Glossary

Glossary: terms of licences, AI costs, invoices and sourcing

Updated Markdown
In short

This glossary defines the terms used on this site and in Cureman, in five groups. Licences covers seats, plans, renewals and playbooks. AI costs covers tokens, caching, the gateway, routing and budgets. Invoices and spend covers e-invoice formats, line items, price agreements and spend analysis. Sourcing covers requests for quotation, offer comparison, negotiation and award. The last group explains how the platform works, from findings and decisions to agent tasks and the audit log. Each entry starts with a definition that stands on its own.

This glossary defines the terms used on this site and in Cureman. The entries are grouped by subject, and each one starts with a definition that stands on its own. Where an article or a module page explains a term in depth, the entry links to it.

Licences

Seat

A seat is one paid right for one person to use a software product. Vendors bill by the number of seats on a subscription, whether or not each holder uses the product. The seat list of a vendor says who holds which seat and on which plan.

Plan (tier)

A plan, or tier, is one of the priced editions a vendor sells of the same product, each with its own features and its own price per seat. One subscription can mix plans, so two people in the same application may hold seats at different prices.

Renewal

A renewal is the date on which a subscription continues for another term, usually with the seat count and plan in place on that day. Most contracts allow fewer seats or a cheaper plan only at renewal, which makes it the date a saving has to be ready for. See preparing a SaaS renewal.

Notice date

The notice date is the last day on which a contract can be cancelled or changed before it renews automatically. It lies ahead of the renewal date by the notice period the contract sets. A finding that has to be acted on before a notice date is a decision with a deadline.

Inactive seat

An inactive seat is a paid, assigned seat whose holder shows no sign-in and no action in the vendor's data for a period the organisation has set. It is a candidate for removal and not a verdict, because activity data can miss use through an integration or a shared login, so the holder is asked first. See finding unused software licences.

Service playbook

A service playbook is the recorded knowledge about one software service: its tiers, the usage that can be measured per seat, the offers its vendor makes and the checks that apply. In Cureman a playbook is data and not program code, and its thresholds are defaults an organisation can change. See the licence optimisation module.

Right-sizing

Right-sizing means moving a seat to the cheapest plan that still covers what its holder uses. It compares the measured usage of each seat with a threshold for the tier and asks the holder where the data cannot answer. See right-sizing SaaS plans.

Shadow IT

Shadow IT is software that is used and paid for in an organisation without the knowledge of IT or procurement. It shows up on invoices and expense claims and not in the application inventory. Comparing the software suppliers on invoices with the inventory makes it visible.

AI costs

Token

A token is the unit in which a language model reads and writes text, roughly a short word or a part of a longer one. Providers price their models per token, so the cost of a request is its token counts multiplied by the price of each kind of token. See LLM token costs explained.

Input, output and cached tokens

Input tokens are what a request sends to a model, output tokens are what the model writes back, and cached tokens are input tokens the provider has kept from an earlier request. Each kind has its own price. Output usually costs more per token than input, and input read from the cache costs less.

Context window

The context window is the maximum number of tokens a model can take into account in one request, the conversation so far and the answer included. A long conversation is sent again in full with every new message, so its input cost grows with every turn.

Prompt caching

Prompt caching is a provider feature that stores the beginning of a prompt, so that later requests starting with the same text are billed at a lower rate for that part. It pays for long, repeated beginnings such as instructions or documents, and it stops applying from the point where the text changes. See prompt caching and context.

LLM gateway

An LLM gateway is one endpoint that sits between an organisation's applications and its model providers and forwards every request. Because all traffic passes through it, it can meter tokens and cost per request, attribute them to keys and users, enforce budgets and route requests. See what an LLM gateway is and the AI gateway module.

Upstream

An upstream is the model provider or compatible endpoint to which a gateway forwards a request. An organisation registers its upstreams with its own provider keys, and each gateway key leads to one upstream.

Model routing

Model routing means serving a request from a cheaper model than the one the application asked for when the request is simple enough. A classifier labels each request, and a policy the organisation controls maps the label to a model. The saving is the price difference for the tokens actually used, minus what classifying cost. See LLM model routing.

Context trimming

Context trimming means sending a model only the part of a long conversation that the current message needs. In Cureman the caller's own history is never changed: only the copy sent to the provider is shortened, and only when the avoided input cost is computed to exceed what prompt caching would have saved and what the selection costs. See the context optimiser module.

Budget

A budget is a limit on what a gateway key or a single user may spend on models in a month. When it is used up, further requests are refused until the next month begins or an administrator raises it. See AI budgets and limits.

