The NAVI AI Enablement Journey · Guide 09A

Efficient AI Use: Token Usage and Management

Use enough AI to create meaningful value—and eliminate usage that does not contribute to learning or accepted work.

Model choiceTask preparationEfficient contextStopping discipline
Core idea

The highest-impact token decisions happen before the work begins: choose the right model, define the result and provide only the context required.

What a token is
and how it is used

A token is a small unit used to measure model processing. In practice, usage grows with what NAVI reads, the reasoning it performs and what it generates.

ReadYour instructions, conversation history, files, retrieved knowledge and tool results.
ReasonThe work required to analyze information, resolve complexity and complete multiple steps.
WriteThe answer, analysis, code or other material NAVI generates.
Read, Reason and Write are practical sources of usage—not formal billing categories. NAVI accounting may distinguish input, cached input, output and reasoning usage.
Choose the lightest
model that fits

Start with the fastest option likely to produce an acceptable result. Move up when complexity, ambiguity or risk requires more reasoning.

Fast

Use for straightforward work where speed matters most.

Typical per-run costLowest—usually considerably less expensive than Methodical.Good fitSimple questions, quick checks, short summaries and direct rewrites.

Balanced

Use as the default for most business work.

Typical per-run costModerate—a practical balance of capability, speed and cost.Good fitSource-backed drafts, moderate analysis, planning and everyday problem-solving.

Methodical

Use intentionally when the cost of a weak answer is high.

Typical per-run costHighest—deeper reasoning and additional work can cost substantially more than Fast.Good fitComplex, high-stakes, ambiguous or multi-source work that requires careful reasoning.
Choose the lightest model likely to succeed. A cheaper run is not a cheaper task if it creates retries or major corrections. Exact differences depend on the current NAVI model configuration.
Not all usage
creates equal value

Token volume alone does not show whether AI is being used well. Protect learning, improve production and eliminate spin.

Tokens that teach

Intentional investment that develops useful capability.

  • Testing representative tasks and approaches.
  • Building approved context, memory and knowledge.
  • Creating reusable instructions and workflows.

Tokens that produce

Usage that contributes to work the organization accepts and uses.

  • Completed research, analysis or deliverables.
  • Reliable workflow and agent results.
  • Business work that improves time, quality, capacity or decisions.

Tokens that spin

Usage that adds cost without sufficient learning or usable output.

  • Idle or unread automations.
  • Overlong sessions and unnecessary data pulls.
  • Repeated corrections caused by unclear scope, missing context or the wrong method.
Order of action: eliminate spin first, improve the reliability of production second and preserve a reasonable learning budget.
Practical habits
for every user

Good token management is mostly good work definition, source selection and stopping discipline.

Before the run

Start a new session when the objective changesKeep related follow-up work together. Start clean when the objective or source set materially changes.
Define the result and acceptance testState what must be delivered, for whom and what makes it usable.
Define the smallest sufficient source scopeName known sources, dates, systems or boundaries. NAVI can locate other relevant approved information within that scope.
Set boundaries for judgment and actionState what NAVI may infer or change, what requires confirmation and what to do when information is missing.
Set a checkpoint for large workAsk NAVI to confirm the approach before broad research, large data pulls or final production.

During the run

Confirm direction before expansionCheck the initial approach before NAVI continues into expensive or time-consuming work.
Stop weak work earlyCorrect the instruction, source, scope or model instead of repeating the same request.
Keep retrieval preciseAvoid entire inboxes, databases or document collections when a useful subset is known.

After the run

Accept, correct or stopDo not continue polishing work that does not meet the stated need.
Make repeated work reusablePreserve a proven instruction, workflow, source pattern or acceptance rule.
Report persistent wasteFlag unused automations, repeated failures and unnecessary complexity.
Too broadReview everything and tell me what we should do.
Better scopedUsing the Q2 pipeline report and sales-call summaries, identify the three largest conversion bottlenecks, cite supporting evidence and recommend one action for each. Ask before expanding to other sources.
Control recurring
work

Automation repeats both value and waste. Prove the work manually before allowing it to run unattended.

Prove it manuallyAccept representative results before scheduling the work.
Use the lowest useful frequencyRun only as often as a decision or business process requires.
Assign review and stop conditionsName the owner, review point and conditions that pause the work.
Remove unread outputPause automations whose results are not reviewed, used or acted upon.
How NAVI helps

NAVI also reduces avoidable usage behind the scenes, while users remain responsible for clear task definition and sound judgment.

01Scoped knowledge retrievalRetrieves relevant knowledge rather than inserting entire document collections into every request.
02Context efficiencyUses caching, pruning and compaction practices to reduce repeated or low-value context.
03Tool and data guardrailsShapes and limits large pulls to reduce avoidable processing and worst-case spikes.
NAVI handles some context optimization automatically. Users have the greatest impact by choosing the right model, defining the work clearly and limiting unnecessary context and rework.
Decision rule:Choose deliberately. Define clearly. Limit context. Stop weak work early. Reuse what works.
Before client publication, GrowthIQ must confirm current model-mode descriptions, NAVI token accounting, included allowances, overage terms and deployment-specific behavior. This guide is an operating aid, not a billing agreement or substitute for client-specific financial and governance decisions.