ES
A glass prism splitting a beam of light into colored rays against a dark background

DiaCroma™ for Executives · How it works

Behind the AI,
in every decision.

The company’s platform sees the business and its AI talks to the executive. Every time the AI proposes a decision, it consults DiaCroma: it checks, adds what is missing and monitors the plan. The executive decides; third parties and agents carry out only what was committed, and only while it still passes the checks.

What it asks

What DiaCroma asks of every decision. On three fronts.

Every time the AI consults it, DiaCroma examines the decision with dozens of checks, questions and reviews on the decision, the people and time, and returns the checked answer.

  1. What situation am I in?
  2. What options are there, and what does each one deliver?
  3. Can it be done, and what holds it back?
  4. With whom?†
  5. Who carries it out, and how?†
  6. How long does it take, and what can go wrong?
  7. How is it done, and how is it monitored?

First front

The decision: is it possible?

First of all, whether the option fits the company as it is today: its situation, its options and what holds it back.

A balance in equilibrium: one large blue sphere on one side and three white spheres stacked on the other

What situation am I in?

  • The company’s situation and the shape of the problem: sudden, gradual, seasonal, or stable but bad.
  • What was done before, and whether it worked: it is the first thing looked at.
  • How similar companies fared† with similar plans: the outside view.
  • How much is known about the cause, and whether this is the right decision†: the bottleneck of the business.
  • Whether the budget and the people match what the mission says.†

Honest Belief (P2) · Watkins · Shewhart · Snowden and Boone · Lovallo and Kahneman

What options are there, and what does each one deliver?

  • The AI’s options, plus the missing families: try small, wait, combine.
  • The numbers for each option, including the ones nobody proposed.
  • The value of keeping doors open†: what an option makes possible later.
  • The minimum data for each type of decision; an effect without data does not compete†: the data is requested.
  • What weighs most for whoever decides.†

Admissibility-First (P1)

Can it be done, and what holds it back?

  • Four gates: feasible, safe, legal and legitimate. They only remove, nothing passes without data, and near the limit they tighten.
  • Execution: capability × will × coordination; the lowest factor rules.
  • Every actor needed multiplies the risk†.
  • The suppliers and systems the option needs.†
  • Not acting is judged too†, and the price of not doing what is forbidden is stated.

Admissibility-First (P1) · Four Gates · Honest Belief (P2) · Multiplicative Execution (P3)

Second front

The people: with whom, and how?

Productivity AI helps each person, or their team. DiaCroma looks at the decision for the whole company and at the people who have to make it possible.

A red umbrella open in the rain; not a single drop falls beneath it

Private by default. Each executive decides what they share about themselves; nobody scores anybody, and each person sees only their part of the plan.

With whom?†

  • The other executives: what it clashes with in what they have under way, and whom to talk to. What is admissible for the company is what everyone can accept.
  • Who approves it, and how to present it to them, according to what weighs on them: moving forward, stability, their people, how others will react, or doing the right thing.
  • Who holds it back, their coalitions, and the strategy with each one.

Human Constraints (P10) · the DiaCroma book (sponsors)

Who carries it out, and how?†

  • The shape of each step, according to what each executive shares about themselves: how they think, execute and relate under pressure, and what counts as winning for them.
  • The team gap: what the option demands that does not come easily to any of the people carrying it out, and the role that would cover it.
  • The executive themselves: their workload, how much depends on them and what they will have to stop.

Multiplicative Execution (P3) · Human Constraints (P10) · the DiaCroma book (Palette, Shape)

Third front

Time: how long, and how is it monitored?

What runs out first, what can go wrong, and the plan that is monitored to the end.

Bars of a series over time, in orange and yellow, over blurred figures

How long does it take, and what can go wrong?

  • The first budget to run out: time, cash, energy or political capital, and how much relieving it is worth.
  • The race: finishing before running out of money, people or support.†
  • The worst case alongside the average; sanctions and reputation.†
  • The probability of success, in %†, and how much it would rise by fixing what holds it back most.
  • Waiting has a cost too; and which assumption decides† the answer, with its range.

