Glossary
What is behind the terms that come up again and again when people talk about AI. Short, without jargon and with everyday examples.
What AI does for you
AI that answers questions and drafts text, in a chat or directly in your software.
An AI assistant such as ChatGPT, Claude or Microsoft Copilot understands questions in plain language. It answers, summarizes or writes drafts. Set up correctly, it works with your company knowledge and cites the source.
Example
An assistant answers questions about contracts and points to the right document.
Find what your company already knows instead of working it out again.
Knowledge search looks through file shares, wikis and emails by meaning, not just by keywords. It finds the right passage in the right document, even when different words are used there.
Example
Find the right solution for a customer request in past projects.
Read, check and pass on documents automatically.
AI reads invoices, contracts or forms, picks out the key details and checks them against rules or master data. The results go straight into the right system.
Example
Incoming invoices are read, checked and prepared for booking.
AI agents complete multi-step tasks on their own, within clear limits.
An AI agent plans the steps, uses tools such as your CRM or email, and completes a task from start to finish. In a workflow, it works within fixed limits. For important decisions, a person reviews the result.
Example
Sort customer requests, gather the information and prepare a reply for approval.
Recurring processes run on their own.
Software takes over regular tasks such as moving data, creating reports or triggering approvals. With AI, you can also automate steps that involve reading text or making simple decisions.
Example
Monthly reports from several systems are compiled automatically.
Connect AI securely to the systems you already use.
To help in daily work, AI needs access to your CRM, ERP, file shares and email, with the right permissions. System integration connects these sources securely to your AI applications.
Example
An assistant knows the customer data in your CRM and the orders in your ERP.
Costs and processes
Making visible which processes cause which costs.
A cost structure analysis assigns operating and personnel costs to the processes where they arise. That reveals cost drivers that disappear in overhead, and savings potential can be evaluated in a targeted way.
Example
A large share of administrative time goes into retyping and matching documents.
A case runs through fully automatically, without a person stepping in.
With straight-through processing, software handles a case from start to finish. Only exceptions go to a person. AI raises the share because free text, PDFs and emails can be processed too. The straight-through rate is a key metric for process automation.
Example
An invoice is read, checked and posted without anyone touching it.
Reconstructing processes from the log data of your systems.
Process mining analyzes which steps a case actually went through in your ERP or CRM. That reveals detours, waiting times and variants. For many midsize companies, structured process mapping is enough; process mining adds to it when good data is available.
Example
On average, an order goes back and forth between sales and back office three times.
The process from incoming order to incoming payment.
Order-to-cash covers order entry, confirmation, delivery, invoicing and payment. Because many departments and systems are involved, this is where hand-offs and long lead times often arise.
Example
AI reads orders from emails and creates them directly in the ERP.
The process from requisition to the paid invoice.
Procure-to-pay covers requisition, purchase order, goods receipt, invoice verification and payment. Matching invoice, purchase order and goods receipt is a classic case for automation with AI.
Example
Invoices are checked automatically against purchase order and goods receipt.
Develop, buy or outsource: calculating the options.
A make-or-buy analysis compares costs, risks and dependencies of in-house development, standard software and outsourcing. For AI solutions, operating costs, data protection and vendor lock-in belong in the comparison too.
Example
A standard tool for invoice checks versus an agent that writes into your own ERP.
How we measure impact
Decide upfront what a use case should achieve.
For each use case, we define which metric should change and by how much, for example processing time or error rate. We also record the baseline value before anything changes.
Example
Processing time per quote: record the baseline, set the target, measure again later.
Measure what has actually changed.
After launch, we compare the metrics with the baseline values: after 4, 8 and 12 weeks, then quarterly if you want. This shows whether a use case delivers and where it needs work.
Example
Usage rate and processing time three months after launch.
You stay independent, of us and of single vendors.
The code and documentation belong to you. We use open standards such as MCP. That way, you choose the model and cloud freely and can export your data at any time.
Example
You switch AI models without setting up the connections again.
The starting value of a metric before anything changes.
The baseline is the starting value you measure against later, for example today’s handling time per case. Without a baseline, no saving can be proven.
Example
Today a proposal takes 90 minutes on average. That is the baseline.
The point from which an investment has paid for itself.
The payback period shows how long it takes until savings offset the costs of delivery and operation. It belongs in every business case and is measured again after launch.
Example
An automation costs once and saves every month. From the month in which both are equal, it pays off.
Technology, briefly explained
Search your own knowledge first, then answer.
With RAG, the AI first searches a document collection for relevant passages and bases its answer on them, with sources. It is the simplest way to give an AI access to your knowledge, and it takes only a few days to set up.
Example
An assistant answers questions about the quality manual from a SharePoint folder.
A layer of knowledge that makes your systems usable for any AI.
A knowledge layer connects several sources such as file shares, CRM and ERP into one layer of knowledge that stays up to date. It keeps permissions, recognizes relationships and makes the knowledge available to every AI application. Our product for this is the Company Brain.
Example
Copilot, ChatGPT and your own agents use the same, verified company knowledge.
An open standard that lets AI applications access data and tools.
MCP defines how an AI such as ChatGPT, Claude or Copilot talks to data sources and tools. A connection built on MCP works with many AI applications instead of being built separately for each one.
Example
A connection to your CRM that works in Claude and in Copilot alike.
A person decides at the important points.
In automated processes, a person approves at defined points whether things move on, for example before a reply is sent or a booking is made. That keeps your team in control.
Example
The agent prepares the reply, a team member checks and sends it.
A small, real test of whether a solution works with your data.
A proof of concept checks on a real use case whether data, integration and model hold up. It is deliberately small and done within a few weeks. Then you decide on the rollout with solid numbers.
Example
For one month, an agent processes real incoming invoices in parallel to the existing process.
Free AI Readiness Check
In the free AI Readiness Check, we find out which applications really make a difference in your company.