Cost attribution

Cost attribution assigns the cost of each model request to whoever caused it: a key, a tag such as a project or a team, or a user. It turns one provider invoice into spend per team and per person. See allocating AI costs to teams.

Chat workspace

A chat workspace is a chat interface for employees that is connected to models through a gateway and paid by the tokens used instead of by a seat licence per person. Usage is attributed to each person, who chooses from the models the organisation allows. See seat licence or pay per token and the AI workspace module.

Invoices and spend

E-invoice

An e-invoice is an invoice that is issued, sent and received as structured data which software can process without reading a picture of a page. A PDF that only shows the invoice is not an e-invoice in this sense. See e-invoices: XRechnung and ZUGFeRD.

EN 16931

EN 16931 is the European standard that defines which information an electronic invoice contains and what each element means. It allows two XML syntaxes, UBL and UN/CEFACT CII. National and sector formats such as XRechnung, ZUGFeRD and Peppol BIS Billing build on it.

XRechnung

XRechnung is the German specification of EN 16931: a purely structured XML invoice without a visual page. It is used above all for invoices to German public bodies. Software can read it completely, so no model has to transcribe anything.

ZUGFeRD and Factur-X

ZUGFeRD is a hybrid invoice format: a PDF a person can read, with the same invoice embedded as XML that software can process. Factur-X is the name of the same format in France. Depending on the profile used, the embedded data follows EN 16931.

Peppol

Peppol is a network and a set of specifications for exchanging electronic business documents such as invoices and orders. Senders and receivers connect through access points instead of connecting to each other, and its invoice format, Peppol BIS Billing, follows EN 16931.

Line item

A line item is one row of an invoice: a description, a quantity, a unit, a unit price and an amount. Checks and analysis work at this level, because a price change or a period billed twice is visible per item and not in the invoice total.

Duplicate invoice

A duplicate invoice is an invoice for the same delivery that arrives or is booked a second time. An exact duplicate has the same supplier, number and amount; a near duplicate has the same supplier, amount and date under a different number. See detecting duplicate invoices.

Price agreement

A price agreement is a unit price agreed with one supplier for one item, valid from a date and, where it says so, only for certain delivery sites. Later invoices are checked against it, and a line billed above it is a finding. See checking invoices against price agreements.

Spend analysis

Spend analysis adds up what an organisation bought, by category, supplier, item, month and delivery site. It is built from invoice line items and shows where the money goes, which is the starting point for tenders and consolidation. See spend analysis basics and the invoice audit module.

Spend category

A spend category is a group of similar purchases, such as packaging, office supplies or software, that can be put to the same suppliers. Categories decide how spend is added up and which items belong in one request for quotation.

Tail spend

Tail spend is the part of spend made up of many small purchases from many suppliers, each too small for anyone to negotiate. Seen together, these purchases can often be bundled with fewer suppliers. See tail spend.

Maverick buying

Maverick buying is purchasing outside the agreements and channels an organisation has set up, for example buying an item from another supplier although a price was agreed for it. On invoices it shows as an item under an agreement that was bought elsewhere at a higher price.

Delivery site

A delivery site is an address to which goods or services are delivered, read from the invoice. In Cureman it is labelled with whether the invoice printed it as a delivery address or as the address the invoice is made out to, and a price agreed for certain sites applies only to deliveries there. See sourcing for multiple sites.

Sourcing

Request for quotation (RFQ)

A request for quotation, RFQ for short, asks suppliers to state their prices and terms for a defined list of items and quantities by a deadline. Because every supplier answers the same list, the offers can be compared by arithmetic. See writing a request for quotation.

Sourcing event

A sourcing event is one round of asking several suppliers for offers on the same items, from the request through comparison and negotiation to the award. In Cureman its items, yearly quantities and today's prices come from invoices. See the sourcing and negotiation module.

Offer comparison

An offer comparison prices every supplier's offer for the same yearly quantities, shipping included, and sets it against today's cost. Only offers that quote every item are ranked, because an incomplete offer cannot be compared as a whole. See comparing supplier offers.

Negotiation round

A negotiation round is one exchange in which suppliers who made an offer are asked to improve it. What is asked for is derived from the comparison, and each message is approved by a person before it is sent. See negotiation rounds with suppliers.

Target price

A target price is the price per item a buyer asks a supplier to meet in a negotiation round. In Cureman it is computed as the lower of the best price another supplier offered and today's price reduced by the target the event sets. It is put to a supplier only where it lies below that supplier's own offer.

