The Portfolio That Did Not Pay Off
An enterprise IT organization launched seven agentic AI initiatives across 2024. Each had a sponsor, a budget, and a roadmap. At the end of the year, the aggregate result was: one initiative had reached production with measurable value, three had reached pilot with unclear value, two had stalled, and one had been quietly closed. The total spend was $9.2M. The total value, optimistically calculated, was $1.4M.
This pattern is not unusual. Deloitte's 2024 enterprise AI survey found roughly 60 percent of agentic AI portfolios produced negative ROI at 12 months (Deloitte, "State of Generative AI in the Enterprise, Q4 2024"). The 40 percent that produced positive ROI shared a recognizable pattern. They had concentrated investment in a small number of workflows from a specific profile rather than spreading across many.
If your agentic AI portfolio is broad and shallow, the pattern is set up to produce the same result. Knowing which workflow profiles actually pay off in year one shapes a more productive year.
The Three Profiles That Pay Off First
Three workflow profiles consistently produce positive ROI in year one for enterprise agentic AI initiatives. The reasons are structural: each profile has a combination of volume, value-per-task, and definable success that aligns with current agentic capability.
Profile one is high-volume support work where the task is well-bounded. Customer support that follows defined playbooks, tier-one IT support with known troubleshooting paths, claims triage in insurance, prior authorization review in healthcare. The volume justifies the engineering investment. The bounded task scope means agentic capability can deliver against defined success criteria. The value per task is moderate but multiplied by volume.
Profile two is research and analysis work where the cost of a comprehensive review is currently the bottleneck. Legal contract review, due diligence work, regulatory analysis, competitive intelligence research, sales call analysis. The value per task is high. The volume is moderate but compelling. The bounded scope (analyze this contract, summarize this filing) means agentic capability can deliver structured output that humans review.
Profile three is internal knowledge work where the cost of finding and applying institutional knowledge is currently slow. Engineering support for queries on internal systems, employee onboarding navigation through company-specific information, executive briefings synthesizing internal data. The value per task is moderate. The volume is high. The reduction in time-to-answer often dwarfs the direct cost savings.
Workflows in any of these three profiles consistently pay back in year one when implemented with production-grade engineering. Workflows outside these profiles can pay back but usually take longer, require more investment, or carry more execution risk.
What Distinguishes Payback Workflows From Payback Theater
Three characteristics distinguish workflows that genuinely pay back from workflows that produce announcements without measurable returns.
The first characteristic is unambiguous baseline cost. The team can answer "what does the current process cost per task" with real numbers, not estimates. The baseline is documented before the agentic initiative starts. Without the baseline, the savings claim is unprovable.
The second characteristic is delivered quality at parity or better. The agentic version produces results at least as good as the human version on measurable dimensions. Speed gains that come with quality regressions trigger user reversion and the value disappears.
The third characteristic is sustained adoption after launch. Six months after the agentic capability ships, the volume of usage matches or exceeds the pre-launch projections. Capabilities that ship and then see usage decline have not actually replaced the prior work.
The payback workflows hit all three characteristics. The theater workflows hit one or two and produce announcements that do not survive year-two scrutiny.
What Pays Off Last (Or Never)
For symmetry, the workflows that consistently fail to pay back in year one share characteristics worth recognizing.
Strategic workflows where the value depends on better decisions over long horizons. Strategy synthesis, executive briefing for major decisions, M&A target analysis. The agentic capability can deliver good analysis. The value depends on whether the analysis leads to better decisions, which is hard to measure and accrues over years.
Creative workflows where success depends on subjective quality judgments. Marketing copy generation at scale, design ideation, brand voice development. Agentic capability can produce output. The success measure is harder to define and the human-in-the-loop overhead is high.
Workflows that span many functions with diffuse ownership. Cross-functional process automation, end-to-end customer journey orchestration. The technical implementation is feasible. The organizational complexity is the bottleneck.
These categories can produce value eventually. They rarely produce value in year one without organizational conditions that most enterprises do not have.
What Concentration Looks Like
The 40 percent of enterprises that produced positive ROI on agentic AI in BCG and Deloitte's tracking concentrated investment in two to three workflows from the payback profiles rather than spreading across seven or ten workflows. The concentration ratio matters because agentic work requires engineering depth that diffuse investment does not provide.
Two workflows at depth produce real production capability. Seven workflows at surface produce announcements and partial pilots. The math of engineering capacity favors concentration.
A practical heuristic: an enterprise with engineering capacity to staff three to four production-grade agentic initiatives concurrently should run two or three rather than five or six. The discipline of saying no to attractive-looking initiatives is what enables the chosen initiatives to ship.
What Logiciel Does Here
Logiciel works with enterprises whose agentic AI portfolios have spread across many initiatives and produced limited results, and who want to concentrate the next year's investment in workflows from the payback profiles. The work is structured around portfolio review against the three profiles followed by deeper engineering on the chosen initiatives.
The Agentic Enterprise Workflows framework covers the four-axis assessment for workflow selection. The Pilot to Production Path framework covers the 12-week sequence for moving selected workflows from pilot to production.
A 30-minute working session is enough to map your current portfolio against the three payback profiles.
Frequently Asked Questions
Should every agentic initiative be in one of the three profiles?
Not strictly. The profiles describe what consistently pays back in year one. Initiatives outside the profiles can be valuable and may pay back on longer timelines or for strategic reasons. The portfolio should be skewed toward the profiles, not exclusively confined to them.
How do I evaluate which profile a candidate workflow fits?
Two questions. What is the current cost (or constraint) per task in the workflow today? Can the agentic version deliver measurable parity-or-better against that baseline? Workflows that produce clear answers to both questions usually fit one of the profiles. Workflows with vague answers usually do not.
What is the typical investment per initiative for payback workflows?
For high-volume support workflows (profile one), $500K-$2M depending on scale. For research and analysis workflows (profile two), $300K-$1M. For internal knowledge workflows (profile three), $200K-$800K. The ranges depend heavily on the specifics of the workflow and the integration complexity.
What is the typical year-one ROI for payback workflows?
2x-5x for profile one workflows at scale. 1.5x-3x for profile two workflows. 1.2x-2.5x for profile three workflows. These ranges reflect the BCG and Deloitte data on the 40 percent that produce positive ROI.
How do I handle stakeholders pushing for non-payback workflows?
Through the portfolio framing. Position the payback workflows as the first wave that funds and demonstrates capability. Position the more ambitious workflows as the second wave that depends on the success of the first. Most stakeholders accept this sequence when the alternative is doing both badly. Sources: - Deloitte, "State of Generative AI in the Enterprise, Q4 2024" - BCG, "AI at Scale 2024 Report"
