How Marketing Wins the Agentic AI Moment: Closing the Loop Between Decisions and Dollars
A response to recent McKinsey research on AI's disruption of ERP — for marketing technology and operations leaders.
IIf you lead marketing technology or operations, agentic AI is finally collapsing the divide between marketing's decisions and finance's books — a divide every other enterprise function closed decades ago. The teams that move first will own a capability their CFOs have been asking for, in some form, for twenty years: a defensible, real-time answer to what did we spend, on what, and what did it return?
To understand why this moment is different, it helps to name what has historically been in the way. Finance and marketing carry two maps of the same city. Finance's map is organized by zip code — cost centers, GL accounts, fiscal periods. Marketing's map is organized by transit routes — campaigns, programs, audiences, channels, funnel stages. Both are accurate. Neither overlays cleanly on the other. The monthly close has been the moment someone translates between them by hand.
To bridge the two maps, most enterprises have stood up a parallel stack — work management platforms for execution, marketing financial management tools for plan-to-actual budgeting, MRM and DAM platforms for content operations, and connected planning platforms for scenario modeling. Each layer earns its keep. Together they form an architecture that, until now, depended on careful integration and a small group of practitioners who held the connecting logic in their heads.
Two recent McKinsey articles — The end of ERP as we know it and Bridging the great AI agent and ERP divide — explain why that arrangement is about to shift in marketing's favor. The articles are written for CIOs and CFOs, but the upside for marketing is arguably larger than for any other function, precisely because the gap has historically been one of the widest in the enterprise. The wider the gap, the larger the prize for closing it.
The thesis, in brief
The two articles work together. The first describes a future in which AI restructures ERP itself, framed by two endpoints: a "SaaSpocalypse" in which agents replicate ERP capabilities on the fly, and a more conservative future where a stable transactional backbone remains but agents handle most orchestration on top. The research projects that AI can reduce ERP program cost and duration by at least half.
The second article is the cautionary half. Companies are shifting investment into AI while quietly defunding the ERP, data, and process work that makes AI useful. The result is what the research calls "pilot purgatory": roughly 80% of companies report using generative AI somewhere, but only about 40% report enterprise-level EBIT impact. The diagnosis is that AI use cases depend on the data, applications, and end-to-end workflows the ERP encodes, and that an agent cannot work around a broken foundation.
Read together, the articles converge on a single point worth holding onto: the unlock is not the agent. The unlock is the shared business ontology the agent operates on.
For marketing, that point matters more than it does for most other functions.
Why marketing's gap has always been structurally hard
Most enterprise functions have, over time, bent themselves to fit the ERP. Marketing has historically operated with the least direct support — not because of vendor failure, but for structural reasons:
Marketing's primary objects do not map cleanly to the chart of accounts. A campaign spans cost centers, periods, agencies, and business units. A program is a portfolio of campaigns. An audience cuts across all of it.
Marketing thinks in commitments; finance sees invoices. When marketing reports "$4M committed to Q4 paid social," the underlying reality is signed IOs and SOWs. Finance sees only the invoices that have posted. The gap between committed, accrued, and actual is where most budget surprises originate.
The unit of analysis changes frequently. Channel one quarter, ICP segment the next, per-product ROI the quarter after.
Soft costs are real but largely invisible to the GL. Internal hours, agency retainer burn, contractor utilization — all belong in an honest program P&L, but few ERPs track them at the grain marketing needs.
These structural realities are why the marketing overlay category exists. The integrations between overlays evolved because no single tool models the full picture. The architecture works, but it depends on careful integration and a small group of practitioners who hold the connecting logic in their heads.
What agentic tooling actually changes
It is tempting to think of an "AI agent" as a smarter assistant bolted onto the existing stack. That framing tends to lead to pilot purgatory.
A more useful framing: an agent that holds a shared map of marketing's objects (campaign, program, channel, audience, vendor) and the ERP's objects (cost center, GL account, PO, accrual, vendor master), can read and write across both, and operates within auditable governance.
