Skip to main content

Token Optimisation:

Reduce AI costs by optimising enterprise knowledge

The fastest path to lower token consumption is rarely a better model. It is better knowledge.

 

Deloitte Digital brings together expertise in AI, knowledge management, information architecture, content strategy, governance and value realisation to help organisations reduce token demand, improve AI effectiveness and increase return on AI investments.

Let's discuss how you can optimise your AI usage today

AI ECONOMICS DEPENDS ON MORE THAN THE MODEL

Understand use-case economics

Assess ROI once token consumption is considered

Reduce knowledge waste
Identify where fragmented knowledge drives avoidable usage

Optimise continuously
Track tokenomics as adoption scales

AI investment is accelerating, but scaling value remains difficult.

Many organisations can prove AI works. Fewer can confidently scale successful use cases while maintaining cost efficiency, quality and return on investment.

01

Unclear AI economics

Long-term operating costs are difficult to predict.

02

Pilot purgatory

Successful proofs of concept fail to become enterprise capabilities.

03

Uncertain ROI

Business cases rarely account for the impact of knowledge on cost and performance.

04

Scaling complexity

Adoption increases operational, governance and cost pressures.

OUR POINT OF VIEW

Token waste is often created upstream in enterprise knowledge.

The way knowledge is stored, structured, governed and retrieved affects how much context AI systems process and how effectively that context supports an answer.

Content duplication

Repeated processing of similar context

Multiple versions of similar information increase retrieval volume and make AI systems process more tokens than necessary.

Knowledge sprawl

Wider search and higher token load

Information spread across repositories forces AI systems to search wider, retrieve more and consume larger context windows.

Poor information architecture

Lower retrieval precision

Weak metadata, inconsistent taxonomy and limited semantic structure reduce precision and increase unnecessary retrieval.

Weak governance

More noise in AI responses

Outdated or low-quality content remains available to AI systems, increasing processing effort and reducing answer confidence.

Oversized context windows

Greater cost for limited value

When retrieval is imprecise, models receive more context than required, increasing token consumption without improving outcomes.

Low intelligence readiness

AI works harder to understand knowledge

Knowledge designed only for human navigation is often not structured for AI systems, copilots and agents to consume efficiently.

HOW WE ASSESS KNOWLEDGE READINESS

 

Assess knowledge readiness before scaling AI and optimise existing initiatives that already consume tokens.


We help clients improve the economics of existing AI initiatives and assess readiness before major investments, providing a clearer view of effort, feasibility, scalability and expected ROI.

 

An end-to-end optimisation of existing initiatives would usually involve:

 

1. Establishing a baseline assessment: We measure your token consumption and establish a current-state benchmark.

 

2. Finding root causes: We identify the knowledge, content and retrieval issues driving demand.

 

3. Prioritising root causes: We focus on the changes with the greatest potential impact. 

 

4. Optimising & measuring outcomes: We implement improvements, test outcomes and track results over time.