InformationWeek dated October 19 from Orlando: Gartner’s annual CIO survey, more than 2,000 IT executives, says only 13% of CIOs have realized significant or transformational value from AI. AI spend is up 18-fold in two years. Kelsey Ziser photographed Kristin Moyer and Daryl Plummer on the Symposium stage Monday. That is a keynote. It is not a go-live.
Gartner’s own newsroom is the same room: more than 7,500 CIOs and IT executives, through Thursday, value compounding slower than risk. CIO on October 20 is Grant Gross on a different object. Rama Donepudi, global CIO at Mead Johnson Nutrition, says an SAP ECC to S/4HANA program that was projected at 17 to 18 months finished in 11, with $5 million to $6 million from the shorter calendar. Three documents. Mix them and you will buy an agent platform because a flywheel slide said “momentum.”
We already filed Gartner’s $2.7T as a rebrand and three-in-ten shadow agents against a ticket. This week the survey bucket is 13%. The named cutover is baby formula.
13% is a bucket, not a program
InformationWeek’s 13% is the annual CIO survey released Monday. Gartner’s related announcement, as the Joplin Globe reprint of the Business Wire notes, adds that use-case success remains one-in-five even as use cases accelerated over two years. Significant or transformational is the top of a scale. “Some positive value” is a different row: Plummer, in the newsroom release, cites 58% of leaders from a Q2 2026 survey of more than 1,000 functional leaders. Do not average 13 and 58. One is CIOs on enterprise-scale value. One is functional leaders seeing pockets.
The 18-fold spend number sits next to the 13% on purpose. InformationWeek calls it unprecedented. It is a two-year multiple, not a dollar table. Gartner did not print your company’s line item in that sentence. Do not put “18x” on a budget slide as if it were your invoice.
Q3 2026, 303 CIOs and IT risk leaders: only 25% align technology risk with business risk. Moyer: alignment is not everyone getting the same value or the same risk. It is a consistent way to shape the relationship. If your risk committee and your AI steering group still use different heat maps, you are in the 75%. A flywheel diagram will not merge the maps.
StockTitan’s Symposium write-up of the CEO survey is a fourth number set. April and May 2026, 203 global CEOs: 44% struggle to demonstrate AI-driven financial returns. 62% still list revenue growth as a top priority. 75% plan to increase technology investment. David Furlonger: the value problem will not be solved by adding more AI to existing ways of working. CEOs want “strategic freedom.” That phrase is analyst copy. The 44% is the row to keep. Your board packet is not 203 people. It is whether last quarter’s AI line has a return you can defend without a keynote.
71 and 21 are the actual gap
Moyer, same stage: a Q2 survey of more than 2,000 CIOs and technology executives found 71% expect to deploy agentic AI within 12 months, and 21% currently have agentic AI governance. InformationWeek quotes her: “That is out of alignment.” “It’s not possible to create value faster than AI can create risk in that situation.”
We already asked where agents are allowed. This week’s version is a calendar. Twelve months is a procurement hope. Governance capabilities are a control. If you are in the 71 without the 21, you are scheduling autonomy ahead of an override. Plummer’s racecar metaphor, per InformationWeek: speed versus drag. Drag is governance. A car without drag still hits the wall. Do not quote the metaphor in the steering committee. Quote 21%.
The newsroom release says Gartner wants CIOs to use a Value-Risk Flywheel to see what is compounding and why. InformationWeek lists the facets they applied it to: reimagining work, reskilling, reducing AI waste, managing shadow AI, establishing governance. That is a keynote agenda. It is not a Jira epic. Shadow AI is the one you can audit this week without a flywheel license. Ask who already spun up an agent on a personal key. We have been through that inventory.
Agentic in 12 months also is not Mead Johnson’s 2024 ERP tooling. Copilots remediating custom programs and UiPath bots validating data are assistive. They are not an agent that files a purchase order. Chandrasekaran, Gartner’s AI analyst in Gross’s piece, still puts hallucination rates on code mapping and QA. If the model invents a function in a custom SAP include, the 11-month story becomes a restore. Donepudi took ownership. Your change-advisory board may not.
11 months is one CIO, one stack
Donepudi’s 2024 program, as Gross reports it: gen AI coding copilots to remediate custom programs for S/4HANA compatibility. UiPath AI-enabled bots to validate data between the legacy environment and S/4. Enable Now and WalkMe for training content. GPT tools for program communications. He told CIO he deliberately applied different AI capabilities across development, validation, training, and communications rather than treating AI as a single use case.
