Automation Discovery Skill: the REAL data analyst

What it actually is

uipath-automation-discovery is not a dashboard and it is not Task Mining with a chat box bolted on. It is an agent with a research loop: it interviews systems and people, assembles evidence about how work moves through a team, and returns a ranked opportunity list with the reasoning attached. The important word is evidence — every candidate it surfaces can be traced back to something it observed, not something a workshop attendee claimed in a room.

That changes who the output is for. A discovery workshop produces a list the automation team believes. This produces a list a finance sponsor can audit.

The evidence loop

The skill runs seven phases, straight from its SKILL.md. It checks in with you after each one, so you can steer before it moves on.

1. P0
   Intake — interactive

   Company context, tool inventory, org structure, privacy scope and scope size (quick scan 10 findings, standard 25, deep dive 35). It tests each source with a read-only call before mining. Rule 2: never assume, always ask.

2. P1
   Mine

   Parallel agents, one per source category. Priority order: messaging help channels → email → CRM/ERP → wiki → issue tracker → HRIS → web research. Raw signal summary checkpoint at the end.

3. P2
   Analyze

   Behavioural patterns per department, single points of failure (pseudonymised by default), proven replicable models, a department coverage map and process deep reads. Sources with fewer than 5 signals are flagged low-confidence.

4. P3
   Reflect

   Layers on public strategy and financials to find strategic gaps that behavioural data alone won't show. These become Tier 4.

5. P4
   Report

   A 4-tier prioritised report. Every Tier 1–3 finding cites a source, a metric and an affected role or team.

6. P4.5
   Estimate — on request

   Opportunity → complexity band → pack-hours → adjustment factors → contingency. The band-to-hours numbers come from YOUR matrices and Pack-Hours catalogue. If you don't supply them, it stops and asks.

7. P5
   Handoff

   Maps each Tier 1–2 opportunity to an implementation skill (uipath-rpa, uipath-maestro-flow, uipath-agents, uipath-human-in-the-loop, uipath-platform), or to uipath-planner for multi-component designs.

Wiring it to Automation Hub

The skill is only useful if its output lands where your pipeline already lives. Push each ranked candidate into Automation Hub as an idea with its evidence payload intact, so the CoE reviews the reasoning rather than re-litigating the score.

from uipath import AutomationHub, DiscoveryAgent

            agent = AutomationHub.discovery("finance-ops")
            hub   = AutomationHub(tenant="prod")

            for c in agent.rank(min_confidence=0.72):
                hub.ideas.create(
                    name=c.process_name,
                    category=c.pillar,
                    evidence=c.evidence_bundle(),   # keep the trail
                    est_hours=c.annual_hours_saved,
                    source="uipath-automation-discovery",
                )
                print(f"{c.rank:>2}. {c.process_name:<38} {c.confidence:.2f}")

Hands-on: from zero to first report

Enough theory. Here is the full path on a clean machine — install the UiPath CLI, wire the skill into your coding agent, authenticate, run a discovery and hand the results off to a build. Five steps, roughly fifteen minutes.

Reading the 4-tier report

The output is a Markdown report your agent writes into the workspace. Tiers aren't a simple ranking. Tier 1 is always replicable models: automation already working in one place that could be copied elsewhere. Tier 4 is strategic gaps tied to documented priorities. If you ask for Phase 4.5, each finding also gets a complexity band, estimated pack-hours of build effort, and a contingency buffer, calibrated from your own catalogue and never invented. Click a row to see why it landed there.

• T1 Replicate AP exception bot to Procurement band · lowfrom cataloguecontingency · low-conf + uipath-rpa

  A proven replicable model: the AP mailbox bot already triages price-variance exceptions, and Procurement does the same triage by hand. Rule 5: replication is always Tier 1.

• T2 Vendor onboarding checks band · mediumfrom cataloguecontingency · per confidence uipath-maestro-flow
• T3 Month-end accrual commentary band · highfrom cataloguecontingency · per confidence uipath-agents
• T4 Faster quarter close band · —not sizedcontingency · — uipath-planner

Where it breaks

Three consistent failure modes so far: shallow system coverage (it can only reason about what it can see, so a team living in an unconnected legacy client is invisible), over-confident ranking on low-volume processes, and sponsors who read the list as a commitment rather than a hypothesis. The first is a connector problem, the second is a threshold you set, the third is a conversation you have up front.

Watch the episode

Episode 2 · with Tertius Nel

Automation Discovery Skill · 45 min

Play on YouTube

Prefer audio? Listen on Spotify