### Cutting the Cost of AI - A Guide to Token Efficiency

**Most organisations run AI or AI agents in production without a clear picture of what it actually costs. Every request consumes tokens both as input and output, and those tokens translate directly into spend that's often underestimated. Because hands-on experience with AI remains limited in many companies, these costs remain unclear and, particularly for agents, can add up to substantial sums before anyone notices.**

In this live session, we show how organisations can make their AI costs visible and bring them under control. Using a live comparison, we demonstrate just how much prompt design and model choice affect token usage and cost. You'll see what to look for when identifying cost drivers, and how this kind of tracking can be set up to run continuously, depending on your organisation’s setup.

Experience live demonstrations on tracking token consumption, measuring costs, and identifying effective ways to reduce expenses while maintaining high standards. Organisations frequently deploy AI in production without a true grasp of the actual costs involved. Every request consumes tokens; each model is priced differently, and with AI agents in play, expenses can escalate rapidly and go unnoticed. 

What if, in one hour, you could see live just how big these differences really are and which levers actually move the needle?

### Agenda

* 5 min — The problem. Why token costs stay invisible inside most organisations, and what changes once agents are in the loop.

* 30–40 min — The cost breakdown.

  * Results across different prompt styles and models, shown as a visual comparison

  * A live agent optimisation — same task, different cost, same output quality.

  * What "token maxing" is and why it's one of the largest controllable cost drivers we see.

* 10–15 min — Q&A and what's next. Your questions on the approach and the method

### Who it's for

* Executives accountable for AI spend who want predictability rather than surprises.

* Teams running or rolling out AI agents who suspect costs are higher than they should be, but lack visibility into why.

* Anyone who currently experiences AI costs as a black box.

### What you'll take away

* A concrete sense of how much prompt wording and model choice can move your costs — backed by real comparison data.

* A clear understanding of "token maxing" and how to spot it in your own organisation.

* A realistic view of where agent workflows put your budget most at risk.

* What a deeper, tailored analysis of your own usage and workflows could look like — same method, your data.

### Practical

- **Duration:** 1 hour
- **Format:** Virtual (Zoom)
- **Cost:** free
- **What to bring:** any questions and experience you have regarding cutting AI token cost


### Your Speaker

**Leona Schulz** – Hitabis GmbH

Leona Schulz is a full-stack software developer at Hitabis and specializes in the development of AI agents and the efficient use of AI.

Among other things, she has contributed to a GreenIT project and brings practical experience in how AI technologies can be integrated into software products in a meaningful and cost-effective way.

[Hitabis GmbH](https://www.hitabis.de), a Berlin-based IT services company, has been developing customized software solutions for businesses, government agencies, and organizations since 2004. Its range of services extends from the development and integration of technical systems and machinery, through database and web development, to the successful implementation of complex digitization and migration projects.

In this live session, we show how organisations can make their AI costs visible and bring them under control. 

Cutting the Cost of AI - A Guide to Token Efficiency

### Customizing the Beer Distribution Game — Live, with AI

**A 1-hour live session for everyone who has played the Beer Distribution Game. We open up the simulation behind the game and change it live, with AI doing the modelling — supply chain structure, product structure, demand patterns, lead times, costs, ordering policies, even the wording. Then we demonstrate our new multiplayer capability, hot off the press.**

You have played the Beer Distribution Game. You watched a small order swing at the retailer turn into a violent one at the factory, and you know by now that the bullwhip effect is not a story about careless players.

But the game you played is fixed. Four tiers, one product, one demand pattern, one set of lead times and costs. The interesting questions start where the fixed game stops:

- What happens with five tiers instead of four? With two? With a network rather than a line?
- What if the retailer could see the factory's backlog?
- What if lead times halved — or demand were seasonal rather than a single step?
- What if it weren't beer at all, but the products, the roles and the language of *your* supply chain?

In this session we open the model and answer those questions live, with you choosing what we try next.

### What the game actually runs on

The Beergame is a simulation model on **Metapad**, our AI-powered platform for modelling and simulation. The model is not locked away inside the game — it is an open template you can read, copy, run and change. That is what makes this session possible: we are not showing you configuration screens, we are showing you the model.

And the changing is done with AI. We describe the change we want in plain language, the AI assistant makes it in the model, and we re-run. That is the workflow you would use yourself.