Temporal and Causal Credit (P7) · Decision Honesty (P8)

How is it done, and how is it monitored?

  • A staged path: do, in parallel, if…, deadline, checkpoint, escalate, roll back.
  • When to change the plan: if two checkpoints in a row miss the minimum; a single one can be noise.
  • Noise and seasonality: what looks like a problem and is not.†
  • When to stop: loops, futile attempts, improvement that flattens out.
  • Habits†: the minimum frequency that sustains a practice, and how long it takes to settle in.
  • What to find out first, and how much knowing it is worth.†

Typed Decision Output (P6) · Habit Thresholds (P4)

The answer

What goes back to the AI: the decision, in 360°.

Everything the executive needs to decide, in a single answer: what to choose, what it takes for it to work, what is still unknown, whether there is something better and how to carry it out.

  1. The choiceThe AI’s, if it passes the gates.
  2. The conditionsWhat has to hold for it to work, and what holds it back most.
  3. What to find outOnly what could change the decision, once, saying what it fixes and what waiting costs†.
  4. A better option outside the listIf there is one, and why.
  5. The staged pathWith its checkpoints.

Typed Decision Output (P6) · Learning Boundary (P9) · Four Gates · Human Constraints (P10) · Rumelt

Back and forth

It is not a questionnaire. Each question comes from your last answer.

In each round, DiaCroma works with what there is and asks only for what could change the decision at that moment. Your answer changes what matters, and with it the next question. And no round comes back empty: it chooses, answers conditionally or says what to find out first.

  1. Your AI consults itYour platform’s AI passes DiaCroma the decision, with the data from the company graph.
  2. DiaCroma answers the AI, not youIt does not write the message: it returns the content of the answer in a fixed format, with the choice, the conditions, what was struck out and why and, if a piece of data that decides is missing, the question. The exact format, field by field, is given to partners under NDA.
  3. The AI tells you, and asks youIn its own words. Before it reaches you, it is checked that it does not leave out what was struck out.
  4. You answer, and another roundThe AI consults again with what is new and DiaCroma works it all out again, with the same memory: it does not ask again what you have already answered†.

At LogiSul, the demo case

Inside

A deterministic engine. Same data, same answer.

What the executive sees are the questions. Behind them, every decision goes through the engine, always in this order.

  1. The company and the optionsThe company graph and the AI’s options. Every piece of data carries its source, whether it is measured, declared or inferred, its confidence and its date.
  2. An honest belief about the situationWith probabilities, updated with every new piece of data.

    Honest Belief (P2)

  3. Four gates

    Admissibility-First (P1) · Four Gates

  4. Will it really be carried out?

    Multiplicative Execution (P3) · Habit Thresholds (P4)

  5. Chooses only among what is admissibleA better score changes which one is chosen; it never makes admissible what is not.

    Patent specification

  6. An answer, not a letterAct, ask first or do not recommend†. It does not flip its recommendation from one day to the next because of noise.

    Typed Decision Output (P6) · Decision Honesty (P8)

  7. Time and learningIt learns from every result outside the core, in a governed adapter.

    Temporal and Causal Credit (P7) · Learning Boundary (P9) · Human Constraints (P10)

Every 90 days

The plan is monitored, and DiaCroma learns.

Checkpoints

The graph measures the indicators and the spending of each budget, and the AI consults DiaCroma again with what was measured.

Cycle close

What was not achieved carries over to the next cycle as a priority. DiaCroma learns the credit against what would have happened anyway, the reliability of each resource, its own calibration and the probability of success†. What third parties and teams do, also outside a query, goes into the learning.

Temporal and Causal Credit (P7) · Learning Boundary (P9) · Decision Honesty (P8)

See it in a case. And take it to your clients.

† Implementation in progress.