Marketing teams handed more of their work to AI agents this year while trusting their own customer data less. Basis’s AI and the Future of Marketing: 2026 Edition, published October 19, 2026, found that 51.1% of marketing organizations now use AI agents, a 155.5% increase over last year. Those agents most often run reporting and analytics (72.3%) and draft campaign briefs (56.4%). In the same survey, only 13.0% of marketers describe their organization’s first-party data as extensive and well-structured, down from 17.9% a year earlier. The agents are reading records that most of the marketers surveyed won’t call well-structured.
Basil C. Puglisi made the case for this foundation years before agents arrived. In April 2021 he published First-Party Data Goldmine: Building Audiences Without Third-Party Cookies, which used his Factics method to turn evidence into actions. Two of those actions were “Unify data from all owned channels into a single customer view for activation across platforms” and “Use consent management tools to maintain transparency and compliance with privacy regulations.” That advice was written for personalization and analytics. In 2026 the same records feed agents that write briefs and plan media. His position for clients follows: before an agent takes on a new task, a named person checks and signs for the customer fields that task will read.
Agent adoption and data confidence moved in opposite directions in 2026
Basis, which sells advertising software, surveyed 184 marketing and advertising professionals from agencies and brands for its fourth annual report. Beyond the agent figures, 47.8% of organizations replaced human tasks with AI in the past 12 months, up from 27.9% the year before. Agents are also creating content (45.7%) and planning media (35.1%). Measurement lags far behind. Just 6.5% evaluate AI investments against defined metrics and benchmarks, and the report notes that autonomous systems “increasingly depend on the quality of what they run on.”
Two days later, Demandbase released research at its GO conference in New York. G2 conducted in-depth interviews for Demandbase with 124 US B2B marketers at companies with 250 or more employees in September 2026. It found that 72% of teams are the same size or smaller than two years ago, while 70% of marketers face shorter timelines. Only 21% said they rarely or never have to correct AI output. Demandbase sells a B2B marketing agent called Mojo. Both samples are small and drawn from different populations, so the two studies can’t be merged into one trend line.
Marketers rank data low as a problem while few call their data ready
One Demandbase figure sits awkwardly beside Basis’s data finding. Asked about their biggest organizational challenge, 56% of Demandbase’s marketers named processes and cross-functional workflows, and 21% named the tech stack, data and tools. Basis’s respondents, meanwhile, rate their own first-party data poorly, and only 13.0% call it extensive and well-structured.
Both readings can hold at once, and the tension is worth keeping in view. Demandbase argues that adding another tool won’t help unless it removes a handoff, and that is a fair point for a team stuck in approvals. Customer data, though, tends to break at exactly those handoffs. A lead moves from a form to the CRM to a sales sequence, and a field gets renamed, a consent flag fails to carry over, or a duplicate account appears. A team that files those breaks under process may never list data as its problem, even while its agents are reading the damaged record.
An agent repeats a flawed customer record at campaign speed
When a person builds a segment by hand, an odd value in a job title field or a stale purchase date often gets noticed and questioned. An agent drafting a brief or a reporting summary reads the same field and uses it. If the segment definition is wrong, the brief inherits the error, the media plan is built on the brief, and the report later grades results against the same flawed segment. Demandbase’s finding that only 21% rarely or never correct AI output suggests most of these marketers already spend time fixing what AI produces. Some share of those fixes likely traces back to the record the AI was handed.
Puglisi separates two kinds of protection here. Automated validation rules, deduplication and format checks are machine checking machine with no human review, which he classifies as Responsible AI, and every client should run them. The decision to let an agent use a field for targeting or for a customer-facing brief belongs to AI Governance. Under his Checkpoint-Based Governance, a named human accepts, modifies or rejects that use, and the decision is recorded.
How Puglisi would ready a client’s customer data before agents expand
The work starts with an inventory by agent task. For each task an agent runs or is about to run, the team lists the fields it reads. A reporting agent reads revenue and source fields, a brief-drafting agent reads segment definitions and customer attributes, and a media planning agent reads audience lists.
