Architecture

Why Orisdale Starts With the Business Decision, Not the Chatbot

Orisdale Editorial Team · Published · 7 min read

Orisdale starts by naming the business decision, the grain, and the person who accepts the result. A chat window is the surface. It is not the decision.

The work being scoped is a decision

A strategy conversation at Orisdale is a scoping exercise. The useful output is a decision brief: the question the business already asks, the grain of the answer, the rule that is allowed to produce it, and the person who still has to accept a consequential result.

The platforms the organization already trusts keep the data, the workflow, and the access model. Orisdale's contribution in the brief is decision design, governed business logic, semantic context, and the review experience that sits on top of those platforms: saying which decision is in scope, which semantic context an agent may use, and where a person remains accountable. A project that begins with a window, a model name, and a list of systems to connect has not started that work. The reference architecture is the stack. This note is the conversation that should happen before anyone argues about where the stack is hosted.

A chat window is not the decision

The window is a place to ask. The decision is something the business will act on, or put in front of someone who acts. A window can accept any sentence. A decision has a stop rule: what a good answer looks like, what it must not invent, and what happens when an input is missing. Without those, the window still produces paragraphs, and the paragraphs are not a governed result. "An agent for everything" has no grain, no owner, and no version of the rule. Variance review, exception prioritization, and a stated contract question each begin with a decision someone already owns.

IllustrationA decision brief feeding a bounded review experience, not a chat transcript.
Business questionOne sentence the owner already recognizes, in their language.
Approved inputs and ruleThe fields and the rule the agent is allowed to read. Nothing invented.
Reviewable outputThe classification or list the rule produces.
Bounded review experienceAn accountable person confirms the output before it is treated as a finding. The agent may explain the fields. It does not accept the result on anyone’s behalf.

What a decision brief records

The brief is short on purpose. An executive should be able to say whether it is the decision the organization actually makes. A technical reader should be able to see what must be true before an agent is allowed to speak.

Decision. One sentence in the owner's language. "Which departments require management attention this period" is a decision. "Use AI on finance" is a theme.

Accountable owner. The function that already owns the rule. Finance owns a variance classification. A land or contracts team owns which contract terms govern a drilling-period exception; a drilling-performance analyst reviews the periods that rule selects, not the rule itself. The agent does not become the owner by answering.

Grain and period. The unit and the time window. A department for one close, a well or rig for a declared contract period, a lease as of a stated date. Undeclared grains get compared as if they were the same number.

Approved inputs. The categories the rule may read, and the system of record. The brief does not paste a company's rows into a scoping document.

Ruled output. The list, flag, or classification the approved rule produces. The agent may explain that output. It may not replace it with a definition invented in the answer.

Missing data. An empty amount, a missing time code, or an absent rate stays missing. The brief says so, so a fluent paragraph cannot fill the gap.

Human acceptance. Who confirms a consequential result before it is treated as a finding, a contract claim, or a management action.

Out of scope. A variance review does not become a forecast redesign. A drilling exception review does not become autonomous field control.

If any line is still "the model will figure it out," the brief is not ready. The governed business logic note explains why a prompt cannot stand in for the missing rule.

Which requests fit

Fit is not a judgment about the model. It is a judgment about whether the decision is bounded enough to govern.

RequestFitWhat has to be true first
Bounded variance reviewSuitableA named grain, an approved variance rule, and a person who accepts the classification before it is treated as the management view
Exception prioritizationSuitableAn exception rule the business already owns, a declared period, and a review list rather than an automatic action
Contract review of a stated questionSuitable when the question is boundedThe question names the period, the fields a reviewer will read, and the decision the review is meant to support
An agent for everythingPoor fitNo single decision, grain, owner, or stop rule. The window would be choosing the question while answering it

Moving a request off the last row is the scoping work. It is a choice to leave most of the enterprise outside this brief, not a smaller prompt. Bounded variance review is the pattern in the synthetic finance worked example. Exception prioritization is the Energy blueprint: a review list, governing fields attached, a person confirming the exception. A contract review fits a named period and named fields, and it stops fitting when the request becomes "read every contract and tell us what matters."

Functions and industries are different scopes

Orisdale publishes five business functions: revenue, finance, tax, operations, and executive intelligence. An industry is where a decision is situated, and the same decision can draw on more than one function at once.

Energy is the clearest public example. A drilling exception can be an operations review and a finance review at the same time, because a contract rate and a cost category sit on an operational time record. A land obligation can carry both an operational commitment and a financial interest. An operational-health queue can be read in a maintenance meeting and again in an executive review. Those links are already drawn on the Energy page. They do not create an "Energy Intelligence" product line beside the five functions, and they do not imply a customer in the sector.

The brief names the function that owns the rule, and names the industry only when the decision is specific to that sector. A generic operations exception and a drilling-period exception are not the same brief.

A proposed Energy exception review

Label: proposed blueprint. Synthetic. Not a customer engagement. Not a demonstrated runtime.

Consider a drilling-performance analyst preparing a contract conversation. The decision in the brief is which wells, rigs, or contract periods a person should look at before that conversation. The user is the analyst, not an autonomous field system. The grain is a well or rig, by day or by contract period, and the period has to be declared before two periods are compared.

The input categories are the pattern names already on the Energy page: well operations, drilling time breakdowns, contract terms, and cost categories. They are not a named operator's records. The ruled output is a review list of periods an exception rule the business has defined would select, with the governing fields attached. This note does not publish a threshold, a rate, a savings figure, or an accuracy rate. The blueprint does not have those numbers.

Missing data is part of the scope, not a footnote. A missing time code, rate, or period stays missing and is labeled. The model does not impute it so the list looks complete. Human acceptance is equally specific: a person confirms the exception before it is treated as an operational finding or a contract claim.

Out of scope: production optimization, predictive maintenance, autonomous control, and any drafted explanation treated as a field instruction. The three-pattern boundary is in AI for Energy Operations. When a business-facing agent is in scope, the reference pattern is Governed Business Agents on Trusted Platforms. That pattern is a design, not a deployment.

A readiness checklist

Use this before a build is scheduled. A no means keep scoping.

  1. The decision is one sentence a business owner recognizes without a product diagram.
  2. The grain and the period are declared, and two different grains are not treated as one number.
  3. The rule has an owner, or the work is explicitly a blueprint and publishes no numeric threshold.
  4. The agent is limited to explaining governed fields. It is not the author of the formula, the grain, or the cause.
  5. Missing inputs stay missing in both the calculation and the narrative.
  6. A person is named for consequential acceptance, and that acceptance is separate from changing the rule.
  7. The request is one of the suitable rows above, not an agent for everything.
  8. The next step is a reviewable output. The brief does not promise a return, a saving, or a production date.

Where to start

Bring one decision. The contact page is the place to request a strategy session. The session scopes the brief. It does not promise a result, a threshold, or a date when a system will be live.

Read the solutions for the five functions, the Energy blueprint if the decision is one of those three patterns, and the architecture for the stack the brief has to respect. The platform-pattern article is the right next read when the question is how a business-facing agent sits on preparation, governed outputs, and human review without becoming the calculation.

Discuss how this applies to your environment