AI Analyst · Isaac

An analyst that only says what the run says.

Isaac is the assistant inside every scenario. Ask in plain language: it reads the answer from your runs and your model, shows the steps it took, and tells you when the data cannot answer the question.

Why the answer holds

Grounded in your runs, checked in code.

  • Grounded by enforcement, not by hope

    Every number Isaac states is read from the run through tenant-scoped primitives. If it cannot find the answer there, it tells you.

  • Reasons in steps

    “Why does the alternative cost more?” is a multi-step question — find both runs, diff them, drill the delta, test whether a capacity explains it. Isaac composes those steps itself.

  • Your data stays where you put it

    Isaac's language model is self-hosted, running alongside Beanome rather than at an AI vendor, so your data stays in your deployment. A customer who prefers another model can point Isaac at it, and a rules tier still answers the structured questions with no model at all.

What you can ask

Ask it the way you would ask an analyst.

Isaac answers from read-only tools over your model and its runs. These are questions it is built for.

  • Explain a result
    • Why is this customer's demand not fully met?
    • Why does Madrid source from Lisbon?
    • Why is this site idle?
  • Compare runs
    • Why does this scenario cost more than the baseline?
    • What changed between runs 12 and 14?
    • How do fill rate and cost compare across my runs?
  • Find the pressure
    • Which resources are at their limit?
    • Which customers are most underserved?
    • What are the biggest cost drivers?
  • Understand the model
    • What does a blank Max Capacity mean?
    • What depends on this supplier?
    • Why did this solve fail?
How it answers

One question, traced.

Follow a single question through Isaac: every read it makes, every check its answer passes, and, in the margin, where a different question would have gone instead.

01Ask
YouWhich customers lose service if we close Venlo?
open: scenario Bsite Venlo
02Frame
Isaacdeciding what this is aboutYou named a site and have scenario B open, so this is about Venlo, in scenario B.
Can’t tell which scenarioIt asks you back, with the choices, before reading anything.
03Planthink → read
up to 6 times
within 45 s
Isaacstep 1First, find everything that depends on Venlo.
The model is unavailableA rules-based floor still answers the common questions.
04Read
Isaac readswhat depends on site Venlo · scenario B
context graphchecked: your permissions
Your data3 lanes · 14 customers served only through Venlo
Not in this dataIt declines and says so, instead of filling the gap.
05Read again
Isaac readsweekly demand of those 14 customers · scenario B
Your data41,200 units a week · 9% of the network
Not something Isaac doesIt says so, and tells you where in the product that lives.
06Check
Isaacevery figure must have been read
  • 14 customers are served only through Venlo.
  • They order 41,200 units a week, 9% of the network.
  • Re-routing them would cost about €380k a year.never read · dropped
Nothing survives the checkYou get a plain summary of what was read, never a guess.
07Answer
You see14 customers are served only through Venlo. They order 41,200 units a week, 9% of the network. answered for scenario B · about site Venlo looked at sites, demand · 2 figures · 1 claim withheld
Pin to dashboardAsk for the project
A change is suggestedIt arrives as a proposal. Nothing is saved until you apply it.
Illustrative example: the site, customers and figures are made up. The steps and checks are the ones Isaac runs on every question.
  • Asks before it guesses

    When a question could mean two things, Isaac asks which one you meant.

  • Reads through a checked door

    The model never touches your data. It asks for one of a fixed set of reads, and each one runs with your permissions, inside your workspace.

  • Says only what it read

    A number appears in an answer only if it was read, or follows directly from figures that were. Anything else is dropped, and the drop is recorded.

  • Changes nothing itself

    Suggestions come as proposals you apply. What you want to keep, you pin.

Privacy

Your data stays in your deployment.

  • Self-hosted, always

    Isaac's language model runs alongside Beanome, not at an AI vendor, so your model, your runs and your questions stay in your deployment. If you would rather use another model, you can point Isaac at it; that choice, and who then sees the prompts, is yours.

  • Answers without a language model

    With no language model configured, a rules tier still answers the structured questions it was built for.

  • Conversations stay private

    A conversation is visible only to the person who had it. Administrators cannot read it, and you can clear it.

Pilot programme

Ask Isaac about your network.

Bring a workbook and a question. We build the model with you, and Isaac answers from your own runs.

Request a pilot