Each field on that list then gets a named owner, a last-verified date and a recorded consent basis, which carries the 2021 consent advice into the agent era. In B2B, that usually means account and contact records, the profiles Puglisi’s 2021 post described building from webinar, whitepaper and email engagement. In B2C, it means purchase history and loyalty data, where a missing consent flag can turn a personalized offer into a compliance problem.
The last step is a readiness gate. An agent takes on a new task only after the owner of each field it reads has reviewed the field and accepted it, sent it back for repair or ruled it out for that use. Basis found that only 6.5% of marketing organizations benchmark their AI investments, and a client that grades its fields before expansion gains a baseline almost no peer has.
Two measures show whether this is working.
The first is the share of agent-read fields that carry a named owner and a verified date, with a target of every field used by an agent that spends money or contacts customers. The second is the duplicate rate and the missing-consent rate in the records agents use, tracked monthly before and after each expansion. If those rates climb as agents take on more work, the client is scaling its errors along with its output.
The open question for 2027 marketing plans
Neither study links data quality to agent results directly. The data question and the agent questions were asked separately, so the gap described here is an inference from two figures. Watch whether Basis’s next edition shows data confidence recovering as agent use grows, or falling further. Watch, too, whether agent vendors, including Demandbase with Mojo, show clients which customer records an agent used for a given campaign, because without that trail nobody can trace a bad brief back to a bad field. As agents take over briefs and media plans, the person who owns the customer record shapes the campaign as much as the strategist who approves it. Few marketing teams have put that person’s name on the record yet.
Sources
- Basis. (2026, October 19). AI and the future of marketing: 2026 edition [Report]. https://basis.com/reports/ai-and-the-future-of-marketing-2026-edition
- Demandbase. (2026, October 21). Nearly three-quarters of marketing teams haven’t grown as marketers face pressure to move faster in the AI era [Press release]. Business Wire. https://www.businesswire.com/news/home/20261021962111/en/Nearly-Three-Quarters-of-Marketing-Teams-Havent-Grown-as-Marketers-Face-Pressure-to-Move-Faster-in-the-AI-Era
- Puglisi, B. C. (2021, April 26). First-party data goldmine: Building audiences without third-party cookies. basilpuglisi.com. https://basilpuglisi.com/first-party-data-goldmine-building-audiences-without-third-party-cookies/
Questions readers ask
What did Basis’s 2026 AI and the Future of Marketing report find about AI agents?
Basis surveyed 184 marketing and advertising professionals and found that 51.1% of marketing organizations now use AI agents, a 155.5% increase over last year. The agents most often run reporting and analytics, draft campaign briefs, create content and plan media, while only 6.5% of organizations benchmark their AI investments.
How many marketers say their first-party data is ready for AI agents?
In the Basis survey, only 13.0% of marketers describe their organization’s first-party data as extensive and well-structured, down from 17.9% a year earlier. That decline came in the same year agent use rose sharply, so more marketing work now runs on customer records that most respondents won’t rate highly.
What did Demandbase’s October 2026 research find about B2B marketing teams?
Demandbase’s study of 124 US B2B marketers, conducted by G2 in September 2026, found that 72% of teams are the same size or smaller than two years ago and 70% face shorter timelines. Only 21% rarely or never correct AI output, and 56% named processes and cross-functional workflows as their biggest challenge.
What is a data readiness gate for marketing agents?
It is a checkpoint a client sets before an agent takes on a new task. The team lists every customer field the task reads, assigns each one a named owner, a verified date and a consent basis, and the owner accepts, repairs or rules out each field under Checkpoint-Based Governance before the agent uses it.
How can a marketing team measure whether its customer data is ready for agents?
Two measures work well. Track the share of agent-read fields that carry a named owner and a verified date, aiming for every field used by agents that spend money or contact customers. Then track the duplicate rate and missing-consent rate in those records monthly, before and after each agent expansion.
#AIg
Leave a Reply