Award

The award is the decision to give the business of a sourcing event to one supplier or, with several delivery sites, to one supplier per site. The awarded prices become price agreements that later invoices are checked against. Awarding is always a person's decision.

How the platform works

Finding

A finding is something deterministic code detected in an organisation's data, stated with its amount, its calculation and its confidence. Examples are seats without activity, a duplicate invoice or a price above an agreement. A finding with an unverified input is labelled and kept out of totals. See how it works.

Decision

A decision is something that waits for a person: an approval, a review or a choice. All decisions of an organisation are in one list, ordered by deadline and worth, and each person sees those their role may decide. See the procurement decision list.

Worth

The worth of a decision is the saving of the finding or the offer behind it, copied with a note saying whose figure it is. A decision without such a figure shows none, and a figure that rests on something unverified is shown and not added up.

Action plan

An action plan is the list of steps that carry a finding to done: manual steps, questions to seat holders, drafted messages and, where the organisation allows it, a change at the vendor. Each step with an effect outside the organisation is prepared, shown in full and approved by a person.

Autopilot

The autopilot is the part of Cureman that prepares the first step of a new finding: a draft campaign, a drafted message or a draft sourcing event. It prepares and never approves. It does not send, change anything at a vendor, switch a policy or book a module.

Conversation

A conversation is everything exchanged with one counterpart about one thing, such as a supplier on a sourcing event or a vendor's contact about a finding. Replies are filed with their conversation, and a clock tracks how long an answer has been outstanding.

Follow-up

A follow-up is a message that asks again after a conversation has gone without a reply for a set number of working days. It is drafted in fixed wording without figures and waits for approval. Only the first one may go out by itself, and only where the organisation has switched that on. See following up with suppliers.

Escalation

Escalation means passing an unanswered question on to someone else, as the approved plan says. A question to an employee is followed by a reminder and then, if the plan includes it, goes to the person's manager; a conversation with a supplier that stays silent comes back to a person. See asking employees about licences.

Agent task

An agent task is a job that can only be done in a website, prepared in full, approved by a person and carried out with every step recorded. The runner is the program that carries it out: a browser driven by a model, which stays on the hosts of the task, enters exactly the approved values and by default stops before a submit until a person confirms. See browser agents for vendor portals.

Access vault

The access vault holds the logins an agent may use on the website of a vendor or a supplier, stored encrypted. A login is released only to the runner and only for a running task whose hosts it covers. The runner types it into the page, so the model never sees it. See security.

MCP (Model Context Protocol)

The Model Context Protocol, MCP for short, is an open protocol through which an AI assistant calls tools and reads data that another system offers. In Cureman it lets an organisation's own assistant read purchasing data and propose requests for quotation. See MCP in procurement.

Assistant key

An assistant key is the credential with which an organisation's own AI assistant connects over MCP. It can read, optionally without amounts, and with the matching scope it can leave a proposal for a person to decide. No assistant key can send, approve, award or change anything.

Audit log

The audit log is the record of every change in an organisation: who made it and what changed. What the platform did by itself appears there as well, marked as automatic.

White label

White label means running a product under an organisation's own name. In Cureman an organisation sets its product name, logo and accent colour and can serve the dashboard and the gateway endpoint under its own host names, once a DNS record has proved control of each name.

Deterministic calculation

A deterministic calculation gives the same result every time for the same data and can be shown step by step. In Cureman every displayed figure is computed this way, and a model never produces a number that is shown as fact. See why AI must not invent figures and the principles.

Unverified price

An unverified price is a price from the supplied catalogue or from research that no person has confirmed yet. It is labelled wherever it appears, and figures that rest on it are shown and not added into totals.

One-off and yearly saving

A one-off saving is an amount recovered once, such as a duplicate invoice, and a yearly saving recurs every year, such as a removed seat. The two are shown and added up separately, so a refund never inflates the yearly figure. See proving procurement savings.

Questions and answers

What is the difference between a finding and a decision?

A finding is something code detected in the data, with its amount and calculation. A decision is something that waits for a person, such as approving the message that acts on a finding. One finding can lead to several decisions over time, and they share the finding's figure, which is counted once.

What is the difference between XRechnung and ZUGFeRD?

XRechnung is a purely structured XML invoice without a visual page. ZUGFeRD is a hybrid, a PDF a person can read with the same invoice embedded as XML. Both build on the European standard EN 16931, and both can be read by software without a model transcribing a picture.

What is the difference between a request for quotation and a sourcing event?

A request for quotation is the message that asks suppliers for prices on a defined list of items. A sourcing event is the whole round around it, from choosing the items and suppliers through the comparison of offers and the negotiation rounds to the award.