By Olivier Gomez (OG), CEO of IAC.ai · 12/08/2026 · First published in his newsletter on Substack
That is not a measurement problem.
That is an accountability vacuum.
And right now, it is eating billions of dollars in AI investment alive.
I have been in that room. The transformation dashboard perfect. Every metric is green. Thousands of hours saved. Hundreds of processes automated. The vendor’s slide deck is a work of art. And then the CFO walks in, opens the P&L, and asks one question: “Where is it?”
Silence.
That silence is not a technical failure. The technology worked. The delivery team did their job. The project is closed in DevOps. The ticket says done.
But the outcome? Nobody owns it. Nobody is tracking it. And six months later, the same executive who championed the project has moved on to the next shiny thing, and the value that was supposed to materialize is sitting somewhere between the business case PDF and a Miro board nobody opens anymore.
This is the conversation I had with Alex Leonida, founder of SilkFlo, an AI Value Realization Platform, and someone who has delivered over £100 million in cost reduction across multinational programs over the last decade. We talked for over thirty minutes about why this problem is getting worse, not better, and what it actually takes to fix it. What follows is the unfiltered version.
The number that should stop you cold
IBM data referenced in our conversation puts the share of organizations achieving real, bottom-line ROI from AI at roughly 5%.
Not 30. Not 10. Five.
We used to say 80% of automation projects fail. With AI, we are now looking at 95% not delivering verified bottom-line impact. That is not a rounding error. That is a systemic governance failure at an industrial scale.
Gartner projects worldwide AI spending to reach approximately $2.5 trillion in 2026. If current tracking discipline holds, the math is brutal. The vast majority of that investment will produce dashboards full of activity metrics that never connect to a single dollar on the P&L.
And this is not a new problem. We did the same thing with ERP. With RPA. With CRM. Every single technology wave followed the same pattern. Excitement. Spending. Activity metrics sold as proof of value. CFO asks for the receipts. Silence.
The difference with AI is that the cycle is ten times faster and the budgets are ten times larger. What used to take years to unravel is now happening in months. Uber reportedly burned through its entire AI budget for 2026 at a pace that would have been unthinkable in any prior technology cycle. Meta and Google are spending at a scale that makes token costs a line item in every employee’s workflow. For a large enterprise or a mid-market company doing serious AI adoption, the pressure to show a return is no longer theoretical. It is showing up in board meetings right now.
Why the KPIs lie
Here is the mechanism that creates the gap between green dashboards and flat P&Ls.
Vendors measure adoption. Finance measures outcomes. Nobody connects the two.
When a technology vendor shows up at renewal time, they bring you adoption metrics. Active users. Queries processed. Hours of usage. Tickets closed. These numbers are designed to justify license renewal. They are not designed to answer the CFO’s question.
Meanwhile, the finance team is looking at OpEx and CapEx. Revenue per headcount. Cost per process. They are not in the room when the transformation team is celebrating 3,000 hours saved per year. And when they eventually ask what those saved hours translate into in the P&L, the answer is almost never ready.
The business case was optimistic, as business cases tend to be. The BA who wrote it has moved on to another project, possibly another company. The delivery team was measured on story points and sprint velocity. Their ticket is closed. By the definition of their job, they are done. Nobody assigned outcome ownership. Nobody set a baseline. Nobody agreed on what the metric would be at month six, month twelve, or month thirty-six.
That is not a technology problem. That is a governance problem. And it is happening everywhere.
Who is accountable? Nobody.
This is the uncomfortable truth Alex put on the table, and I agree with it completely.
Right now, in most organizations running AI programs, nobody owns the outcome. The business case gets signed off. The project kicks off. The delivery happens. The ticket closes. And then the outcome just floats, unowned, in the space between the technology team and the finance team.
When the CFO eventually calls and asks why four million in AI spend has not shown up in the P&L, the best answer available is usually something like: “We automated 10,000 quotes. We saved 3,000 hours.” That answer does not connect to OpEx reduction. It does not connect to revenue growth. It does not connect to anything that appears in a financial statement.
PMBOK, the standard project management framework, has only recently started treating value management as an official closing activity. That tells you everything. For decades, project closure meant delivery closure. The idea that someone should still be accountable for the outcome six months after go-live was not even formally recognized in the methodology.
The accountability gap is structural. It is not a matter of individuals failing. It is a matter of organizations not having built the function, the ownership model, or the tooling to hold the outcome thread from day one of ideation through to multi-year post-live performance.
What the top 5% are doing differently
The companies achieving real AI ROI are not smarter. They are not using better models or better vendors. They are doing something fundamentally different in how they think about AI investment.
They treat AI like capital, not like a technology project.
That distinction matters enormously. When you treat something as a technology project, you measure delivery. Specs met. Timeline hit. Budget consumed. Done.
When you treat something as a capital investment, you measure return. You start with a financially modeled business case before anything is approved. You define the baseline metric. You set the target. You assign ownership of that target to a named individual or function. You put governance checkpoints into the delivery process. And post go-live, you run monthly audits comparing actual performance against the original forecast.
Before. Target. Actual. That is the entire framework. It is not complicated. The discipline to maintain it consistently across a portfolio of dozens or hundreds of projects is where most organizations fall apart.
The concept of a Value Realization Office is starting to emerge, and it makes complete sense. A dedicated function, owned by finance rather than the CIO or the transformation team, with the mandate to govern AI spend as capital allocation and track outcomes with financial discipline. This function should have existed from day one of the first enterprise transformation program, not just now. But better late than never.
