Support owners
Test internal summaries for a selected queue. A small team reviews results before wider use.
Record: Use case, responsible people and approved users
AI & data protection · An example approach
Using AI in support:
a five-step example.
A suitable online service, selected knowledge sources and an agreed setup make the first rollout manageable. This example shows the decisions to take in order.
Discuss your rollout →Example: internal summaries and approved service guides for a support team.
- 01
Start with a clear use case
Choose internal initial proposals or ticket summaries for a small support team, for example. Define which requests are included and who reviews the results.
Outcome: a manageable pilot with named owners.
- 02
Prepare business access and knowledge sources
For company knowledge, we recommend starting with Chat Data, with Chatbase as an alternative. Set up a business account and begin with a few approved manuals or FAQs. OpenAI is also available for direct AI tasks.
Outcome: a suitable service with a small set of maintained sources.
- 03
Agree the data processing arrangements
Include your data protection and IT teams. Record the purpose, permitted data, legal basis and responsibilities. Agree the necessary processing terms and review locations, subprocessors and access. For relevant international transfers, include the appropriate mechanism and, where needed, SCCs and a transfer assessment.
Outcome: documented approval and suitable provider agreements.
- 04
Configure access and the permitted data
Use business accounts with limited permissions. Define which ticket content and attachments may be processed, retention periods and deletion. Review available safeguards in the chosen plan. Start testing with fictional or approved examples.
Outcome: a documented data flow for the selected use case.
- 05
Test real tasks and approve the rollout
Review results against typical support questions. Show staff how to assess suggestions and keep sources current. Record approval and regular reviews. Expand gradually once the first workflow works for your team.
Outcome: a tested starting point with clear team guidance.
A PRACTICAL ROLLOUT EXAMPLE
From the first summary to internal approval.
A support team initially wants internal ticket summaries only. It starts with a selected queue and a few approved sources. Existing reviews and contracts can be reused where they actually cover the planned use, avoiding unnecessary repeated work.
Privacy and IT
Check whether existing provider agreements cover the chosen plan, models and data flow. Add missing arrangements.
Record: Provider review and required agreements, including SCCs and transfer assessment where needed
Technical setup
Configure suitable business access, permitted data and approved sources. Review the service’s retention and deletion options.
Record: Approved settings and a description of the data flow
Test team
Use fictional or approved cases to review usefulness and the handling of sensitive information.
Record: A short test record and staff guidance
Internal approval
Approve the defined use for the intended users. Review additional functions and sources when needed.
Record: Approval record covering scope, owners and review date
There is no universal timetable. With owners, provider documents and test access already available, the steps can be prepared and coordinated without unnecessary repetition. Addonflow can assist with technical preparation and testing on request.
Internal review checklist
Review these points with the responsible people. Tick items here as you go; selections are not saved or submitted.
Planning aid for an initial use case. Agree the required reviews with the responsible teams in your organisation.
Planning example, not blanket privacy approval. The responsible teams determine the reviews required for your organisation.
Common questions about getting started.
Do we need every provider?+
No. Choose the service for the features you use. Chat Data is our preferred starting point for company knowledge, with Chatbase as an alternative. OpenAI can provide direct AI access. Add translation services only where needed.
Are our documents used to train an AI model?+
A document knowledge base differs from training a foundation model. The service terms, selected models and settings matter. OpenAI states that API data is not used for model training by default unless you explicitly opt in. Retention is a separate consideration.
What should we check with Chat Data or Chatbase?+
The actual plan, selected models, processing agreement, retention and deletion, subprocessors and access. Enable only models and additional features covered by your approval. Current provider documents are linked below.
Is an EU endpoint or a contract enough?+
You need suitable agreements and matching technical settings. If processing in the EU is required, clarify the appropriate provider options and internal approvals before integration. OpenAI has eligibility requirements for EU processing and special retention options.
Which records should we keep?+
A short use-case and data-flow description, provider agreements, the documented privacy assessment and a DPIA where required, approved settings, test results and staff guidance. Agree the details with the responsible teams.
Can Addonflow help with rollout?+
Yes. Integration, testing and technical setup assistance are available on request. Your responsible teams approve the data and use cases. We agree the required assistance in your offer.
Provider documents and official guidance
This guide describes an example approach. Your responsible teams adapt it to your organisation’s data, agreements and use.
AI SERVICES & COMPANY KNOWLEDGE
Connect an AI service
and use company knowledge.
Our recommended approach connects Zai-LLM to a suitable online service, without an AI server of your own. Choose access for the features your team needs.
Chat Data
Provide approved content from resolved support tickets, internal knowledge articles, manuals or service guides. Your team can use these sources through the configured Zai-LLM features.
The service also offers visual workflows. We can help you assess suitable workflows on request.
Chatbase
Another option for making company documents available for AI-assisted answers, including when your organisation already uses Chatbase.
OpenAI
Use your own business account for OpenAI’s interface (API) for summaries, reply drafts and other AI tasks. Usage is billed separately by OpenAI.
A ChatGPT subscription and API access are separate offerings.
How can we connect our own documents?
Start with selected sources in Chat Data or Chatbase. You generally do not need to train your own AI model. Maintain the content and check answers against typical support questions.
DOCUMENT TRANSLATION
DeepL & Google Cloud Translation
These services provide the dedicated translation features. Suitable API access is needed only if you use those features.
UNDERSTAND THE COSTS
Separate the module and service costs.
Buy the module licence from Addonflow and pay the selected AI or translation provider directly. Optional setup and customisation are agreed separately.
AI assistance does not have to require a large budget. A suitable model and plan can keep running costs low. Costs depend mainly on usage, the amount of content processed and the provider you choose. We can help you estimate your requirements before you start.
Already running a local AI model?+
We can integrate local AI models, for example through Ollama, OpenAI-compatible services and additional APIs on request. We review the interface and develop and test the required integration. Advice on model and hardware selection is available to suit your tasks, planned use and budget. The online services above remain our recommended starting point.