The Work-Shape Question: Why AI Investment Starts Before You Talk to Vendors 

The question isn't which AI vendor to choose or what model to deploy. Before any of that: how does this work actually accomplish value? Most organizations skip straight to comparing capabilities, pricing dashboards, and proof-of-concept demos. They're solving for the wrong variable. The result is predictable, according to Gartner, more than 40% of agentic AI projects will be discontinued by the end of 2027, driven primarily by cost overruns, unclear business value, and inadequate risk controls. The technology isn't the problem. The investment logic is.

Why teams skip the workflow decomposition step

There's no good vocabulary for describing the shape of work. We have plenty of language for outputs (reports generated, invoices processed, leads routed) but almost none for the structure of decision-making, judgment, exception-handling, and coordination that sits underneath. So when a CFO says "we need AI in order-to-cash," and three vendors arrive with three different solutions, none of them are actually describing the work. They're describing their capabilities.

The mismatch is structural. Vendors are incentivized to sell what they've built. Internal teams don't have a framework to articulate what they actually need. The conversation defaults to feature lists and integration promises, bypassing the fundamental question entirely.

What does "work shape" actually mean?

Consider an accounts receivable team. From the outside, it looks like one function. From the inside, it's at least half a dozen distinct work patterns: collections prioritization (ranking which accounts to pursue based on likelihood of recovery), invoice matching (reconciling purchase orders against delivery and payment), customer follow-up (contextual outreach that escalates or de-escalates based on relationship history). Each of these has different decision structures, different data dependencies, different tolerance for error.

Collections prioritization might benefit from a machine learning model that ranks accounts by propensity to pay. Invoice matching might need deterministic rules with human review on exceptions above a certain threshold. Customer follow-up might require generative AI for tone and context, but with strict guardrails on what commitments can be made. Treating this as "one AI problem" leads to one of two failures: over-engineering a general solution that's mediocre at everything, or under-scoping a point solution that solves 20% of the need and leaves the rest unchanged.

The same pattern shows up in marketing operations

Marketing automation fails for the same reason. Teams bolt AI onto existing processes (lead scoring, email sequencing, content personalization) without first understanding whether those processes are structured in ways that create value or just activity. A demand gen sequence that sends seven touches over fourteen days isn't a workflow worth automating faster. It's a workflow worth redesigning.

The work-shape question forces a different conversation. Which parts of our content operations are actually about judgment, deciding what message will resonate with which segment at what stage? Which parts are coordination overhead, routing approvals, updating spreadsheets, syncing systems? Which parts are currently invisible because we've structured the work around tool constraints instead of outcomes?

This decomposition reveals where AI investment makes sense and where it doesn't. Some work benefits from speed and scale, processing inbound form submissions, tagging assets, matching intent signals to content. Some work benefits from consistency, applying a scoring rubric, enforcing brand guidelines, triaging support tickets. Some work requires human judgment and shouldn't be automated at all, relationship decisions, strategic pivots, ethical edge cases.

When the right answer is "wait"

Work-shape analysis also surfaces a less obvious decision: whether to invest at all right now. If the current workflow is poorly defined (if nobody can articulate what good looks like, if exceptions are frequent and unstructured, if success depends on individual expertise rather than repeatable logic) then AI investment is premature. You're automating confusion.

The strategic value of this analysis isn't just better vendor selection. It's the clarity to say "we're not ready" or "this part isn't worth automating" or "we should build this internally because it's core to how we compete." Those are higher-ROI decisions than choosing between two vendors who both promise the same generic uplift.

If you're evaluating where AI investment makes sense in your marketing operations, or trying to articulate why the vendor demos aren't landing, we'd be interested to hear what shape your work is actually taking. Sometimes the clearest path forward is just having the right conversation first.

Sources

• Gartner (2026) More than 40% of agentic AI projects will be discontinued by the end of 2027, driven primarily by cost overruns, unclear business value, and inadequate risk controls. Referenced in source material; original report not specified.