AI as a step in the process, not a tool beside it
What separates an experiment from an operating procedure is not the model but everything around it: when it is asked, on what, who checks the result and where it lands. That is exactly what a flow defines.
OpenAI — without a contract of your own
“OpenAI” is an action like any other in a flow: you write the prompt, insert values from the entry through expressions and optionally add further instructions. Unlike Slack or Hubspot, this step needs no account and no API key of your own — it runs through ApptiveGrid.
- Ready immediately — no OpenAI contract, no API key, no separate billing
- A prompt with your data — field values flow into the request through expressions
- Additional instructions — set the format, the language and the answers you allow
- From the Pro plan — as with the other AI steps
Categorise this support request and summarise it in two sentences:
{{Message}} No OpenAI contract and no API key needed — the step runs through ApptiveGrid.
The Document Analyzer reads what comes in
Invoices, delivery notes, completed forms: the “Document Analyzer” reads an attached document with AI and returns the values it contains, ready to be written into fields. For plain prose there is “PDF to text”, for recordings “speech to text”.
- Document Analyzer — pulls the values out of a document instead of having them retyped
- PDF to text — when the prose itself is needed, for a summary say
- Speech to text — a voice recording from a form becomes text you can work with
- Right inside the process — the flow writes the values into the same entry the document hangs off
Straight into fields — no retyping
The result lands in fields, not in a chat
The flow writes the answer back into the entry — as a category, a summary, a draft. That makes it a value you can filter, report on and reuse in the next step, rather than a fleeting piece of text.
- Into defined fields — single select for the category, text for the summary
- Reportable — filter, sort and show it in views like any other value
- Reusable — later steps read it and act on it
- Traceable — result and timestamp stay with the case
Approval before it counts
Between a suggestion and its effect, most processes want a person. A page as a flow step puts the result up for review: confirm, correct or reject. Only then does the process continue — and what was decided sits in the entry.
- A suggestion, not a fact — the result waits for confirmation before it takes effect
- Correctable — the page accepts an amended version
- A router in front — conditions check whether the result is plausible at all
- Documented — who approved it and when stays with the case
The AI steps and what belongs around them
What is ready to use, what the process lays down and what stays in your hands. Actions marked “Pro” are only available on the highest plan.
AI steps in a flow
- OpenAI Pro
- Sends a prompt to ChatGPT and returns the answer. Inputs: a prompt and, optionally, additional instructions.
- Document Analyzer Pro
- Analyses an attached document with AI and returns the values it contains.
- PDF to text Pro
- Reads the text content of a PDF — the step before, when the AI should get the prose.
- Speech to text Pro
- Turns an audio recording into text, such as a voice recording from a form.
- Generate with AI Pro
- Describes the JavaScript you want in words instead of typing it.
- HTTP request
- The open route to any other AI service with an interface — included on every plan.
What the process defines
- The trigger
- When the AI is asked at all.
- The data
- Which fields go into the prompt — and which do not.
- The target field
- Where the answer is written, and stays reportable.
- The conditions
- What happens when the result is not plausible — the router branches.
What stays in your hands
- The approval
- A page as a flow step puts the suggestion up for review.
- The data
- Only what is in the prompt leaves the platform. Everything else stays in the grid.
- The provider
- OpenAI as a ready-made step, or another service through an HTTP request.
- The audit trail
- Result, timestamp and approval sit in the entry.
From the blog
Two pieces on making AI usable in a company without giving up control. The articles are written in German.
KI im Unternehmen: So vermeiden Sie gefährliche Anbieterabhängigkeit
Warum Unternehmen KI nicht direkt in kritische Prozesse einbauen sollten – und wie ApptiveGrid stabile Workflows mit flexibel austauschbaren KI-Diensten verbindet.
DigitalisierungClaude Code, Codex & Replit Agent: KI nutzen, ohne die Kontrolle über Ihre Prozesse abzugeben
KI-Coding-Agenten bauen heute in Minuten ganze Apps. Warum die wichtigere Frage nicht lautet 'KI oder Plattform?', sondern wie Sie KI kontrolliert in Ihre Geschäftsprozesse einbinden.
DigitalisierungDigitalisierung in KMU
Die Corona-Pandemie hat die digitale Transformation in deutschen Unternehmen, insbesondere in KMU, beschleunigt. Was das bedeutet und wie Sie davon profitieren.
Questions about AI in workflows
How does AI get into an ApptiveGrid process?
As a step inside a flow. There are two ready-made AI actions: “OpenAI” sends a prompt to ChatGPT and returns the answer, and the “Document Analyzer” reads attached documents. Both belong to the Pro plan.
Do I need an OpenAI contract of my own?
No. Unlike Slack, Hubspot or Magenta Business, the OpenAI step needs no account and no API key of your own. It runs through ApptiveGrid and is ready to use as soon as your plan includes it.
Which data leaves the platform?
Exactly the content of the prompt — and that goes to an external service. So do not hand over personal or confidential data without checking the data protection implications first. What is not in the prompt stays in your grid.
Can I use a different AI provider?
Yes. The “HTTP request” step reaches any service with an open interface — you set the URL, headers and body yourself. That is how you plug in a European model, a self-hosted one or a specialist provider. This step is included on every plan.
Is the result taken over unchecked?
Only if you build it that way. The usual route is an approval: the flow writes the suggestion into a field and shows it on a page. The process only continues once it is confirmed.
Does the result stay reportable?
Yes. Because the answer is written into fields of the entry, it can be filtered, sorted, shown in views and used in later steps — unlike a chat transcript that ends up nowhere.
What if the AI classifies something wrongly?
The same logic applies as to any other input: a router checks the result and branches, an approval corrects it, and the entry records what was actually decided.
Build your first AI step
Start for free, create a flow, add an AI step. No credit card required.