2026.27: Two Days and a Facilitator
Happy Friday friends! I'm back in "The Six," and I'm enjoying the last couple of weeks of summer.
I ask (a version of) the same question in every room I stand in front of. Doesn't matter if it's a cohort at Alberta Women Entrepreneurs, a session with WeBC, or a classroom of my students at SAIT.
Think about the last time your team did real strategic planning. The offsite, the retreat, the two days with the butcher paper and the facilitator nobody liked. How many genuinely different options did you walk out with?
Somebody always says three. Sometimes four. Memorably, a founder said one, and then added, "well, one that my board would let me say out loud," which is a different newsletter entirely.
Then I ask the follow-up, and the room goes quiet.
Were there only three good options? Or were three all you had the hours for?
There's a new piece in the upcoming September–October issue of Harvard Business Review by Felipe Csaszar, who chairs the strategy area at Michigan's Ross School. He puts a proper name on what I've been circling in those classrooms for many years: bounded rationality. Our brains and businesses have a ceiling. Finite attention, finite memory, finite processing, finite resources.
And then he says the thing that made me put the iPad down.
SWOT has four quadrants, the growth-share matrix is a 2x2, and Michael Porter's famous framework has exactly five forces, not because the competitive world is that tidy. Because those frameworks had to fit on a whiteboard, in a morning, in front of tired people.
Every one of those frameworks is a photograph of a human attention span. Four boxes. A 2x2. Five forces. Not a complete map. Just a map of what a tired team can hold in its head at once.
And we've been making our most important decisions off them ever since.
Let's break it down.
Signal:
YOU DIDN'T RUN OUT OF OPTIONS. YOU RAN OUT OF TUESDAY.
Csaszar names three constraints on any strategic decision. I've watched all three kill good companies, and not one of them is about how smart the people in the room are.
The first is SEARCH. How many alternatives can you actually generate and judge? A planning cycle surfaces a dozen ideas, narrows to three or four, and the team spends every remaining ounce of energy arguing about the finalists. The rest of the possibility space never gets touched. Not because it's worthless. Because there are no hours left in the day.
Csaszar points to an M&A case documented by McKinsey. A software company ran a scouting system across a database of more than 40 million public and private companies. Patents, filings, expert transcripts, all the messy stuff. It surfaced and scored more than 500 acquisition targets in under a day. Narrowed to 15 serious leads. Three of those closed within months.
The speed is not the story. The story is that no human team was ever going to see target number 340.
Running out of Tuesday is the cheap version of this problem. Here's the expensive one.
The second constraint is REPRESENTATION. How good is your model of the world you're operating in? For most companies it's a market map from last year, a segmentation deck somebody built in 2023, and a forecast that arrives a quarter after it would have been useful. Those tools capture the big dynamics and drop everything else. They drop the nuance, the context, and worst of all, the exponential change happening right now.
MYbank, the digital lender inside Ant Group, is the example that stuck with me. Traditional lenders score a small business on a handful of variables: revenue, collateral, credit history. Miss one and you're not a bad risk, you're an unknown one. Unscorable. Unfundable. Invisible.
MYbank rebuilt the model on more than 3,000 variables: transaction data, supply chain relationships, business networks, satellite imagery. They call the result 3-1-0. Three minutes to apply, one second to approve, zero humans in the loop. More than 53 million small businesses have taken credit through it. Seventy-two per cent of them had never borrowed before. Default rate sits around one per cent.
Read that again, because the lesson isn't about lending speed.
They didn't serve their existing customers faster. They saw a customer segment that their old model had rendered invisible. Tens of millions of viable borrowers who were sitting right there the whole time, under a lens too coarse to pick them up.
Unilever did the same trick in the ice cream business, of all places. Weather probabilities, demand signals, and telemetry off roughly three million connected freezers across 35 factories, refreshed continuously instead of quarterly. Ten per cent better forecast accuracy in Sweden. Sales up as much as 30 per cent in some regions.
