Admin and operations
The Keeper
Order from chaos; coordinates the unseen.
Motif · Interlocking keys, a quiet mechanism
Figure — Reported time saved
~11 hrs
a week reported by digital workers — minus 6.4 hrs a week spent managing the tools
Handles data entry, scheduling, invoice matching and status-chasing, so people are freed to make judgement calls and design better processes.
A week reported by digital workers — minus 6.4 hrs a week spent managing the tools. Every figure on this page is traced to a named source below, with its method and its limits.
Read the evidence01
What these agents do
Operations agents run back-office processes: matching invoices, handling exceptions, onboarding vendors, checking policy compliance, coordinating schedules and handovers, and producing audit trails.
02
Genuinely good at
- High-volume, rule-shaped work with a clear definition of done
- Producing a complete, timestamped record of what was done
- Straight-through processing where master data is clean
- Chasing, reconciling, and coordinating without fatigue
03
Genuinely bad at
- Genuine exceptions, which is where the cost usually sits
- Operating on inconsistent or duplicated master data
- Judgement calls that carry financial or contractual consequence
- Anything that assumes an undocumented internal convention
04
The risks that matter
- Financial: an agent with payment authority is a control question, not a productivity question.
- Control: segregation of duties and four-eyes rules must hold for non-human actors too.
- Audit: if the action log is incomplete, the process is not auditable.
- Concentration: back-office agents accumulate access across many systems.
05
How to evaluate one responsibly
- Ask for the value threshold above which a human must approve, and whether it is enforced in the product.
- Confirm the action log is immutable and exportable.
- Pilot on exceptions, not on the clean path.
- Involve internal audit before the pilot, not after it.
Now compare what is actually declared.
The registry holds each agent’s stated facts — autonomy, oversight, compliance, residency, sustainability disclosure — with provenance on every field.
Compare admin and operations agents06
The evidence
- Independent research
- No commercial interest in the outcome. Method published.
- Analyst research
- Structured method, but the publisher sells advice in this market.
- Practitioner survey
- Self-reported. Good for direction, unreliable for magnitude.
- Vendor or customer material
- A claim about the publisher's own product. Not a finding.
Figures below are reproduced as published. We label who paid for the work and what the method can and cannot show. Where a number is self-reported or vendor-published, we say so next to the number rather than in a footnote.
4 of 4 sources
- 01Analyst research
Currently demonstrated technologies could automate activities accounting for about 57% of US work hours — 44% via agents, 13% via robots — implying around $2.9tn in annual US value by 2030.
- Source
- McKinsey Global Institute — Agents, robots, and us: skill partnerships in the age of AI25 November 2025
- Method
- Activity-level modelling across occupations, scenario-based value estimate.
- What it does not show
- A modelled ceiling under adoption assumptions, not a forecast and not an observation. Value is technical potential, not booked savings.
- 02Practitioner survey
Digital workers report AI saving about 11 hours a week — but spend 6.4 hours a week “botsitting”, and only 13% say AI significantly improved company performance despite 87% using it.
- Source
- Glean Work AI Institute — Inaugural Work AI IndexJune 2026
- Method
- 6,000 digital workers across the US, UK and Australia, surveyed December 2025 – January 2026.
- What it does not show
- Self-reported, and published by a supplier in this market. The gap between hours saved and performance improved is the most useful figure in it.
- 03Practitioner survey
About 67% of knowledge workers spend three or more hours a day on manual coordination that automation could remove.
- Source
- Aggregated reporting (McKinsey, Gartner, Formstack) — Manual-work and coordination-burden reporting2022–2024
- Method
- Aggregation of separate surveys with differing definitions of manual work.
- What it does not show
- The sources define “manual work” differently, so the aggregate is directional only. Widely re-quoted without its original method.
- 04Vendor or customer material
Invoice processing drops from around 45 minutes to under three minutes per invoice.
- Source
- Practitioner and supplier case reports — Accounts-payable automation write-ups2024–2025
- Method
- Single-process before-and-after, usually on straight-through-eligible invoices.
- What it does not show
- Excludes exceptions, which is where most accounts-payable cost sits. A per-invoice figure is not a departmental figure.
We take no payment from any organisation named on this page, and no source is listed or omitted on commercial grounds. If a figure here is wrong or out of date, tell us and we will correct it with the date of the change.
Ideas in this domain
Continue