The calendar: 17 to 18 months projected without those tools, 11 months completed. $5 million to $6 million from compression. Cutover story, told at the CIO 100 Awards in August: down Friday evening, Monday the business asked if they were on S/4. They had not noticed. That is a change-management anecdote. It is not Gartner’s 13%. It is one nutrition company, one ERP move, one CIO who said he was fully accountable instead of parking a $50 million-a-year IBM deployment as the backstop.
He still named Deloitte, IBM, and Accenture as the firms CIOs hire for program management, change, and communications on company-wide finance, supply, procurement, and sales transformations. The AI tools did not replace the big firms in his telling. They replaced some of the hours inside development, validation, and content. If your RFP assumes agents delete the SI, you are not reading Donepudi. You are reading Avaya.
Content was the piece he could see immediately. GPT for communications, Enable Now and WalkMe for role-specific training, updates as the project moved. He told Gross that change management got easier because the prompts produced the materials. That is a document factory sitting next to an ERP. It is not 71% deploying agents. If your transformation stall is actually “we cannot write the training,” buy that stall’s tool. Do not buy an agent mesh.
Vijay Thangella, head of AI and machine learning at Avaya, tells Gross that AI-assisted contact-center modernization can cut a months-long rebuild of workflows, routing, integrations, and business rules to a few days. That is a vendor claim from a contact-center provider. Put it in a different column from the $5–6 million. Brian Klingbeil, chief strategy officer at Ensono, says most organizations still have not used AI to assist digital transformation, and that Ensono uses it for code discovery and legacy cleanup. “Most haven’t” is the base rate. Donepudi is the exception Gross went looking for.
Donepudi also said he had to walk the walk inside IT before telling finance, supply, or marketing to use AI for productivity. That is a political sentence, not a benchmark. It explains why a CIO would put copilots on the ERP that feeds the company rather than on a chatbot in marketing first. It does not make 11 months portable. Your peer who handed the cutover to IBM for $50 million a year, in his contrast, still has someone on the hook. Different hook. Same plant if the include files are wrong.
Chandrasekaran: interest is higher among midsize CIOs with limited budgets for multi-year programs and more risk tolerance than IT at multibillion-dollar multinationals. “You need to accept or tolerate a certain level of risk to trust your AI to do some things that you were doing yourself.” Mead Johnson took that bet on an ERP that feeds baby formula. Your core banking stack may not. Hallucination rates are why “customers are not fully there yet” in his quote. Governance and data infrastructure are the two steps he puts before using AI on the transformation itself. That sentence agrees with Moyer’s 21% more than it agrees with a few-days contact-center pitch.
A flywheel is not a cutover window
Symposium through Thursday is a conference. Gross’s piece is a feature about people already in the work. The collision this week is CIOs treating Orlando as permission to skip the 21%. It is not. 71% planning agents is a poll. An 11-month S/4 cutover is a program with copilots, bots, WalkMe, and a CIO who stood up on a Friday.
Upcoming Symposium dates in the StockTitan note, Yokohama, Barcelona, Kochi in November, are travel. They do not change the 13%. If your team is flying to one of them, take the 71/21 split, not the flywheel poster.
Client Zero was the first bill. This pair is the same distinction with survey percentages attached. If you cannot show a return, you are in the 44% of CEOs Furlonger is describing, or in the 87% of CIOs who did not check “significant or transformational.” Adding another model to the same process is the thing Furlonger said would not fix it.
ServiceNow’s workflow factory already taught us that shipping is not EBIT. An 18-fold spend multiple is not EBIT either. Donepudi’s $5–6 million is the rare row that even pretends to be cash, and it is still a compression of SI calendar, not a new product line.
Do not steal Donepudi’s $5–6 million. He tied it to a compressed SAP calendar, not to an agent fleet. Do not steal 18-fold as your growth rate. Do not steal 13% as a reason to stop. 13% is how many CIOs already claim the top of the scale. It is not a ceiling Gartner set for you. It is also not a majority. If your board heard “transformational” in a vendor demo, put the survey next to the demo. Then put the ERP timeline if you actually have one.
What you can steal from Orlando this week: the 71/21 split as a go/no-go. No agentic rollout without a named override, a named owner, and a place shadow AI already goes to die. What you can steal from Gross: AI on transformation is currently code mapping, test validation, and training content, with hallucination still on the Gartner analyst’s list. What you cannot steal is a Friday-to-Monday S/4 story unless you have the same custom-code inventory, the same bots, and the same person willing to be on the hook instead of IBM.
Plummer said pockets of success are hard to replicate across the enterprise. Donepudi’s Monday morning is a pocket that landed. The 7,500 people in the hall are not that pocket. If your ticket queue still cannot say which data an agent may touch, you are in Chandrasekaran’s “not fully there yet.” Stay there until the maps match. The flywheel can wait.