The critical point on ownership: value realization cannot sit with the CIO. It is a conflict of interest. The technology leader is evaluated on delivery, on deployment, on adoption. Giving them ownership of the ROI tracking is like asking the vendor to grade their own renewal pitch. Finance needs to own this. The financial discipline is the point.
The failure happens at the beginning and the end
Alex was specific about where the breakdown occurs, and it matched my own experience running automation programs at scale for large global corporations.
The middle is fine. The delivery teams are good. AI practitioners are capable. The technology works.
The failure is at the start and at the finish.
At the start, poor use cases get approved. Not because nobody saw the problems, but because organizational politics pushed them through. A senior stakeholder wants a particular use case. Nobody has an intake engine rigorous enough to filter it out. The business case is optimistic. The project goes live. The value is not there. And the same stakeholder who forced it through is the first one asking why the ROI did not materialize.
If you fix the intake stage, if you build a rigorous prioritization engine that filters use cases by financial viability before they consume delivery resources, you eliminate a significant portion of failed investments before they start.
At the end, the tracking stops the moment the project is marked complete. The thread between the original business case forecast and the actual post-live performance is cut. Nobody picks it back up. The outcome disappears.
AI is not a project. It is an operation. There is no clean end date. You need to track performance continuously, understand which solutions are delivering, which need investment, and which need to be terminated. That requires a platform and a discipline, not a one-time close-out report.
What good looks like in practice: the Aviva case
Aviva’s claims innovation team is a real example of what this looks like when it works.
When Alex first engaged them, their entire AI and automation pipeline lived in a combination of Miro boards, Post-it notes, and spreadsheets. The manager’s words were direct: “I can’t do anything with this information.” There was no engine to prioritize, no structured intake, no way to know which use cases to pursue or what the financial case was for each one.
Within a day or two of deploying SilkFlo, 102 qualified opportunities were structured and visible inside a single platform. Employees logged their pain points directly. An AI intake layer cleaned up the raw inputs into project charters that business analysts could actually work with. Over the following two to three weeks, the team ran structured one-on-ones across business units, building out fully modeled business cases across that pipeline.
The result: approximately £12 million in identified AI pipeline, ready to be prioritized and tracked through to delivery and post-live value realization. The entire process costs 90 times less than using a single contracted business analyst from an external consultancy.
That is not a technology story. That is what happens when you apply the right governance framework and the right tooling to the intake and tracking problem.
The outcome-based model is now the only model that survives
Something important is shifting in how consulting and technology engagements are being structured.
For years, I have been vocal about outcome-based models. Clients want to buy results, not effort. They want to pay for impact, not time and materials. The industry is finally catching up. Even large firms are transitioning toward outcome-based commercial structures because clients are no longer willing to fund lengthy engagements with uncertain returns.
That shift makes value tracking infrastructure non-negotiable. If you are a consultancy or a software vendor selling outcomes, you need to be able to prove what you delivered. If you are a client buying outcomes, you need to be able to verify what was delivered. The outcome-based model and the value tracking capability are two sides of the same requirement.
JIRA is excellent at tracking tasks. Azure DevOps is excellent at tracking delivery. Neither of them tracks ROI. They mark things as done. For them, done means delivered. For a CFO, done means nothing. Day one of post-live is where the ROI clock starts, and most organizations have no infrastructure to track what happens from that point forward over the following three years.
The framework, plain and simple
For anyone who wants to start getting this right without a major platform investment, here is the minimum viable governance model:
Before any AI project is approved, define the metric you are trying to move. Is it the cost per transaction? Conversion rate? Headcount per revenue unit? Compliance incident rate? Name it. Measure the current baseline. Set the target. Assign a named owner.
During delivery: run stage gates that check whether the use case still makes financial sense as requirements evolve. Kill it early if it does not.
Post go-live: run a monthly audit for at least twelve months comparing actual performance against the original forecast. If it is underperforming, diagnose why. If it is overperforming, understand why and look for replication opportunities.
That is the framework. A spreadsheet can support it for the first few use cases. Once you are managing a portfolio of ten, twenty, or fifty AI initiatives simultaneously, you need infrastructure designed for this purpose. The complexity of tracking diverse use cases across business units, each with different metrics, different baselines, and different timelines, is not a spreadsheet problem anymore.
The question every technology leader needs to answer this week
Not: “What AI tools are we deploying?”
This one: “Who in this organization owns the outcome of each active AI investment, and what is our process for tracking performance against the original business case after go-live?”
If you cannot answer that clearly, you are in the 95%.
You may have green KPIs. You may have a vendor dashboard full of impressive adoption numbers. You may have a delivery team that is proud of everything they have shipped.
But if your CFO cannot see it in the P&L, none of that counts.
What to do right now
Two actions.
Watch the full conversation with Alex Leonida. Everything covered in this article goes deeper in the video, including the full Aviva case walkthrough and the mechanics of the SilkFlo platform. Link in comments.
Then go to SilkFlo.com. If you are scaling AI adoption and you do not have a structured intake, prioritization, and post-live tracking process, book a conversation with Alex’s team. It is the fastest way to understand where your value tracking stands and what it would take to close the gap.
I have been delivering AI and automation programs on an outcome-based model for years. I do not sign off on technology that cannot be connected to a verifiable result. SilkFlo is the infrastructure that makes that discipline operational at scale.
That is why it is OG Approved.
More from OG: all articles · subscribe on Substack