Why CFOs Should Fund Agents Like Value Streams

Koko: An enterprise can be spending heavily on AI and still be unable to fund it. Those are two different problems.

Sam: That sentence stopped me cold. Because those feel like the same thing, right? You're spending — you're funding it.

Koko: And that's exactly the trap. Spending is easy — you buy seats, tokens, licenses, run a pilot, spin up a copilot. Funding, in the CFO's sense, means committing scarce capital against a return you can name and defend.

Sam: And most AI spend today can't clear that bar.

Koko: Because it's measured by usage, not outcome. And usage is not a return.

Sam: Seats provisioned. Tokens consumed. Prompts served. Sessions logged.

Koko: Every one of those is a measure of activity. Not one of them tells you whether the using produced value.

Sam: And a cost that can't point to value is, by default, a cost to cut.

Koko: Which is why the pilot stalls. Not because it failed — it may have worked beautifully — but because its impact was never attributed to a number anyone owns.

Sam: It loses the budget fight to a project that can name its number. Every time.

Koko: So that's the problem the piece is built around. The piece is called Why CFOs Should Fund Agents Like Value Streams, and its thesis is pretty surgical: stop funding agents like IT projects and start funding them like value streams — with a value card on every agent, and a blunt gate. No value card, no scale funding.

Sam: No card, no money. I appreciate a framework that comes with a bouncer.

Koko: The underlying premise is that the frontier model — the actual AI — is basically a commodity now. The advantage lives in the harness around it.

Sam: And the harness is what attaches a value card to every agent.

Koko: Right. And there's this line that I think explains why this argument lands right now: adoption is near-universal, agent deployment is in the single digits, and the gap is not capability — it is fundability.

Sam: So a thousand people happily using a copilot, and the enterprise still won't fund a single agent at scale.

Koko: Because the copilot's usage is visible and its value is not. The constraint on scaling isn't the model. It's the absence of a credible answer to what does this buy, and who owns the answer.

Sam: Which is a funding-model failure, not a technology failure.

Koko: Exactly. The agent was scoped, built, and measured like an IT project. And IT projects are justified by delivery — did we ship it — not by which lever they move.

Koko: So let's get into the mechanics, because this is where it gets interesting. The value card is five fields. Baseline, target, value lever, owner, and attribution method.

Sam: Five fields. Deliberately small enough to fit on one line of a portfolio review.

Koko: And the piece is very specific about what each one means. Baseline is captured before the agent touches the process — not reconstructed afterward. A target with no date is a wish. And the owner is a named executive accountable for the number, not for the agent.

Sam: That last one is load-bearing. The owner already owns the working-capital line or the close calendar. Not the team that built the bot.

Koko: Engineering owns whether the agent runs. The owner owns whether it mattered.

Sam: I want to sit on that for a second because I think that distinction is where most internal AI programs are failing right now. You're asking the builders to own the business outcome.

Koko: And then you're surprised when the outcome is measured in builds.

Sam: Right. Okay so the card is the instrument. What's the spine?

Koko: The value spine is the chain that makes the card credible. It has five links: agent activity, then process metric, then financial or risk metric, then executive value lever, then shareholder value.

Sam: So you walk it end to end and see where it breaks.

Koko: That's exactly how the piece frames it — it's a test, not a diagram. And the most common failure is what it calls a missing middle vertebra: a process metric that improves but can't be connected to anything in a management report.

Sam: The agent is busy, not valuable.

Koko: And the harness's job is to refuse to fund those. Not dress them up. A spine with a gap is a story. A spine that holds end to end is an investment thesis.

Sam: I like that framing. And the piece actually works through a concrete portfolio — five agents in the office of the CFO.

Koko: Finance is where it's easiest to make concrete, because the levers are clear and the processes are well understood. You've got the Month-End Closer agent moving close cycle time, owned by the Controller. The GL Reconciler clearing recon aging. Billing Dispute Triage pulling dispute aging down — that's working capital.

Sam: Because every day off dispute aging is cash pulled forward.

Koko: Then Cloud Spend — pure margin agent, watching commitment utilization and idle resources. And Usage and Rating Integrity, which is reconciling metered usage against what was actually invoiced.

Sam: Catching the gap between what you should have billed and what you did.

Koko: Unbilled consumption is revenue left on the table and margin quietly bleeding. The piece calls it leakage rate. Which, honestly, is a great name for a metric.

Sam: Very evocative. You can feel the money leaving.

Koko: And notice what the table isn't — it's not a list of AI capabilities. It's a budget, sorted by lever, with an owner behind each row.

Sam: Okay, here's where I want to push a little, because all of this depends on the attribution method being honest. And attribution is where I've seen a lot of business cases quietly cheat.

