How it works
Most manual data work is a handful of steps, repeated. If your organisation has repeating workflows, even if loosely defined or messy, automating them usually brings significant savings in time and cost, and gains in quality. Agents run those steps as software and keep running them without you. The graphic below compares each step, by hand and automated.
Search
Manual
- Hunt for the right documents and data — known and unknown
- Constrained by cognitive load
Automated
- Sweeps known sources, finds new ones
- Keeps logs of findings
- Can quickly filter large volumes
Sources
Manual
- Opened individually by hand
- Constrained by cognitive load, language, domain expertise
Automated
- Fetched in bulk, scheduled, multilingual
- Can be fed internal domain expertise to identify key information
- Handles large volumes and messy structure with comparative ease
Extract
Manual
- Read it, find the data, copy it
- Error prone
- Requires contextual knowledge
Automated
- Every field pulled at once
- In-line validation
- Contextual knowledge can be codified
Structure
Manual
- Paste into a spreadsheet or database
- Limited fields due to time and effort
Automated
- Automated entry into a typed and easily extensible data schema
- Validated on the way in
- Unconstrained number of fields and rows
Verify
Manual
- Spot-check a sample
- Hope the rest holds
Automated
- Every row can be verified
- Exceptions flagged for human in loop
Act
Manual
- Analysed, reported, emailed at a later time
Automated
- Instantly consumable via the systems or people that need it
Time & cost
Manual
- Hours, days or weeks every iteration
- Costs and volume scale together
Automated
- Minutes, unattended or human-piloted
- Marginal cost increase with scale