM operations emphasizes QA pass rate, pipeline velocity, and handoff efficiency as the metrics that matter here.

Run: agent networks adapt, humans steer. Agents make real-time decisions across the funnel and adjust as results come in. People set strategy and monitor drift. Ops owns traceability from signal to intelligence to action to outcome, plus reporting that finance will accept. BCG's research underscores how rare this stage actually is: only 8% run campaigns in which multiple agents operate autonomously.

Some workflows may never belong in the run tier. A workflow where a wrong move is expensive (record updates, pricing actions, contract terms) can stay in walk indefinitely. That's a sound answer, not a failure of ambition.

The Real Maturity Curve Is Measurement Maturity

Microsoft's guidance describes a progression that maps directly onto this: moving from anecdotal productivity wins to defined KPIs, then to standardized value reporting, then to continuous optimization as adoption scales. Most teams are stuck in the anecdotal phase. The Basis survey found that 75% of marketers said AI made them moderately to significantly more efficient, but only 6.5% said their organization evaluates AI investments against defined benchmarks.

Close that gap before worrying about multi-agent orchestration. If you can't measure what a single agent contributes to qualified pipeline (not form fills, not "time saved," but pipeline), adding more agents just multiplies the measurement debt.

Moving Up Without Tripping

Move one workflow at a time. Prove it on a first-touch email sequence before you touch record updates or routing logic. The hypothesis format works here: "If we move this outbound sequence from crawl to walk, then reviewer override rate will stay below X% and pipeline velocity from this sequence will hold or improve, because the agent's context layer now includes the signals reviewers were manually checking."

Write the exit criteria before the test starts. Name the metric, the threshold, the time period, and the approver. "Agent output override rate below 8% for 30 days, validated by the RevOps lead" is a criterion. "Feels ready" is not.

Plan the step back. If guardrails fire at unexpected rates or humans start overriding more, move the workflow down a tier, find the cause, and re-validate. Microsoft's maturity model emphasizes risk-based controls, monitoring, and clear human-agent roles as agentic systems scale. That guidance exists because the failure mode is real: an agent that drifts without anyone noticing creates pipeline noise, not pipeline.

Where Most Teams Actually Win

The walk stage. Less exciting than autonomous agent networks, but orchestrated multi-step workflows with QA and human-in-the-loop controls produce the most reliable lift. BCG's findings confirm the pattern: only 20% of B2B CMOs said their agentic marketing transformation was already having significant, measurable revenue impact. The other 80% are still building the infrastructure, the measurement, and the governance to get there.

Agentic GTM maturity isn't about reaching the run tier. It's about knowing, with evidence, which tier each workflow belongs in today, and what specifically has to be true before it earns the next one. The teams that get this right won't be the ones with the most agents deployed. They'll be the ones who can tell you exactly what each agent changed, and prove it.

Frequently asked questions

How do you define crawl vs. walk vs. run for agentic GTM in terms of funnel impact and risk tolerance?

Each tier is defined by decision authority and evidence requirements. Crawl means agents assist but humans approve every action. Walk means agents run workflows autonomously within guardrails while humans handle exceptions. Run means agent networks make real-time cross-funnel decisions with minimal human intervention. The tier for each workflow should reflect your risk tolerance: high-stakes workflows (record updates, pricing) can stay in walk indefinitely.

What KPIs determine whether an agentic workflow is ready to scale from pilot to production?

Key metrics include agent output override rate (how often humans change what the agent produces), pipeline velocity from agent-driven workflows, QA pass rate, and handoff efficiency. Write exit criteria before the test: specify the metric, threshold, time period, and approver. Anecdotal time-saved reports are insufficient; tie measurement to qualified pipeline outcomes.

What governance controls are required before agents take higher-impact GTM actions?

At a minimum: frequency caps, audience filters, data access permissions, escalation triggers, audit trails with rollback capability, and a shared context layer so every agent works from the same account data. Microsoft's maturity model emphasizes risk-based controls, monitoring, and clear human-agent role definitions as autonomy increases.

What are the minimum data and process prerequisites before deploying agents broadly if GTM infrastructure isn't AI-ready?

Clean signal data matched to accounts and contacts, documented routing and scoring logic, a named owner for the shared context layer, and the ability to record and audit every agent action. LeanData's survey found 94% of B2B ops leaders say their infrastructure isn't AI-ready, largely because of gaps in data quality, process documentation, and tool connectivity.

How should the team be structured to operate and improve agentic GTM workflows?

Apollo's guidance on agentic GTM team structure emphasizes cross-functional ownership across marketing ops, RevOps, growth, and sales, with operational metrics like QA pass rate, pipeline velocity, and handoff efficiency tracked as you move from pilot to production. Clear ownership and documented handoffs matter as much as prompts and models.