Koko: The piece agrees with you. It says attribution is where most AI business cases quietly cheat — by claiming the full movement of a metric while ignoring everything else that moved it.

Sam: Dispute aging fell thirty percent! Was that the agent or was that a seasonal lull?

Koko: Or a pricing change, or a billing-system upgrade, or a headcount addition. The honest method is baseline versus counterfactual — attribute only the delta against what the metric would have read without the agent.

Sam: Not the raw before-and-after, which sweeps in everything else that happened in the period.

Koko: And where a clean counterfactual is impossible — which is often — the next-best methods are stated on the card with their limitations. A holdout segment, a staggered rollout, confounders netted out explicitly.

Sam: Stated, not hidden. That's the discipline.

Koko: And the corollary is uncomfortable: an agent whose impact cannot be attributed gets no scale budget. Not a reduced budget — no scale budget.

Sam: It can keep running as a pilot. It can keep gathering evidence. It just doesn't graduate to a line in the plan.

Koko: Until its card has a number that would survive an audit.

Sam: Now here's where I'd push back slightly on the piece. That standard — survives an audit — can become a convenient reason to never fund anything. Attribution is genuinely hard. If the bar is airtight causality, you're going to leave a lot of real value on the table.

Koko: That's a fair tension. But I think the piece anticipates it — the bar isn't airtight causality, it's a defensible number. The holdout and the staggered rollout both qualify. It's not demanding a randomized controlled trial, it's demanding that you've actually thought about what else was moving.

Sam: So the test is: would you be embarrassed to show this attribution to a skeptic.

Koko: And the piece makes a really sharp point about why this matters beyond any individual agent. One over-claimed business case, discovered later, taxes the credibility of the entire portfolio. And the trust tax on AI is already high enough.

Sam: That's not abstract. If your first three agents oversell and underdeliver, you don't get to fund the next ten.

Koko: So let's make this concrete for the person listening. What does this actually change about how you run the portfolio?

Sam: Three things. Fund by lever, kill by metric, and own the conversion.

Koko: Fund by lever means capital goes to the five levers — revenue, margin, working capital, productivity, risk — not to technology. The portfolio review isn't which AI projects are we doing. It's what are we spending against working capital, and which agents carry that spend.

Sam: The lever is the unit of allocation, not the model.

Koko: Kill by metric is exactly what it sounds like. An agent that moves its target metric earns more scope. One that doesn't, after a fair window, gets killed. Not iterated forever, not left running because someone built it, not protected because it demos well.

Sam: Killed, and its budget recycled to a lever that is moving. The willingness to kill is what makes the funding credible.

Koko: A portfolio with no kill rule is a slush fund. And a board reads it as one.

Sam: And the conversion — this one I think is the most strategic of the three.

Koko: Funded as a project, an agent is an expense that expires. The budget is consumed, the quarter closes, and the capability is effectively rented for the life of the invoice. Funded as a value stream, the same agent is an owned, governed capability that compounds.

Sam: It keeps moving its lever quarter after quarter. Next year's plan starts from a higher baseline, not a fresh ask.

Koko: And the piece makes the case that this conversion is the CFO's to own specifically. Only finance can decide whether a given dollar is being spent or invested. Only finance can hold the kill rule.

Sam: So if you're a CFO and your AI budget is still sitting in IT, that's the first thing to change.

Koko: And if you're on the business side — if you own the DSO line or the cloud P&L — the piece is asking you to pick up ownership of the agent that moves your number. Not just the number. The agent.

Sam: Because otherwise you're waiting for IT to tell you whether your own metric improved. Which is a strange way to manage a business.

Koko: The piece ends with a forward view that I think is worth taking seriously. It says expect the value card to become a procurement and governance artifact — agents that arrive without a defensible card simply don't get production access.

Sam: So it's not just a finance discipline. It's the gate at the door.

Koko: And it says expect owned AI capability to enter the CFO lexicon as something reported — the way other intangible assets are — as the compounding side of the value-stream model accrues.

Sam: Which would be a real shift. Going from AI spend as a noisy line item to AI capability as a governed asset on the books.

Koko: The card is the funding instrument. The spine is the test that keeps it honest. The kill rule is what makes the portfolio credible. Together they are how the CFO converts AI spend into something the enterprise actually owns.

Sam: Which is the entire argument in Why CFOs Should Fund Agents Like Value Streams. And I think the question it leaves me with is: what happens to the enterprises that don't make this shift? Because if some CFOs are building compounding, governed capability and others are still running slush funds — that gap doesn't stay even.

Koko: That's exactly the right question to sit with. The full, source-verified piece — every number, every framework, every attribution method — is on Koko Knows. Go read it, especially if you're heading into an AI budget conversation this quarter. You'll want the spine.