# Token management & optimisation: The right AI for every step. Lower costs per result.

Official page: https://aitje.com/en/services/token-management-en-optimalisatie

> AITJE Maps out AI usage and costs for content, knowledge, images and code. Optimise subscriptions, models and workflows. Available as a standalone service or as a follow-up to the AI scan.

Where are your AI budget and tokens going? ‘AITJE’ identifies which teams are using AI, for which tasks, and via which subscriptions, APIs and in-house models. From writing text and searching for knowledge to generating images, building code and other applications.

AITJE then investigates which setup is most suitable: multiple smaller subscriptions, a shared model budget, groups of specialised agents, or a combination of external, local and edge models. The choice is based on the workload, quality and total cost per usable result.

The workflow itself is also important. Dividing tasks, providing specific information and carrying out fixed actions using code can reduce unnecessary token usage. This service can be purchased separately; the [AI-scan](https://aitje.com/en/services/ai-scan) provides a broad overview of your usage and opportunities for optimisation.

## Who it is for

- Teams with individual AI subscriptions and little overview
- Organisations with high or fluctuating API costs
- Those who use AI for content, knowledge, images or code
- Companies with agents and automated workflows
- Anyone wishing to combine external, local and edge models
- IT companies that manage the use of AI for clients

## The approach

### Usage guide

Which teams use AI, for which tasks and at what cost? ‘AITJE’ identifies subscriptions, models, tokens and other usage, such as image credits. Where available, quality, turnaround time and repetitive work are also measured.

### Choosing the setup

Based on your requirements, AITJE compares subscriptions, APIs and suitable external, local and edge models. Shared budgets and groups of specialist agents may also form part of the approach.

### Tailor as appropriate

Within the agreed scope, tasks are allocated, prompts are refined and data is retrieved in a targeted manner. Code handles the routine work; the chosen model is provided with the correct context. Where useful, limits, the reuse of results and central model routing are incorporated.

### Checking the result

AITJE It compares the costs, quality and speed before and after the adjustments, including any additional attempts. Documentation, logging and agreed management help to monitor usage going forward.

## Deliverables

- Overview of AI usage by team and task, where measurable
- Insight into subscriptions, tokens, API costs and other model usage
- A well-founded proposal for models, budgets and workflows
- The agreed changes to model calls, context and task allocation
- A comparison of costs and quality before and after implementation
- Logging and usage limits as agreed
- Documentation and agreements regarding data, access and, where applicable, management

## Prices and conditions

Fixed hourly rate. All prices exclude VAT.

An hourly estimate is provided in advance for assessment, comparison, implementation and, where applicable, management. You can also opt to have just an analysis carried out. We reuse findings from the AI scan. Subscriptions, model usage, hardware and hosting are quoted separately.

## Agreed separately

- External API costs, hardware, hosting or software licences
- A fixed saving or the same outcome for every workflow
- Ongoing management outside the agreed scope of the assignment

## Frequently asked questions

### What is token management and optimisation?

AITJE It maps out how your organisation uses AI and what that costs. It then examines subscriptions, model selection, agents and workflows, and makes adjustments where agreed. The aim is to reduce unnecessary usage whilst maintaining the required quality. Read more about [Token management & optimisation](https://aitje.com/en/services/token-management-en-optimalisatie).

### Is this part of the AI-scan or a separate service?

Both. The [AI-scan](https://aitje.com/en/services/ai-scan) provides a broad overview of your current AI usage, subscriptions and opportunities for optimisation. A comprehensive model comparison, workflow adjustments and ongoing management are covered under a separately agreed engagement. You can also purchase this service directly, without first carrying out a scan.

### Is it just about language models and APIs?

No. ‘AITJE’ also considers AI usage via subscriptions and applications for content, knowledge, images, code and other tasks. Not every service is billed in tokens: for images, audio or video, for example, credits, the number of images or minutes may be relevant. This usage data is factored into the cost per usable result.

### Could a different combination of subscriptions be more cost-effective?

That’s possible. Multiple smaller subscriptions, a shared model budget or API usage may be a better fit than the current setup. AITJE compares actual usage, included capacity, limits, quality and terms and conditions. Which combination is right depends on your team and tasks.

### What does AITJE mean by ‘agent clusters’?

Groups of agents, each carrying out a specific part of the work, such as retrieving information, writing text and checking the result. Each step can be assigned a suitable model and specific context. AITJE assesses whether this division of labour adds value to your workflow; additional agents may also mean extra calls and management overhead.

### Does this work with frontier, local and edge models?

Yes. AITJE can combine advanced models via external APIs with models on its own servers or local devices. Edge models run on, or close to, the device where the task takes place. The task, capacity and data flows determine the most suitable approach. With in-house models, hardware, power, hosting and management remain part of the costs.

### Does AITJE always choose the cheapest model?

No. ‘AITJE’ tests your tasks and compares quality, speed, usage and the number of attempts. A model with a lower token price may require more calls; a more expensive model may actually work out cheaper for a particular step.

### Can a workflow be set up to send results from €4 to €0.40?

This is a sample calculation, not a firm commitment. Assuming a hypothetical flat rate of €4 per million tokens, one million tokens would cost €4 and one hundred thousand tokens would cost €0.40. The achievable savings depend on the workflow, output, quality and retries. Costs for code, hosting, management and development are separate from this.

### What does the code do instead of the LLM?

For example, retrieving data via an API, selecting relevant fields, performing calculations and checking the output against fixed rules. The model is given a clearly defined task with specific context, rather than all the raw data and steps in a single large prompt.

### Can AITJE also build new workflows and manage API calls?

Yes. AITJE can build a central layer for model calls and a new workflow, including routing, logging, budgets and limits. Access, key management and a form of ongoing management are also agreed in advance.

### What is the difference between this and optimisation?

This service focuses on AI usage and costs within your organisation: subscriptions, models, agents and workflows. [Optimisation](https://aitje.com/en/services/optimalisatie) assesses an existing AI solution more broadly, for example in terms of reliability, stability and human oversight.

### How does AITJE determine whether a cost saving comes at the expense of quality?

By testing the end result. ‘AITJE’ compares the original and the adapted workflow on the same tasks and assesses whether the outcome still meets the agreed requirements. Fewer tokens or a cheaper model only constitutes an improvement if the result remains good enough.

### Can we see how much AI usage costs for each workflow or department?

Yes, provided that usage is measurable and can be attributed. AITJE monitors subscriptions, model calls, token usage and other consumption per task or team. Where necessary, logging can be set up to gain that insight.

### What happens if a model API goes down or reaches its limit?

AITJE A fallback can be built: an alternative route or a different model that takes over the task. For critical steps, a local or edge model may be a good choice. Which fallback is possible depends on the task and is agreed upon during the setup.

### Can ‘AITJE’ prevent a stuck workflow from making endless API calls?

AITJE It can incorporate limits on attempts and costs, logging and clear stopping points. At key moments, an employee may first need to assess whether the workflow can continue: ‘human in the loop’. In this way, technical limits are combined with human oversight.

## Contact

[Contact AITJE](https://aitje.com/en/contact) — contact@aitje.com