> **Try it before you come.** The Beergame model is public in the Metapad template library — [open it here](https://app.metapad.ai/models/16cf74a7-e642-4277-b22a-8b5576bca422). You land straight in the app, where you can read the model, run the simulation and change things yourself. It is the same model we will be taking apart in the session. Nothing to prepare — but if you arrive with a question you ran into yourself, so much the better.

### Agenda

- **5 min — What's behind the game.** The model itself: the stocks and flows, the ordering policies, and the delay structure that produces the bullwhip.
- **25 min — Customizing it live, with AI.** We change the game and re-run it, with the audience calling the shots — and we talk through what is worth customizing in the first place:
  - **The wording** — roles, products, units. It does not have to be beer.
  - **The supply chain network structure** — more tiers, fewer tiers, branches, multiple suppliers.
  - **The product structure** — more than one product, with shared or separate capacity.
  - **Demand patterns** — a step, a ramp, seasonality, noise, a shock.
  - **Delays and lead times** — shipping, production, and how fast information travels.
  - **Costs and ordering policies** — what players are optimising, and what they are allowed to see.

  One change, one run, one look at what happened to the bullwhip — then the next.
- **5 min — Making it your own.** Metapad supports **custom branding**, so facilitators, trainers and consultants can run the game under their own brand rather than ours. And because Metapad is **multilingual by design**, the same game can be offered in several languages at once — the model is translated, not rebuilt.
- **15 min — Hot off the press: the multiplayer capability.** We are extending Metapad's simulation layer so that any simulation model can become a multiplayer game, with each player taking a role. You define the roles, decide who decides what and who sees what, and put people in the seats. We will demonstrate it fresh — you will be among the first to see it. The Beergame stops being a one-off product and becomes one instance of a general capability.
- **10 min — Q&A and what's next.** Your questions on the model, the platform and the method — and, if you want a game built around your own supply chain or operating model, where that leads.

### Who it's for

- **Educators and trainers** who teach with the Beergame and want variants — a different structure, a different demand pattern, a debrief that shows *why* rather than just *what*.
- **Supply chain professionals** who have run a multigame with their own team and wished the game looked more like their own network.
- **Facilitators and consultants** who run the game regularly, want to build their own presets rather than pick from ours, and would rather run it under their own brand.
- **Modelling practitioners** — simulation engineers, operations research analysts, business engineers — who want to see the model underneath and how it is edited.

### What you'll take away

- **A sharper picture of what actually drives the bullwhip** — because you will see it change as we change the model, rather than hear it asserted.
- **A working knowledge of how the Beergame is built**, and how it is changed with AI. The model template is open, so you can go and do it yourself afterwards.
- **A concrete sense of what a branded, multilingual variant of the game would look like** for your own teaching, training or client work.
- **A first sight of the new multiplayer capability** and what it opens up beyond the Beergame.
- **A route** if you want a simulation or a game built around your own supply chain, process or operating model.

### Practical

- **Duration:** 1 hour
- **Format:** Virtual (Zoom — link sent on registration)
- **Cost:** Free
- **Dates:**
  - **English** — Thursday, 12 November 2026, 16:00 CET
  - **German** — Thursday, 19 November 2026, 16:00 CET
- **What to bring:** the change you have always wanted to make to the game — or one you tried yourself in the model beforehand. We will take suggestions from the audience and run them live.

## Your Host

**Dr. Oliver Grasl** — Founder & Managing Partner at transentis

Oliver has spent over 25 years in enterprise transformation, bringing business engineering, operations research and analytics to bear on how enterprises actually run. He is the architect of the transentis Beer Distribution Game and of Metapad.

His passion is building models that bring human intelligence and machine intelligence together — computational knowledge graphs that turn complexity into clarity and make change steerable. A simulation you can play is one of the most direct ways to feel that: you do not read about the dynamics, you run into them.

Live session: we open up the simulation behind the Beergame and change it live, with AI doing the modelling — then demonstrate our new multiplayer capability.

Customizing the Beer Distribution Game — Live, with AI

### Re-Think Your Operating Model — Live, with AI

**A 1-hour interactive session for Operating Model Owners, Transformation Owners, and modelling practitioners. We start with an empty canvas and build a knowledge graph of an operating model from scratch — live, with you, with AI in the loop.**

Most enterprises run on operating models that live in tacit knowledge, scattered documents, and the heads of senior people.

