--- title: How to prune your AI pilot portfolio before Q3 board reviews slug: how-to-prune-ai-pilot-portfolio-q3-board date: 2026-04-30 excerpt: "Series C boards are no longer asking how many AI pilots are running. They are asking which ones to kill before Q3, and most teams do not have a defensible answer yet. Here is the five-step pruning protocol." featured_image: "https://bbtxujdxvidaghmhxkqs.supabase.co/storage/v1/object/public/generated-images/blog-1777531645603-how-to-prune-ai-pilot-portfolio-q3-board.webp" canonical_url: https://cerevisor.com/blog/how-to-prune-ai-pilot-portfolio-q3-board updated_at: 2026-04-30T06:47:27.839172+00:00 --- # How to prune your AI pilot portfolio before Q3 board reviews TLDR Series C boards entering Q3 are no longer asking how many AI pilots a company is running. They are asking which ones should be killed, and most teams do not have a portfolio-level inventory yet. The five-step pruning protocol below runs in roughly four weeks and produces the slide a Q3 board most wants to see. Three Series C CEOs sent me their AI portfolio review decks last week. Two of them had the same blank slide in slot four: “AI Initiatives We Are Sunsetting.” Neither had filled it in. That blank slide is the Q3 board review problem. Boards are no longer asking how many AI pilots a company is running. They are asking which ones are still alive in twelve months, and which ones should be killed before the burn keeps going. A Gartner study of 782 [infrastructure](/blog/series-b-ai-infrastructure-cost-reality) and operations leaders, reported by The Register on April 7, found that only 28 percent of [enterprise AI](/blog/series-c-ai-integration-bill) projects fully succeed and deliver ROI. One in five fail outright. The remaining slice, the largest one, is the dangerous slice: alive enough to keep getting funded, dead enough that nobody can point to a clean win. That is the slice your board is now pointing at. 11-25% of enterprise AI pilots reach sustained production (FifthRow, April 28, 2026) --- ## What Series C teams actually built The shape of a Series C AI portfolio in April 2026 is reasonably consistent. Eight to fifteen active initiatives. Most launched between the second half of 2024 and Q1 2026. Funded through a mix of departmental budgets, central IT, and one or two CEO-level priority bets. Few of them ever appeared on a single page together until a board member asked. The reason that page never existed is structural. ProcureAbility’s 2026 CPO-CIO Report, released April 29, found that while 96 percent of [procurement](/blog/series-b-ai-vendor-58-percent-2026-04-29) and IT teams collaborate to some degree, 54 percent of them do not collaborate at all on AI governance. Initiatives got funded one P&L at a time. Procurement did not see them. Finance saw them as line items, not as a portfolio. The result is a long inventory of half-started projects nobody owns at the portfolio level, which is exactly what the audit committee is going to flag. A separate analysis from FifthRow, published April 28, put the production conversion at 11 to 25 percent of enterprise AI pilots reaching sustained production. The same study found that 54 percent of failed projects stall 3 to 9 months after they appear to succeed. That is the painful number for Series C. Most of the pilots a board is about to ask about are inside that 3-to-9-month danger zone right now, looking healthy on a dashboard. There is also a quiet financial number under all of this. FifthRow estimates Fortune 1000 failed AI projects average $2.1 to $2.3 million in sunk cost per project. At a Series C company, that is roughly a percentage point of runway per stalled pilot, which is the part the CFO is starting to compute on a separate spreadsheet nobody else has seen yet. --- ## What the surviving portfolios actually do The teams that have a defensible portfolio entering Q3 share a pattern, and it is not a technology pattern. They committed to fewer bets earlier than they had to. Eaton, profiled in Fortune in March, moved from proof-of-concept thinking to a private-equity-style portfolio: three to five high-impact initiatives with the rest sunset deliberately. Stanford’s enterprise AI playbook from 51 successful deployments points the same way. Nothing about the surviving initiatives required heroic engineering. They had executive sponsorship, they fit existing workflows, and they were measured against pre-pilot baselines somebody had actually written down. The teams without that pattern share the opposite signature. Gartner’s April data attributes most failures to a single root cause: expectations. As Gartner research director Melanie Freeze put it, “AI that doesn’t fit into the organization’s operations simply can’t deliver ROI.” Translating that for a Series C audit committee: most of the alive-but-unhealthy pilots in your portfolio were never going to fit. They were funded because someone in the org was nervous about being slow on AI, not because