Neither company went out and found a new market. They built a lens sharp enough to see the one that was already sitting in front of them.
The third constraint is the one no tool has ever solved, because it isn't cognitive. It's social.
AGGREGATION. How does your organization turn a group of smart people into one decision?
You already know how it actually goes. The most senior voice carries more weight than the argument deserves. Half the room is measured on numbers that quietly conflict with the other half's. Dissent costs something, and everyone in the room has already priced it. So the group converges fast, and calls the speed alignment.
Three ceilings. Search, representation, aggregation. Your team hits all three within the first hour, and every one of them is a limit on the process, not on the people.
Scale:
BUILD A BOARD THAT ARGUES BACK.
Let's stay with the last one for a minute. The third one. Aggregation is the ceiling I get asked about most, and it's the cheapest one to fix. No new data. No new frameworks. Just a different room.
Csaszar's answer is what he calls synthetic deliberation. Assign roles: one agent makes the strongest case for the plan, one tries to take it apart, one plays the competitor or the regulator. There's real evidence behind it. A field experiment at Procter & Gamble with 776 commercial and R&D professionals found that one person working with AI matched the average output quality of a two-person team without it. Teams using AI ran about 12 per cent faster and produced more top-decile work. The finding I care about most: it cut the silo effect between the commercial people and the R&D people. Each side finally had access to the other's lens.
I've been running my own version of this for a while. I call it my Virtual Board, and it's ten seats:
Mark Cuban. Elon Musk. Steve Jobs. Oprah Winfrey. Seth Godin. Cal Newport. Mitch Joel. Tom Peters. Nilay Patel. Scott Galloway.
Now, the part that matters.
I do not agree with all of those people. I have real, load-bearing values conflicts with a few of them, and I'm not going to pretend otherwise to make the list look tidier. Musk is on my board precisely because he is on my board. He questions limits in a way that offends my sense of how you're supposed to treat people, and roughly one time in five he breaks a frame I didn't know I was standing inside.
That's the job. The board is not there to agree with me. It's there to refuse to.
Cuban cuts to whether anyone will pay. Newport asks what I'm giving up to do this. Galloway says the uncomfortable number out loud. Patel asks what happens when the lawyers show up. Peters asks whether I've talked to a single actual human who does the work. They never sleep, they cost me nothing, and when one stops being useful I retire the seat.
But here's the risk in all of this, and I want to say it plainly rather than let it slide past.
Csaszar's own research found that when AI evaluated business plans, its assessments lined up with the investor panel's average judgment more closely than individual investors' did. He offers that as evidence of consistency, and it is.
It's also a warning.
A system that reliably converges on the mean is a system that will never hand you the one dissenting read that saves you. That's not a board of directors. That's a very polite committee that happens to be made of software. The failure mode of a human strategy meeting is groupthink. The failure mode of a synthetic one is a machine that agrees with you in ten different voices.
Which is why the diversity has to be designed on purpose, the conflicts have to be real, and the whole thing is worthless the day you start using it to feel better about a decision you already made.
I don't want a board that praises me. I want a board that engages me.
Deep Dive:
EVERYONE HAS THE SAME AI. SO WHAT.
Csaszar reaches for the internet in 1995 to explain why everyone having the same AI doesn't level the field. Access was becoming universal. The technology was not anyone's secret. Amazon, Netflix and Google did not win the next two decades because they had a website. They won because they treated the thing as a foundation to rebuild on, not a feature to bolt onto what they already had.
Let me add a chapter he didn't write, because I was in the room for both of them.
Yes, both. I have shipped software on a CD, in a box, with a printed manual inside it. I am, functionally, a fossil with a Substack.
For anyone who was in kindergarten during dial-up, let's use the wave you might actually remember.
Then we did it again with mobile, between 2005 and 2008. Enterprises bought phones and tablets by the pallet, then handed people the same workflow they'd had on the desk, now on a smaller screen, in the rain. It failed fast and it failed loudly. The organizations that got anything out of mobility were the ones who threw out the process and rebuilt it around what a person could now do standing in a field.