What if you could turn that mess into a queryable knowledge graph — in real time, with AI doing the heavy lifting?

In this 1-hour live session you'll watch (and join in on) the build itself. We put a fictitious company on the table, share the kind of artefacts most enterprises actually have — a leadership memo, a process write-up, a KPI snapshot, a strategy slide or two — and build a knowledge graph live with AI. 

No prepared demo, no polished result. Just the workflow, as it happens.

We:
  
  - Describe the situation to the AI assistant.
  - Share the documents and watch the assistant parse them.
  - Build the knowledge graph node by node — capabilities, processes, decision rights, KPIs, dependencies — with the audience suggesting what to ask next.
  - Surface tensions, gaps, and the decisions that fall out.


### Agenda

- **5 min — The problem.** Why operating-model conversations stall on slide decks and out-of-date documents — and what changes when you can build the model collaboratively, with AI, in the room.
- **30–40 min — The live build.** A fictitious company (we'll share the brief on the day). A handful of typical artefacts. An empty metapad model.
- **10–15 min — Q&A and what's next.** Your questions on the build, the platform, the method; a walk-through of our **AI-Driven Operating Model Workshop** (our full-day version on *your* operating model, with your real documents)


### Who it's for

- **Operating Model Owners** — CEOs, COOs, CTOs and other senior leaders accountable for how the enterprise actually runs, who want their operating model to be a queryable, improvable artefact rather than tacit knowledge spread across people and slide decks.
- **Transformation Owners** — accountable for a specific multi-stakeholder enterprise transformation, looking for ways to keep the model of the work explicit and useful as the programme evolves.
- **Modelling practitioners** — enterprise architects, business engineers, knowledge engineers and operations researchers who already work with structural and process models and want to see what an AI-grounded, computational version of that practice looks like.
- **Anyone curious** about how AI assistants behave differently when grounded in a typed knowledge graph of an enterprise rather than scraping a pile of documents.

### What you'll take away

- A first-hand view of the *workflow*: how a knowledge graph gets built from prose-and-doc inputs, live, with AI.
- A clear picture of the difference between AI-on-documents (RAG) and AI-as-co-modeller (a *computational knowledge graph* being built in front of you).
- A concrete next step if you want to do this on your own operating model — same workflow, your data, more time.

### Practical

- **Duration:** 1 hour
- **Format:** Virtual (Zoom)
- **Cost:** free
- **What to bring:** questions you'd ask your own operating model if it could answer; a willingness to suggest what to ask next during the live build


## Your Host

**Dr. Oliver Grasl** - Founder & Managing Partner at transentis

With over 25 years in enterprise transformation, Oliver brings deep expertise in business engineering, operations research, and analytics to reshaping enterprises operating models.

Oliver's passion is building digital twins that bring human intelligence and machine intelligence together. 

He believes the best decisions emerge from this collaboration — and builds "computational knowledge graphs" that make this possible, turning complexity into clarity and making change steerable.

Live demo: See how we build a live, queryable knowledge graph of an enterprises operating model

Re-Think Your Operating Model with AI

Prototyping Business Models And Market Strategies

Make Your Professional Service Firm Grow

Business Prototyping Manifesto Hero Image

How can you unleash the power of business prototyping? Check this post!

How to Unleash the Power of Business Prototyping

How can you unleash the power of business prototyping?

How to Unleash The Power of Business Prototyping

How the path from a customer's order to the payment in the bank account is automated step by step, plus why the limit here is set by the customer, who can ask for the reason behind every decision.

How AI Changes the Order-to-Cash Value Stream

How accounts payable moves from typing every invoice by hand to posting the standard ones automatically, plus what has to stay under human control once a process ends in a payment.

How AI Changes the Procure-to-Pay Value Stream

How customer service shifts from an agent handling every ticket to AI handling the routine ones with a human on call, plus where that quietly goes wrong.

Services

Products

Training

Resources

Company

How to Unleash The Power of Business Prototyping

How can you unleash the power of business prototyping?

Where to Start

transentis consulting

Geisbergstraße 9

10777 Berlin