Automation Anywhere Control Room and IQ Bot: A Developer’s Guide

Quick Answer: This Automation Anywhere Control Room IQ Bot guide covers the platform’s two halves. The Control Room is the web-based nerve centre — it manages bots, devices, credentials, roles and schedules — while IQ Bot (now branded Document Automation in Automation 360) handles semi-structured documents through “learning instances” that classify documents and extract fields with ML assistance. A working developer needs both: the Control Room to deploy and monitor, and a learning instance when a process starts with a PDF or scanned form instead of a database query.

What the Control Room Actually Controls

New developers treat the Control Room as “the website where I upload my bot”. It is much more than that, and most production incidents trace back to someone misunderstanding one of its layers.

If you are coming from UiPath, the mental mapping is: Control Room ≈ Orchestrator, task bots ≈ workflows, Bot Runner ≈ Robot, Credential Vault ≈ Orchestrator assets/credentials. The vocabulary differs; the operational problems are identical.

Task Bots, MetaBots and Where IQ Bot Fits

A task bot is the unit of work: an action tree built in Bot Creator from the command library — loops, conditionals, Excel and database actions, REST web service calls, screen and web recorders. Every action has configurable error handling (retry count, log, go-to-page, end task), which is the part that separates demo bots from production bots.

A MetaBot was the older reuse mechanism — an application component with its own visual capture and logic, called from task bots with input/output variables. In Automation 360 the ecosystem has shifted toward shared sub-tasks, packages and the AARI interface for citizen developers. If you are maintaining a legacy Enterprise A2019 estate you will still meet MetaBots; if you are starting fresh, invest your reuse effort in packages and well-structured task bots instead.

IQ Bot is the document brain. In current Automation 360 branding it has become Document Automation, built around learning instances — a configured object that says: these are the document types, these are the fields to extract, this is how low-confidence output gets handled. Task bots then consume the extracted JSON/XML and push it downstream. The two halves meet in the middle: a task bot emails or receives documents, hands them to the learning instance, waits for extraction, then acts on structured data.

How a Learning Instance Actually Works

1. Create the learning instance and name the document type — invoice, purchase order, bank statement, contract.
2. Upload sample documents. Twenty clean samples train a field; fifty mixed-quality samples are better than two hundred perfect ones, because reality is messy.
3. Define the fields — invoice number, date, total, line items — and let the engine propose extraction zones, then correct them. This is the labelling step and it is where the accuracy is won or lost.
4. Train and review. The classifier assigns a document type; extractors assign per-field confidence scores. You review what falls below the threshold you set.
5. Set confidence thresholds. A common production setup: fields above 0.95 auto-pass, 0.75-0.95 go to a human validation queue, below 0.75 get rejected for reprocessing. The exact numbers depend on the field — a total amount is worth more human eyes than a page number.
6. Publish and integrate. The task bot calls the learning instance, receives structured output, and handles the validation-queue exceptions separately.

Debugging a Failing Run

When a bot fails at 2 AM, work this sequence rather than guessing:

• Bot Insight run detail — which action in the tree failed, and the actual error text. Nine times out of ten the answer is here.
• Credential Vault — did a password rotation expire the entry the bot uses?
• Device state — was the unattended machine patched, left logged out, or has its screen resolution changed?
• Package versions — did the Control Room upgrade a package under the bot since the last successful run?
• Target application — did IT patch the app over the weekend and move the field the recorder captured?

A real-world version of that last one: a finance bot keyed invoice data into an ERP screen. The ERP’s quarterly patch inserted an extra column. The bot dutifully typed an invoice number into the amount field, the ERP accepted it without validation, and the error surfaced three days later as a ₹14 lakh reconciliation gap. The fix was not clever — an assert-before-write check comparing the parsed amount to a sanity range, plus a Bot Insight business metric on “records posted with non-numeric amounts”. The lesson is that screen automation has no type system; you have to build your own.

