CogentIntelligence
AI agents for admin and back-office teams
Every business has an admin layer: copying data between systems, sorting the inbox, chasing documents and building the weekly report. Cogent takes the repetitive part so your admin team can handle the work that needs a person.
01Where the week goes
We work with office managers, operations teams, PAs, virtual assistants and any business with a busy back office. The details differ from firm to firm, but the same few jobs come up on almost every discovery call:
- Double data entry. Typing the same details into the CRM, accounts and a spreadsheet.
- Inbox overload. Shared inboxes where nobody is sure who's replying.
- Document chasing. Signatures, forms and approvals that sit in someone's inbox.
- Reporting. Pulling numbers from four systems every Monday morning.
None of these need your expertise. They need doing quickly, the same way every time, and that's exactly the kind of work AI agents do well, with your Business Map telling them how your business runs.
Key idea
If a job is done the same way every time, a person shouldn't be doing it every time.
02Agents built for admin and back-office teams
These are the agents we most often switch on first in this sector. Each one reads your second brain, works around the clock and asks you before anything that matters.
- Trigger
- Every email to a shared inbox
- Does
- Labels, routes to the right person and drafts replies
- Asks you
- Before anything is sent
- Trigger
- Data needs entering into a system with no integration
- Does
- Keys it in as a person would and checks it
- Asks you
- If anything doesn't match
- Trigger
- Monday at 7am, or month end
- Does
- Pulls numbers from your systems and writes a plain-English report
- Asks you
- Nothing; it only reads
Your first agent is the one with the highest score on your discovery call, not the one on the top of this list. Most firms add a second and third over the following months.
03Ask your business anything
Because every agent writes to the same memory, you can ask questions in plain English that used to mean an afternoon in spreadsheets and inboxes.
YouWhich supplier invoices haven't been approved yet?
CogentSix, worth £3,920. Four are waiting on the same manager.
YouWho's our contact at Midland Fleet and when did we last speak?
CogentSarah Doyle, accounts. Last contact 2 October about a credit note.
04Your Monday briefing
At 7am every Monday you get a short briefing built from your own tools: what came in, what's stuck and what changed.
05Find your peak
On the call we score every repeating job by hours a week × how easy it is to hand off (1 hard, 2 medium, 3 easy). The highest score gets built first. Here's how a typical list might look.
| Job | Hours a week | Ease | Score |
|---|---|---|---|
| Copying data between systems | 6 | 3 | 18 |
| Sorting the shared inbox | 5 | 3 | 15 |
| Chasing signatures and forms | 3 | 3 | 9 |
| Weekly reporting | 3 | 3 | 9 |
| Diary management | 3 | 2 | 6 |
Example · dummy data. Score = hours a week × ease.
In this example, copying data between systems is the peak. Run your own numbers with the calculator on our home page, or bring them to the call.
06Where approvals sit
Every agent starts on a short lead. Here's a typical starting point for this sector. Yours is written into your Business Map.
Does alone
- Sorting and routing email
- Entering and checking data
- Building reports
Asks first
- Anything sent outside the business
- Payments and approvals
- Deleting records
Never
- Approving spend
- Changing bank details
Changes to bank details and payment approvals always need a person. That's a fraud control, not just a preference.
07Works with the tools you already use
We connect to what you have rather than asking you to move. Common tools in this sector include Microsoft 365, Google Workspace, Xero, QuickBooks, Sage, HubSpot, Pipedrive, Monday.com, Asana, DocuSign and legacy portals. If a system has no integration, a Portal Operator agent can work it the way a person would.
08Questions
Can AI do data entry between two systems?
Yes. Where systems connect, agents move data directly. Where they don't, a Portal Operator keys it in as a person would and checks it.
Is an AI admin assistant better than a virtual assistant?
They work well together. Agents take the repetitive work around the clock, and people handle judgement and relationships.
Can AI manage a shared inbox?
It labels, routes and drafts replies, and nothing is sent without approval until you've set it to.
Will it replace my admin team?
It removes the repetitive part of their week. Most teams use the time for work that was always being put off.