Strategy moves fast. Funding moves annually. On November 5, Adam Mattis, Head of Product and Strategic Alliances at TSG, and Olivier Fleischmann, Head of Revenue, Digital Business and IT at Planisware, showed portfolio and finance leaders how to close that gap with one decision discipline: fund the experiment, not the plan.

Instead of committing a year of money to a bet nobody has tested, you write down what would end it, release money one round at a time, and let evidence decide what happens at each gate. Watch the full session below, or read on for the highlights.

Key takeaways

• Approved money rarely moves. Once a bet is funded, stopping it gets harder every month, so most portfolios track programs constantly but judge them almost never.
• Write the stop condition first. Every bet gets a one-page evidence charter, signed before the build, that names the result and date that would end it.
• Every gate has four exits. Go, Pivot, Stop or Retest, and the only agenda item is the next round of money.
• Fund in rounds, not years. Finance keeps its annual plan. Money inside one budget line is released in steps, each tied to evidence written down in advance.
• Score every Go. Compare what each bet predicted with what it delivered, and use the gap to set sharper thresholds next time.
• AI prepares the evidence. People make the call. AI can draft claims, forecasts and recommendations, but a named person owns every output and the gate stays human.
  
  

Why funding lags strategy

Adam opened with a meeting most portfolio leaders have sat in. A program has run long enough that few people remember its original business case. Its status is green because the team keeps hitting milestones it wrote for itself. Nobody raises the real question, because that would mean asking a sponsor to admit, in front of peers, that a call they championed was wrong.

Adam then polled attendees: in the last 12 months, how many of your gate or steering reviews ended in a Stop? Most answered none, or one or two. One or two, he noted, is actually a healthy sign, because it means an organization is willing to walk away from something. The more common pattern is that once money is approved, it stays deployed as originally planned. TSG's Tiffany Huckaby explores this in her recent post on strategic portfolio management, including how leaders can decide what to prioritize, pause or stop when plans change.

The pattern is common, and the research behind it is long-standing:

• Money stays where it was put. A McKinsey study of more than 1,600 US companies from 1990 to 2005 found an average 0.92 year-to-year correlation in business-unit capital allocation. Yet the top third of reallocators earned roughly 30% higher annual shareholder returns than the bottom third.
• Spending raises the price of admitting a mistake. In a 1976 experiment, Barry Staw found that people personally responsible for a bad funding choice put more money into it than people who inherited the same choice.
• Overruns are not rare. A Harvard Business Review study of 1,471 IT projects found that 1 in 6 ran a cost overrun averaging about 200%. And as HBR reported, RCA's videodisc player consumed $580 million over 14 years as the company kept expanding it through a weak market response.

AI raises the stakes

Adam pointed to AI pilots as the sharpest example. For the last three years, almost anything with AI in the title made it through procurement. Those pilots got funded, but few were ever judged, and many organizations are now cleaning up after them.

Gartner reported in January 2026 that at least half of generative AI projects were abandoned after proof of concept by the end of 2025, well above the 30% it had predicted in July 2024. The reasons it cited map directly to missing discipline, not missing technology:

In other words, a better model fixes none of these.

The fix isn't new, either. Stage-gate, discovery-driven planning and venture-capital staging all rest on the same idea: when a claim is untested, the right to stop has value, and releasing money in stages is how you keep that right.

How to fund the experiment: charter, gate, rounds, score

The core of the session walked through four practices, using an illustrative example: a fictional mid-market distributor testing whether independent clinics would pay a monthly fee for guaranteed supply availability.

1. Before the money moves: the evidence charter

Every bet starts with a one-page charter, signed before the build. It has six fields:

1. Claim: the result the bet must produce, stated so it can fail.
2. Stop condition: the result, by a date, that ends the bet. Write this one first.
3. Data: the real records the test will run on.
4. Method: how the claim gets tested, starting with simulation and then the real-world test it can't replace.
5. Thresholds: the results that mean Go, Pivot, Stop or Retest.
6. Predictions: what a Go should deliver later, economics included.