anyone had drawn the production-side workflow on a whiteboard. > "Only 28 percent of AI infrastructure projects 'fully succeed and offer return on investment (ROI),' [and] one in five AI projects in IT infrastructure and operations fail outright." The Register, reporting on Gartner's 782-leader I&O survey, April 7, 2026 Key Insight The [pilot-to-production](/blog/second-pilot-trap-series-b-2026) attrition is structural, not technical. The teams that survive Q3 board scrutiny are the ones who killed three things last month and can name the two indicators that told them. --- ## The five-step pruning protocol The pruning protocol that works is short, public, and runs on the calendar of the next board cycle, not the Q4 planning offsite. It has five steps and almost nothing about it is technical. - **Build the actual inventory** Pull every AI initiative across every business unit onto a single page with owner, sponsor, funding line, status, and pre-pilot baseline. Most Series C teams will discover one or two initiatives nobody can remember approving. That is normal. ProcureAbility's April 29 report shows 54 percent of organizations have no procurement-IT coordination on AI, so the inventory has to be built by hand the first time. - **Apply the production threshold** Use the FifthRow benchmark as a forcing function: 11 to 25 percent is the realistic survival rate. If your portfolio has fifteen active initiatives, you are budgeting as if four will reach sustained production. Mark the rest as "must defend or sunset" rather than "in progress." - **Run the 9-month stall test** For every pilot older than three months and younger than nine, ask one question: what is the next decision gate, and what evidence will trigger it? FifthRow found 54 percent of failed projects stall 3 to 9 months after apparent success. The pilots without a defined next gate are the ones that will quietly slip into that bucket. - **Score against Gartner's three markers** Gartner's April data identifies what separates the 28 percent that work from the rest: workflow integration, executive sponsorship, and a mature application area. Score every initiative on those three. Anything missing two markers is a candidate to kill, not coach. - **Kill, consolidate, or graduate, in writing** Each surviving initiative gets a one-page graduation memo with named owner, baseline, and exit criteria. Each killed initiative gets a one-paragraph sunset memo with the indicator that triggered the decision. Both memos go to the board pack. The kill list is the artifact, not the spreadsheet. The hard part is not the framework. It is that step one tends to surface a number nobody had a clean answer to. I have not yet seen a Series C team where the inventory took less than two days, and most of them found pilots whose original sponsor had already left the company. That is also normal. That is, in fact, the point. --- ## What I would tell you over coffee If you are heading into a Q3 board, the slide your board most wants to see is not the new initiative slide. It is the kill list. That is the slide that signals you are running AI as a portfolio rather than as a wishlist, and it is the one a few Series C peers are walking into their boards with for the first time this quarter. The teams that look weakest going in are the ones with the largest active portfolio and no sunsets. The teams that look strongest are the ones who killed three things last month and can say which two indicators told them. Portfolio discipline is the part most Series C teams underinvested in for the last eighteen months. It is also the part that takes the least time to fix. Four weeks of inventory and scoring is enough to walk into a Q3 board with a clean answer to the kill question. Most peers will walk in without one. #### Sources - [Agentic AI's Enterprise Tipping Point: How April 2026 Redefined Systematic Innovation](https://www.fifthrow.com/blog/agentic-ai-s-enterprise-tipping-point-how-april-2026-redefined-systematic-innovation-and-production-scale-adoption) - FifthRow, 2026-04-28 - [ProcureAbility's 2026 CPO-CIO Report: Lack of Procurement and IT Collaboration on AI Governance Is Limiting Deployments](https://www.prnewswire.com/news-releases/procureabilitys-2026-cpo-cio-report-finds-that-lack-of-collaboration-between-procurement-and-it-teams-on-ai-governance-is-limiting-ai-deployments-and-adoption-302756327.html) - PR Newswire, 2026-04-29 - [Only 28% of AI infrastructure projects fully pay off](https://www.theregister.com/2026/04/07/ai_returns_gartner/) - The Register, 2026-04-07 - [From pilot mania to portfolio discipline: how the best companies are escaping AI purgatory](https://fortune.com/2026/03/19/from-pilot-mania-to-portfolio-discipline-ai-purgatory/) - Fortune, 2026-03-19 - [The Enterprise AI Playbook: Lessons from 51 Successful Deployments](https://digitaleconomy.stanford.edu/app/uploads/2026/03/EnterpriseAIPlaybook_PereiraGraylinBrynjolfsson.pdf) - Stanford Digital Economy Lab, 2026-03-15