Which brings me back to the tailboard project I told you about back in 2026.20, Same Rodeo, Different Bronco. An energy company asked for their paper safety form on an iPad. We refused the brief and rebuilt the process from the crew up instead; the iPad turned out to be almost incidental. In under five years that app has mitigated 847,000 hazards and taken 185,000 sheets of paper a year out of the field.
Refusing the brief and widening the option set before you narrow it are the same muscle. That's the connection I hadn't made until this week.
There's a model I've used here before and it fits Csaszar's moat argument better than anything in the article itself. Dr. Ruben Puentedura's SAMR: Substitution, Augmentation, Modification, Redefinition. I used it in 2026.10, You Lost the Plot to explain feature stuffing, and it does the same work here.
Buying the same AI your competitor bought and pointing it at the process you already have is Substitution. It raises the floor for everybody and distinguishes nobody.
Modification and Redefinition are where the moat lives, and every example in the article is one of those two. Morgan Stanley wired an assistant into the research workflow 16,000 advisers already lived in. A hundred thousand internal documents, 98 per cent adoption, retrieval efficiency from 20 per cent to 80 per cent. John Deere's See & Spray scans 2,100 square feet a second and makes 5,000 spraying decisions a minute, cutting herbicide use by 50 to 77 per cent, and quietly turns an equipment manufacturer into an agriculture platform. Duolingo shipped fast after GPT-4, but the moat is Birdbrain, trained on 500 million users doing 1.25 billion exercises a day.
The AI tools everyone can buy are the commodity. What you build on top of them, out of your own data and your own process, is the thing nobody can.
Here's the instruction I give from the stage, and I'm giving it to you now.
Before the next big call, whether that's an acquisition, a market entry or a product bet, make the machine produce the long list your team never had the hours to build. Filter it hard on your criteria. Then spend your people's judgment on the short list.
AI expands the search. Humans choose.
That's the whole move, and it's the one to start with.
I'm working on the other three, and they'll come in the weeks ahead: replace the snapshot with a living model, build the structured challenge in on purpose, and shift the strategist's job from analyst to architect.
You don't have to wait for any of that to start on the board, though. My entire Virtual Board document is yours, today, HERE. All ten seats, how I brief them, how I make them argue instead of agree, and a blank template at the end.
Build your own rather than borrow mine. Mine is calibrated to my blind spots. Yours should be calibrated to yours.
Picture your next offsite.
Same two days. Same facilitator. Nobody likes him any better.
But your team walks in having already read a scan of five hundred possibilities, cut to the fifteen worth arguing about. The market map on the wall is current, not six months stale. And nothing gets presented until it has survived a critic, a competitor, and a customer who was never going to be impressed.
The conversation in that room gets sharper. Not because the machine made the decision. Because the machine cleared out the work that was eating the time you needed for the only questions it can't touch.
What do we believe. What are we willing to risk. What kind of company are we trying to become.
Those are still yours. They always were. You finally have the two days free to answer them.
So, how many real options came out of your last planning cycle? Be honest about the number. Hit reply and tell me. I especially want to hear from anyone whose answer is one.
The newsletter isn't the conversation. The conversation is the conversation.
See you next Friday.
Best,
JT
PS - If someone forwarded this to you and you want it in your inbox directly, subscribe HERE.
Sources
Felipe A. Csaszar, AI Is Revolutionizing Strategic Decision-Making, Harvard Business Review, September–October 2026 (Reprint R2605B). All company examples in the Signal and Deep Dive sections (McKinsey's M&A scouting case, MYbank, Unilever, Procter & Gamble, Morgan Stanley, John Deere, Duolingo) are drawn from this article.
Csaszar et al., experiments on LLM-generated business plans and investor evaluation, via the HBR article above.
Dr. Ruben Puentedura, SAMR model, via 2026.10
Tailboard / energy company project figures, via 2026.20
