Bayesian Leaf

The problem is rarely the interface. It's the decision infrastructure underneath.

Fractional product leadership for AI-native products that decide, recommend, and automate. I work from what your system actually does, find the thing your team hasn't seen, and hand back something buildable.

For founders, CEOs, CTOs, and product leaders at seed to Series B AI-native companies.

Your team already ships with AI. What is usually missing is the decision infrastructure: the product logic, evidence, and controls that turn raw model capability into something customers can actually use.

Book a working session See engagement and pricing

Thirty minutes. Bring what's stuck. No deck, no pitch. If it isn't work I do, I say so.

Currently taking one new engagement.

Previously at Meta, Pinterest, and Opendoor.

Why people reach out

The product works. It still needs too much help to succeed.

Most useful when adoption depends on trust, control, workflow, or operational judgment.

Usually someone has already tried the obvious moves: a contract designer, a redesign, another round of onboarding. Activation still does not move, or every new customer still needs a person in the loop.

Case study

NºWho it was forCame in withLeft with

An applied AI platform used in high-consequence work.

The product worked. Nobody could point to the screen where it lost people, and every new account still took hands from the team.

A scoped onboarding became standing product and design direction.

An opaque system became legible. The screens barely changed.

The Decision Read v2.2.3

Bring a situation. See what the method notices.

Start with how your team makes product decisions. If it is not work I do, I will tell you that too.

I test each version against cases it should turn away. If it recommends me too often, I don't publish it.

Your model works in the demo and stalls in production, and you can't tell where the trust breaks.

You're shipping AI features faster than your team can decide which ones actually matter.

You need senior product and design judgment in the room now, and you're not ready to hire a full-time executive.

Get in touch Free. No signup.

The Decision Read applies the Bayesian Leaf methods to your situation. Your answers are used only to generate the read, and I don't see them unless you choose to send it to me afterwards.

01 / 06
Reading your situation…

The Decision Read v2.2.3 · running the Bayesian Leaf method

Your read

Send me your read

Leave it with me and I'll reach back out.
Your read and your answers come with it, so when I get back to you, I already know your situation.
The read you're sending

Your answers are sent to Anthropic's API to write the read. I never see them unless you choose to send the read to me. No signup.

Got it. I'll be in touch.
I read every message personally, and I'll come back to you with your read in hand.
Ready

Selected writing

ForthcomingNotes and essays as new thoughts, discoveries, and outcomes arise.

Subscribe on Substack

Engagement

What it costs, and what you get

You're buying the judgment that decides what gets built.

The working session comes first. I scope the proposal after I understand the work.

Direction Sprint Four weeks
Set a direction the team can carry.
$30K
Good for
The consequential decision is known. A direction the company can commit to is not.
What's produced
  • A direction document with named principles the team can argue with.
  • A running high-fidelity prototype with the direction inside it.
  • The sequence the team can carry after I leave.
What changes
The team can commit to a direction and argue with it, instead of relitigating strategy every sprint.
Embedded Monthly
Keep the hard decisions moving while the team ships.
Works alongside an existing product lead who has more surface than time.
$18K to $25K/mo Set by scope
Good for
Product decisions are stacking up faster than the team can get to them, and hiring is not the answer this quarter.
What's produced
  • Decisions written clearly enough for the team to build against.
  • Product direction and release sequence kept current as the work changes.
  • Hands-on product and design work when the direction needs proving.
How it works
The work can move from advisory through hands-on execution. Priorities, ownership, and price are set in the monthly scope. The context does not reset.
What changes
Product decisions stop waiting on one person's calendar.
System Read One week
Read the product from the codebase outward.
$12K
Good for
The right decision is not yet clear, and you want the gap named before committing to either of the other two.
What's produced
  • A written read of the system: file paths, line references, the gap between product and interface.
  • The decision the product is waiting on, named.
  • A working high-fidelity prototype that proves the smallest important point.
What changes
You know what the gap actually is before you spend a quarter on it.
First two weeksWhat happensWhat I need from you
Day one

A 30-minute working session. No deck, no pitch. Bring whatever is stuck, and we find the one decision the rest is waiting on.

Thirty minutes, and whoever owns the product decision.

Week one

A read-only pass through the codebase, then the first written read. The work starts from what is already true.

Repo access, read-only. Product access.

Week two

The first direction, or a working prototype, in front of the team. For a System Read, week one is the whole engagement.

For Embedded, standing time from your side, set in the monthly scope.

The system knows more than the interface says, and more than the people running it remember.
Book a working session

The method

The work moves from the system itself
to a direction the team can carry.

01

Map the system

Most of the problem is already visible. It has rarely been drawn in one place.

02

Find the load-bearing decision

Until it is named, the roadmap stays full of plausible directions.

03

Make it legible

A person should understand what to do and why, without being taught. This is where design earns its place. Not as the fix, but as how the decision becomes something a team can build against.

04

Build the trust layer

AI changes what a person has to believe before they act. Trust has an anatomy: the evidence, the controls, the way back when it's wrong. I design that.

05

Ground it in measurement

If we can't tell what changed after release, we don't know whether the decision made the product better.

06

Sequence for compounding

I order the work so each release makes the next one easier. That's what compounds.

About | View full work »

Hew Suber, MBA

Seventeen years building decision-heavy product systems across Meta, Pinterest, Opendoor, and early-stage companies. At Pinterest that meant an AI automation product built from nothing, where more than thirty setup decisions collapsed into three inputs. It cleared its first-year target by 57 percent. The work spans automation, self-serve, operational modeling, and AI-native products, including embedded product and design work across a Techstars Atlanta cohort.

I work where product, design, and engineering have become the same problem. Every engagement is with me. I hold the direction and move into the work without handing off the context. I read the codebase before I design, so what I hand back is grounded in what the system can actually do. The team knows what is hard and what only sounds hard before it commits.

Outside client work, I build consumer products, Sage Orange and Presence, under Human Commons.

Hew Suber

Let's talk

Bring what's stuck.

A 30-minute working session. No deck, no pitch. We find the one decision the rest is waiting on, and I scope the proposal after I understand the work.

Not a fit: teams still looking for the product, or teams that want a pair of hands for a backlog they have already decided. If it isn't work I do, I say so on that call and point you to someone better.

Send a note

Atlanta, GA · Working worldwide

Book a working session