Choosing your components
Work down these tables in order. The orchestrator comes first, because a document is always uploaded or imported into one.
Which orchestrator?
| Kind | Use it when | What you get |
|---|---|---|
| Smart Extract | You need named values from every document. | A value for each field you defined, which a person can correct before it leaves. |
| Smart Classes | You only need to know which type a document is. | A class for each document, with no fields. |
| Smart Queue | One stream carries several document types, or needs both fields and redaction. | Each document sent down a route, sorted by category first if you choose, that extracts with a Smart Extract, redacts with a Smart Redact, or both. |
| Smart Redact | Sensitive content must not be seen downstream. | A redacted copy for everyone, and the original only for roles granted IntELIEDocs · Original document - View. |
A Smart Queue uses orchestrators you have already published: the Smart Extract or Smart Redact each route runs and, if it sorts documents first, a Smart Class with a different Smart Extract for each category. So you set up two or more records rather than one. All four kinds share the same tabs; see Orchestrator basics.
Connector, or upload by hand?
| Situation | What to use |
|---|---|
| A few files now and then | No connector. Use Upload Document; see Uploading documents. |
| Documents already arrive in a mailbox, on a file server, in a cloud drive or from an API | An import connector, attached on the orchestrator's Import Channels tab |
| Results must reach another system | An export connector, attached on Export Channels |
| Both ends unattended | One of each; see Import and export channels |
A connector's type is fixed once you create it, so read Choosing a connector type first. Database connectors (SQL Database and MongoDB) bring in rows rather than documents, so they feed a knowledge base and cannot be attached to an orchestrator.
Knowledge base, or orchestrator?
| You want to | Use |
|---|---|
| Read the same fields from every document | An orchestrator, above |
| Ask questions in plain language and get answers that show their sources | A knowledge base |
| Both | Both. They keep separate copies, and a knowledge base is filled by its own uploads or by a source |
Voice IntELIEgence?
Only if your input is recorded conversations rather than documents. A voice application transcribes and analyses them, and its Knowledge base setting can send what it finds into a knowledge base you already have.
Human review?
| Where | What turns it on | What the reviewer does |
|---|---|---|
| Smart Extract, Smart Classes | Review Required, in Edit Settings on Overview | Selects Mark as Reviewed on a document on Waiting for Review |
| Smart Redact | Is review required while you create it, or Review Required in Edit Settings afterwards | The same |
| Smart Queue | Review on its Pipeline tab: Auto, Always or Never | The same, on each route's document |
| Voice application | Require human review in its settings | Selects Approve on a conversation on Waiting for Review |
See Reviewing extracted fields and Changing a Smart Queue's pipeline.
Putting them together
An orchestrator, something to feed it, review where judgement is needed, and an export connector to deliver the result. Build them in that order. Leaving one out only shortens the chain: an orchestrator you upload into by hand already works.
Next
- Build your first workflow - all four, in order.
- Connecting the pieces - where each join is made.