Automation Anywhere Control Room IQ Bot Failures in Production

Document extraction degrades quietly. A new supplier sends invoices in a two-column layout; a scanned batch arrives at lower resolution than the training set; a field moves because a template changed. None of these throw an exception — the model just returns lower-confidence values, and if your validation queue is under-staffed, low-confidence items sit and the SLA slips.

Two habits prevent this. First, track extraction confidence over time as a trend, not a one-off — Bot Insight business metrics make this straightforward. Second, budget for retraining the way you budget for patching: a learning instance trained once and never revisited will drift out of usefulness within a few quarters. For how WorkFusion approaches the same labelling-and-scoring problem, see the WorkFusion IDP and orchestration explainer.

Control Room Housekeeping You Will Eventually Automate

Every team that runs more than about twenty bots ends up scripting Control Room housekeeping through its REST API — listing bots, exporting them for Git-based version control, checking device status. A representative snippet:

# 1. Authenticate — the JWT comes back in the x-auth-automationanywhere-jwt header
curl -s -D headers.txt -X POST "https://your-cr.example.com/v2/authentication" 
  -H "Content-Type: application/json" 
  -d '{"username":"svc_admin","password":"****"}'

TOKEN=$(grep -i "x-auth-automationanywhere-jwt" headers.txt | cut -d' ' -f2 | tr -d 'r')

# 2. List bots in a workspace (paginated)
curl -s "https://your-cr.example.com/v2/bots?workspaceId=12345&offset=0&limit=100" 
  -H "Authorization: Bearer $TOKEN"

# 3. Export a bot's .atmx for Git-based promotion (dev -> test -> prod)
curl -s "https://your-cr.example.com/v2/bots/67890/export" 
  -H "Authorization: Bearer $TOKEN" -o invoice_bot.atmx

The export step is the one experienced teams care about most. Keeping .atmx exports in Git gives you diff history and rollback; teams that rely only on the Control Room’s internal versioning find out the hard way that “restore previous version” is not the same as a code review trail.

How This Fits Your Career

Control Room fluency plus one real learning instance is, in practice, the mid-level threshold in Indian RPA hiring — it is what separates a ₹9 LPA bot builder from a ₹15 LPA automation developer. That salary picture is laid out in the Automation Anywhere salary guide, and if you are weighing whether to buy pre-built Bot Store content rather than build, read Bot Store vs custom bot builds before you commit.

Training Resources

Get hands-on with all of this in SkilBrill Automation Anywhere Training — Control Room administration, task bots, Document Automation (IQ Bot) learning instances, Bot Insight and real project work.

FAQ

What is the difference between the Automation Anywhere Control Room and IQ Bot?

The Control Room is the management layer — bots, users, devices, credentials, schedules. IQ Bot (now Document Automation) is the document-processing engine that classifies documents and extracts fields; the Control Room hosts and administers its learning instances.

Is IQ Bot deprecated in Automation 360?

IQ Bot has effectively been absorbed into Document Automation in Automation 360. The concepts — learning instances, classification, extraction, confidence review — carry over directly, so skills learned as IQ Bot remain marketable under the new name.

How do you call an IQ Bot learning instance from a task bot?

A task bot hands documents to the published learning instance through the document-processing command, waits for extraction, then reads the structured output (field name, value, confidence) and routes low-confidence fields to a validation queue for human review.

Do you need Control Room admin rights to develop bots?

No — developers need Bot Creator access to a workspace. But understanding roles, licenses, devices and credential vaults is what lets you debug failed runs instead of raising tickets, and it is a real mid-level salary separator.

What license is needed for IQ Bot or Document Automation?

Document Automation capacity is licensed separately from Bot Creator and Bot Runner licenses, typically per document or capacity unit. Confirm the entitlement before promising a business case — pilot teams regularly discover this only at production rollout.

Written by SkilBrill Training Institute

October 15, 2026 · 8 min read