Loading

Language and Image Analysis

People often expect a business platform to detect what language a message is in, or label what a photograph shows. On ActiveManage neither is a feature you can switch on today: there is no screen, setting, workflow step or automatic use for them. This page says plainly what is and is not available, and which features cover the nearby needs.

Where to find it

Architect Panel → Configuration:

  • AI Settings — the ocr role, for reading text in images, and the image and video generation settings

Architect Panel → Automation:

  • Tasks — Document Text Extraction, which reads text in scanned images

Architect Panel → Activity:

  • AI Usage — the cost of anything read by an AI model

What is not available

  • Language detection: there is no setting that records or acts on the language of a message or record.
  • Image labelling or screening: nothing tags photographs with what they contain, or checks uploaded images before they are published.
  • Sentiment and entity detection: not available as features either.

The platform does include building blocks for these, based on AWS language and image services and using the installation's AWS credentials, but nothing in the platform calls them. Using them means custom development, and the cost lands on your AWS account.

What covers the nearby needs

  • Text in an image. A sign, a serial plate or a scanned letter is read into the document index by the configured OCR reader, so it becomes searchable. See Document & Image Intelligence.
  • What a message is about. Inbound messaging rules can match on what a message is about using TypeSafe Jev; see Is About (AI) Message Rules.
  • A model looking at a picture. The AI gateway accepts images as input, so a custom feature can ask a model a question about a photograph, logged and charged like any other AI call.
  • Making images. Generating pictures and video is a separate feature; see AI Images & Video.

If you need one of the missing capabilities

  1. Write down the decision it would drive: route a message, tag a record, hold an image for review.
  2. Check whether one of the features above already answers it, for example matching on what a message is about instead of its language.
  3. If not, raise it as custom development, and settle where the data will be processed and what it costs before work starts.
  4. Keep any automated result as a suggestion a person confirms, and store its confidence alongside it, so a threshold can be tuned later from real data.

Using results from a model

Whatever reads text or images produces a likelihood, not a fact. Short text is genuinely ambiguous, and photographs taken at an angle or in poor light read badly. Use results to pre-fill, prioritise or suggest, with a person deciding, and never to publish or refuse something unattended. Where automated analysis affects how someone is treated, say so in your privacy information.

What goes wrong, and how to tell

  • A process assumes language is recorded: it is not. Check with your developer what the custom code actually stores.
  • Text inside photographs is not searchable: no OCR reader is configured, or the extraction task is off.
  • Unexpected AWS charges after custom work: the AWS building blocks bill your AWS account directly and do not appear on AI Usage.

Worked example

A service wants replies sent in the language a customer wrote in. Reading this page, the team finds there is no language detection to switch on. They instead add a preferred-language field to the contact record, filled in at first contact, and route on that. It is simpler, never guesses wrong on a three-word message, and needed no custom development.

Recommendations

  • Do not plan around language detection or image labelling as if they were features.
  • Prefer a field a person fills in where the answer drives how someone is treated.
  • Use OCR for text in images, and the gateway for custom questions about pictures.
  • Settle cost and data location first if you commission custom analysis.