Judson Tomaiko · AI, decisions, and how work gets done

The usual question

Where can we add AI?

The one I ask

What decisions, workflows, and business models would we redesign?

I'm Judson Tomaiko. I look at how work actually happens, find the decisions where a better prediction would change the outcome, and build working prototypes to test the idea before anything scales.

See the projectsRead the framework
01Point of view

What I believe about AI in business

  1. 01

    AI lowers the cost of prediction. Value appears only when a better prediction changes a decision.

    A model that produces a number nobody's workflow is set up to act on adds cost, not value.

    See it in: Market Expansion Prioritizer →
  2. 02

    Start with the workflow, not the model.

    Ask what would change if a useful prediction became cheap, fast, and available at the moment of action.

    See it in: Workflow Assessment Tool →
  3. 03

    Redesign before automating.

    A broken process with AI is still a broken process, only faster. A simpler fix often gets most of the benefit first.

    See it in: Diagnosis before tools →
  4. 04

    Treat automation as a risk decision.

    Data quality, the cost of an error, reversibility, and human oversight decide how far automation should go.

    See it in: Customer Escalation Triage →
  5. 05

    Build the feedback loop.

    Outcomes, corrections, and exceptions should improve the next decision, not get logged and forgotten.

    See it in: Model Drift Monitor →
02The adoption ladder

Using AI is not the same as being changed by it

Only 39% of companies report any enterprise EBIT impact from AI (McKinsey, 2025). The first two rungs make individuals faster without changing how the business works. The value is further up.

  1. Level 1

    Individual productivity

    People use AI to write, summarize, brainstorm, code, and analyze. It makes individuals faster. Nothing else changes.

  2. Level 2

    Task assistance

    AI helps with pieces of work, but the underlying workflow stays the same. The process is unchanged.

  3. Level 3

    Workflow integration

    AI enters real operating processes and starts changing handoffs, reviews, approvals, and exceptions.

  4. Level 4

    Decision redesign

    AI changes when a decision happens, who makes it, what data supports it, and what action follows.

  5. Level 5

    Organizational learning

    Daily use creates feedback. Feedback improves prediction. Better prediction changes the business. This is AI-native.

03Automation

Automate it, or keep a person in the loop?

Automate when

  • Data is strong and prediction confidence is high
  • Mistakes are low cost and recoverable
  • The action is reversible
  • Feedback arrives quickly so the model can learn
  • Rules are clear and the environment is stable
  • Ethical risk is low

Keep a person in the loop when

  • Stakes are high and mistakes are hard to reverse
  • Data is thin or the situation is genuinely novel
  • Trust or fairness is at risk
  • Accountability must be traceable to a person
  • Customer relationship depends on human judgment
  • Regulatory or legal exposure attaches to the outcome
04Projects

Ideas I've tested by building them

Each prototype starts with a business decision. Some use AI and some deliberately don't, and each one says plainly what it can't do.

  1. 01

    Market Expansion Prioritizer

    Which markets should your organization enter first, and how sure should you be? This weighted model runs on synthetic data and shows why the ranking moves when the criteria change.

  2. 02

    Customer Escalation Triage

    When is it safe to let software route a customer complaint before a person sees it? Paste in a complaint and see how Claude and a simple rule set would handle it.

  3. 03

    Workflow Assessment Tool

    Where is your process stuck, and what kind of fix does each step need? Describe a workflow and get a structured breakdown, step by step.

  4. 04

    Visual Quality Control

    Once a camera flags a possible defect, who decides whether the unit ships? This simulation lets you set the confidence it takes to ship a unit untouched.

  5. 05

    Data as Fuel for AI

    How much should you trust a model's output when you can't see what it learned from? Train a small model in your browser and watch what happens when the data runs short.

  6. 06

    Model Drift Monitor

    A model worked at launch. How would you know when it stops working? This simulation tracks accuracy over 24 weeks so you can decide when to retrain.

  7. 07

    Adaptive Personalization with AI

    What does the feedback loop behind personalization look like for your customer experience? This browser-based replica shows an example.

05Writing

Writing on AI and decisions