Ask Theralon
The fastest way to get an answer is to ask a direct question.
Not a general-purpose chatbot layered on top of the platform. Every answer is resolved against your own live model, current anomalies, risk register and operational history — and every factual claim has to carry a real figure before it's shown. Where the evidence isn't there, Theralon says so rather than filling the gap with something plausible-sounding.
Beyond retrieving what's already computed, it can:
- Define a cohort on the fly, by whatever condition is asked for
- Calculate expected loss from calibrated probability against real exposure
- Rank the likely explanations behind a pattern, not just describe it
- Surface a relationship between two variables — with an explicit reminder that correlation isn't cause
- Hold context across a conversation, the way a colleague would
Every answer is computed live from your own model, so the same question asked a week apart returns a genuinely different, specific answer — never a templated one. The effect is to collapse the distance between having intelligence and using it: instead of interpreting a chart or exporting a report, an operator just asks.
"Why did on-time delivery drop in the Midwest region this week?"
- Three carriers below 90% OTIF, concentrated on one corridor
- A 12% rise in dwell time at two connected cross-docks
- Order volume is within its normal range — not a demand issue
What a question can turn into
The operator doesn't need to know which part of the platform to ask.
A question is the entry point to the reasoning loop. Theralon works out which stage the question is really about, composes the reasoning it needs, and answers from the same live model everything else is built on.
"What's getting worse?"
Trajectory
"What does that affect?"
Propagation
"How much are we exposed to?"
Exposure
"What should we do?"
Intervention
"Have we dealt with this before?"
Operational memory
"How do you know?"
Evidence and provenance
An investigation, not a lookup.
"Investigate why European delivery performance has deteriorated" isn't a search for matching records. It becomes a structured investigation across trajectories, dependencies, exposure, historical events and outcomes — comparing entities and periods, tracing dependencies across multiple hops, weighing competing explanations.
The conversation is part of the reasoning.
"Which suppliers are deteriorating?", then "which facilities depend on them?", then "which one creates the greatest exposure?", then "what happens if it fails?" — each question extends the same investigation, carrying the entities, findings and time periods forward, rather than starting again from an empty context.
Ask what's limiting the answer.
"How do you know?" · "Which sources disagree?" · "What's limiting your confidence?" · "Show me the evidence." Every material finding traces back through the reasoning chain to the measurements, source observations, model versions and confidence components that produced it. Where the evidence is insufficient, Theralon states what's known, what's uncertain, and what's missing.
Move from what is to what if.
"What happens if this supplier fails for seven days?" · "What if demand increases by 15%?" · "What if we move 30% of this inventory to another facility?" The resulting propagation, constraints and exposure are evaluated against the current model — so a question can move from what's true now to what's likely next without leaving the same system.
Then into a decision.
Asked what to do, Theralon identifies and evaluates candidate interventions against expected effectiveness, cost, downstream consequences, feasibility, reversibility, historical effectiveness and confidence. It doesn't jump from anomaly to action — it reasons through the consequences first, grounded in the same evidence as the finding underneath.
One reality, phrased for the audience.
A board summary, a finance leader's question about what's driving exposure, an operations leader moving straight into the underlying entities — different levels of detail, but never different versions of reality. Reports are generated on demand from the same live intelligence, scoped to what the recipient is authorised to see, with the supporting chain still attached.
As autonomy develops, the same interface carries an authorised decision through simulation, approval and execution — and the outcome returns to operational memory, keeping the whole chain from observation to result intact. The conversation doesn't end at the recommendation. It closes the loop.