When that foundation is in place, several historically difficult workflows become more tractable:
Continuous plan-to-actual reconciliation. What is currently a monthly fire drill becomes a live view, with variances surfaced by program rather than by cost center as they emerge.
Commitment-to-accrual automation. An agent monitoring contracts, IOs, and SOWs posts soft commitments and accruals to the ERP in near-real-time, with campaign and program metadata preserved.
Scenario re-planning on demand. "What happens to Q4 if we shift $2M from paid social to events?" moves from a multi-day modeling exercise toward a conversational interaction, with assumptions documented for audit.
Program profitability stitching. Answering "what did Campaign Alpha cost and return?" requires data from CRM, ad platforms, agency invoices, internal labor, and the ERP. With a shared ontology, an agent can produce a per-program P&L on demand. Without one, it will produce confident but unreliable outputs.
Vendor and agency governance. Agents can monitor retainer burn, flag scope creep, route approvals against thresholds, and surface anomalies — off-contract spend, unusual invoice patterns, duplicate work across agencies — without anyone running a report.
The common thread is closing the loop between what marketing decided and what the financial system records. That loop is where value tends to leak today, and where agents are best positioned to help — provided the ontology underneath is sound.
How to think about the vendor landscape
Major platforms in this space are all investing meaningfully in agentic capabilities. Without rank-ordering them:
Work management platforms (such as Adobe Workfront) position themselves as the system of record for marketing work, with integrations into financial planning tools.
Marketing financial management platforms (such as Uptempo, which absorbed Allocadia and BrandMaker) focus on plan-to-actual discipline and program-level budgeting for distributed global marketing organizations.
Marketing resource management and content operations platforms (such as Aprimo) are embedding agents across planning, creation, review, and distribution, and exposing managed content to external LLMs via MCP servers.
Connected planning platforms (such as Anaplan) are rolling out role-based AI agents that turn natural language into structured planning models, with autonomous agents for anomaly detection and recommendation arriving over 2026.
The basis of competition is shifting. Useful procurement questions in 2026:
Can the agent take an action that posts to the ERP without a human re-keying, with full audit trail?
What does the platform's business ontology look like, and how does it map to enterprise master data?
What happens at the seams between this platform, finance systems, and other marketing tools?
How is governance handled for autonomous actions — thresholds, reversibility, accountability?
Platforms with strong answers are likely to consolidate value. Those without will face increasing pressure as the overlap zone between ERP and marketing overlays narrows.
The build-versus-buy calculus is shifting, but not as dramatically as some takes suggest
A common reaction is that "agents will replace the marketing financial stack." This is probably premature.
The overlay platforms encode years of marketing-specific process knowledge — how media commitments differ from production commitments, how retainers amortize, how campaign hierarchies actually roll up. That knowledge is itself an asset, and a parallel point holds for ERP: leaders often underestimate the "equity" inside core systems — the embedded process knowledge AI needs to be useful.
What does shift is the cost of extending the ERP toward marketing constructs. If the projection that AI can cut ERP program cost and duration by half holds, the historical reason organizations could not natively model campaigns in their core ERP starts to weaken. Some of what marketing overlays do becomes contestable on cost grounds, particularly for organizations with simpler taxonomies.
The likely outcome is not displacement but a shrinking overlap zone. Overlay platforms that differentiate on depth of marketing process — rather than on a generic LLM bolted on — will continue to be valuable.
Five practical moves for the next twelve months
The single takeaway: agents amplify whatever they sit on. A clean foundation makes them dramatically faster and more reliable. A messy foundation produces fluent, confident, unreliable outputs at scale.
1. Invest in the taxonomy first. Audit the campaign and program hierarchy. Reconcile it with cost center and project structures in the ERP. Get marketing, finance, and IT to agree on a single business ontology — what constitutes a campaign, a program, a channel, how they roll up, how they map to the GL.
2. Pick one closed-loop pilot rather than ten experiments. Plan-to-actual reconciliation is a defensible starting point: bounded scope, measurable ROI, and doing it well forces taxonomy decisions to the surface.