Two people sign. The bet owner holds the money and decides at the gate. The evidence steward opens access to the data. In the distributor example, the stop condition was fewer than 4 letters of intent from 40 clinics, or a replenishment model that predicted under 60% of stockouts 72 hours ahead.

2. At every gate: four exits

Each gate has one agenda item, the next round of money, and four possible outcomes:

• Go: the claim held against its thresholds. Release the next round.
• Pivot: the evidence changed the claim. Fund the new claim with its own test.
• Stop: the stop condition was met. End the bet and keep its evidence.
• Retest: the evidence is too thin to read. Run once more, under rules written in advance.

Why four exits and not three? Most portfolios already have a fourth exit called "continued," which arrives with no new claim, no plan for new evidence and no date. Retest is that exit with rules: the charter allows it in advance, each bet gets one, it comes with a revised charter, and it has a decision date.

Roles matter too. The portfolio office recommends; the bet owner decides. In a McKinsey study of 1,048 major decisions, the quality of the decision process mattered more than the quality of the analysis by a factor of six.

3. Fund in rounds

The distributor never funded a full year up front. Round 1, carved from an existing growth budget line, ended in a Retest because scan data was too thin. Round 2 ended in a Pivot: only 3 of 40 clinics signed letters of intent, but 7 of 25 dental and veterinary groups did. Round 3 earned a Go, with 9 of 30 letters and 74% of stockouts flagged 72 hours ahead.

Finance keeps its annual plan. What changes is that money inside one line moves in steps, each released against evidence written down before the round began.

4. After the Go: score the prediction

A Go isn't the end. Each one is scored against what its charter predicted:

The scorecard also checks the methods. Here, the market simulation overstated demand by about a third, while the digital twin held. So the next charters get tighter market-fit thresholds, and simulated demand gets discounted.

Making the discipline stick

Olivier Fleischmann of Planisware picked up from there with a practical warning: a discipline kept in spreadsheets fades by the third gate. For the charter, the gates and the funding rounds to last, they have to live where the money, capacity, benefits and decisions already live, in whatever portfolio system your organization runs on. In the session, Olivier walked through how that looks in Planisware's platform, which you can see in the recording.

Where AI fits

Olivier also showed how AI can take on the preparation work across the funding cycle. Before a round, it turns intake ideas into draft claims and flags thresholds that no result could miss. During a round, it forecasts cost, resources, timing and outcomes as evidence arrives. At the gate, it lays the evidence against each threshold and drafts a recommendation. After a Go, it pulls actuals at each checkpoint and shows which methods overstated.

Two rules keep that in check. First, a named person owns every AI output, and the bet owner decides: the gate stays human. Second, AI is itself a bet. Every AI initiative gets its own charter, with a stop condition measured on your own data.

Three moves you can make this week

You don't need a transformation program to start. Adam and Olivier closed with three steps any portfolio or finance leader can take before next Monday:

1. Find one stop condition. Take your longest-running funded initiative and look for its written stop condition. If you can't find one in an hour, draft one: a result plus a date.
2. Count your AI pilots twice. First count all of them. Then count only those with a written result, measured on your own data, that would end them.
3. Carve the first round. For the next bet heading to approval, propose one round from an existing budget line, with the next release tied to a gate date.

Adam added that much of this testing can happen before any money moves. Organizational digital twins, simulated customer personas and your own customer data let you run a hypothesis through several rounds at little or no cost, so you can be reasonably sure of a bet before you fund it. It's the same principle behind TSG/ControlBurn, which uses a digital twin to test OT security against a copy of the environment instead of the live system.

Where to start: test the method on your own history

The safest way to try this discipline is on decisions you've already made.

As Adam put it, the AI tooling is the easier part. The hard part is getting your own data somewhere your portfolio team can access it and use it to test.

With TSG: a blind decision backtest. We rebuild three of your past funding decisions from what was known at the time, and a reviewer who doesn't know the outcomes applies the gate. Then we write charters for two or three live bets and recommend which to test first. You can stop there, with all of it in hand.

If your portfolio already runs on Planisware, the charter fields, gate exits and round-by-round releases can live on the investment record itself, so every decision and its evidence stay together.

Want to talk through your portfolio? Talk to our team.