3. Push for API access at the seams. Most of the value lives between systems. The right question for IT and vendors is whether an agent can read commitments from the ERP and write accruals back, pull spend data from ad platforms, and reconcile against the work management tool — without manual export-import steps.
4. Treat governance as a product requirement. Decide in advance what an agent can do autonomously, what requires approval, who is accountable when it acts incorrectly, and how actions can be rolled back. This is harder than the technical work and frequently more consequential.
5. Architect for interchangeability. Every overlay vendor will pitch a consolidation story. None has clearly won the broader marketing operating system. The shared ontology should belong to the enterprise; the tools on top should be replaceable.
Implications for agencies and systems integrators
For the agencies, consultancies, and SIs that serve marketing leaders, the shift is consequential. The traditional services portfolio — campaign execution, MarTech implementation, integration build-out, monthly reporting — relied on the same gap between finance's and marketing's maps. A meaningful share of billable hours has gone into reconciling them by hand. As agents close that loop, the volume of that work will compress.
What grows in its place is higher-leverage advisory work: ontology design, governance frameworks, agent orchestration architecture, change management for the operating model shifts agents introduce, and the cross-functional facilitation required to get marketing, finance, and IT to agree on shared definitions. SIs that lean into ontology and governance as a service offering — not just implementation — will likely capture disproportionate value. Those that continue to bill primarily for integration plumbing and reporting cycles will see margins compress as that work gets absorbed by the platforms themselves. For agencies specifically, the implication extends to how engagements are scoped and measured: when an enterprise agent can monitor retainer burn and scope creep in real time, the conversation about agency value moves up the stack toward strategy and creative outcomes, away from hours and deliverables.
The bottom line
For the first time, marketing has a credible path to giving finance the answer the CFO has been asking for since the budget conversation began: real-time clarity on what was spent, on what, and what it returned — at the grain of campaigns and programs, not just cost centers. The backbone is becoming flexible enough to model what marketing does. Agents are emerging as the interface where the two functions can speak a common language. And the ontology underneath — the work of agreeing what a campaign actually is — is becoming one of the more strategically important assets a marketing organization can own.
The leaders who capture this moment will be the ones who treat 2026 as a foundation year: clean the taxonomy, close one loop end-to-end, push vendors to operate at the seams, and refuse the temptation to chase ten generative-AI experiments instead of building one defensible plan-to-actual capability. Marketing has waited twenty years for the tools to make its decisions and its dollars speak to each other. Those tools are now within reach.
Sources and citations
Primary sources
1. "The end of ERP as we know it? Five ways AI is disrupting ERP" — McKinsey & Company, McKinsey Technology, May 2026. Source for the five disruption vectors, the SaaSpocalypse vs. stable backbone framing, and the projection that AI can reduce ERP program cost and duration by at least half.
2. "Bridging the great AI agent and ERP divide to unlock value at scale" — McKinsey & Company, McKinsey Technology, January 2026. Source for the "great divide" and "pilot purgatory" framing, the 40%/80% EBIT-impact and gen AI usage figures, and the shared ontology argument as a precondition for agentic value.
Vendor-specific references
3. Adobe Workfront and Anaplan integration positioning — including Workfront Fusion as the integration layer between work management and financial planning.
4. Uptempo positioning as a marketing financial management platform — formed from the combination of Allocadia and BrandMaker.
5. Aprimo's 2026 release with agentic capabilities — embedded across planning, creation, enrichment, review, and distribution, including an MCP server to expose managed content to external LLMs.
6. Anaplan's role-based AI agent suite — including CoModeler and the planned 2026 rollout of autonomous agents.
Supporting context
7. Independent market overview of marketing budgeting and resource planning tools — covering positioning of Uptempo, Anaplan, Workfront, Aprimo, and adjacent platforms.
Note: Vendor capability descriptions reflect publicly available information as of mid-2026 and are illustrative of category positioning rather than comprehensive product evaluations. Readers evaluating these platforms should validate current capabilities directly with vendors.
Want more from The Wolf Tank? Click here to be notified when new posts are published.