MSDSystems
sales@msdsystems.co.uk Talk to MSD
Menu

Explore MSD Systems

sales@msdsystems.co.uk Talk to MSD

AI worker · managed service

An AI worker for the repetitive work your team keys by hand.

One workflow automated end to end by an AI digital worker, mapped and agreed before anything is built, running across the systems you already use, with every exception routed to a named person. No AI worker has run for a client yet: we scope the work and discuss it candidly, and do not claim otherwise.

Book a scoping call See AI workers
Talk to MSD
Optional
The more specific you can be, the more useful the first conversation will be.
Read the privacy notice
  • 1 named process per first engagement
  • 2 directors on every engagement
  • 0 day rates — fixed price against a written scope
Where automation stalls

Four reasons automation stops short of the actual work.

Automation products all promise the same thing. These are the four places delivery usually stops, and they are the same four places the work actually is.

01 PROBLEM 01

The tool automated the easy half

A rules-based tool handles the submissions that arrive correctly. The exceptions still land on a person, and the exceptions are where the work actually is.

02 PROBLEM 02

Nobody mapped the process first

Automation gets bought before anyone writes down what the process actually does, so nothing can be checked against it afterwards. We map the workflow, and every exception in it, before anything is built.

03 PROBLEM 03

The platform has no API

The system holding the work was never designed to be automated against. Technical discovery checks whether authorised exports, an integration or controlled screen interaction can support the process, with actions logged.

04 PROBLEM 04

Every change needs the supplier

The process moves, the automation does not, and the queue goes back to being manual. Rule changes are agreed in writing and applied as part of the service.

One workflow at a time
How one worker operates

The timesheet worker, end to end.

This is the shape of every worker: a trigger, a small number of stages across real systems, one escalation path to a person, and a logged outcome.

Trigger TIMESHEET SUBMITTED · EMAIL / WHATSAPP / PORTAL
01
Read Extracts hours, shift, site and worker from whatever format it arrives in. OUTLOOK · WHATSAPP
02
Validate Checks it against the rota, the rate card and the contract terms. ROTA SYSTEM
03
Key Writes payroll-ready lines into the run, with the source document attached. PAYROLL
04
Report Posts the batch summary and everything it could not resolve. FINANCE
Exception → payroll lead A missing shift, a rate mismatch or an unreadable submission stops and is queued for a named person, with the reason attached.
Outcome Logged. Exceptions escalated to a named person, not silently corrected or dropped.
Specialist capability

Where an AI worker has to actually work.

This is the part most automation products cannot do. Operational back offices run on platforms that were never built to be automated against, and that is exactly where the repetitive work lives.

01 PLATFORMS

Where the workers operate

Workforce and rota systems, VMS and MSP portals, Xero, Sage, payroll bureaux, CRM and ATS platforms, Outlook, WhatsApp, SharePoint.

02 NO API? STILL AUTOMATED

Portal automation

Where a platform has no integration story, the worker signs in and operates it the way a member of your team would, under its own credentials, with an audit trail of every action.

03 OWNERSHIP

Managed as a service

We run the worker, monitor the exception queue and report against the process record you agreed at the start. You get the report; your team gets the workflow back.

How the service works

From identifying the process to running the worker.

  1. 01 STAGE 01

    Identify

    We find the workflow with the most repetitive volume and the clearest rules, and map it with the people doing the job.

  2. 02 STAGE 02

    Map

    Every step, every exception and every unwritten rule, written down and agreed before the build starts.

  3. 03 STAGE 03

    Build

    The worker is built against your systems, your approvals and your tone of voice.

  4. 04 STAGE 04

    Shadow run

    It runs alongside your team while they still check the output. Nothing is trusted from day one.

  5. 05 STAGE 05

    Deploy

    It takes the work. Exceptions route to a named person from the first day it runs.

  6. 06 ONGOING

    Manage and report

    We monitor the exception queue, handle rule changes and report against the process record you agreed at the start.

One workflow at a time

We do not attempt a whole department at once. One worker is built, checked and handed over, then the next — which is how the first one goes live without waiting for a department-wide programme.
Governance

What the audit trail and the escalation path look like.

Applies to every worker built under this service
AreaHow it works
CredentialsThe worker holds its own least-privilege account in each system. It is never a shared login and never a person's own account.
Audit trailEvery read, decision and write is logged with a timestamp, the input it worked from and the rule it applied.
EscalationExceptions route to a named person with the reason attached. An exception with nowhere to go does not exist, by design.
ChangeWhen your process changes, we change the worker. Rule changes are agreed in writing before they go live.
DataUK data residency where required, processing terms and a data protection impact assessment supplied before anything runs against live data.
ExitAn agreed notice period, set out in the scope. The process map, rules and audit history are yours to keep.
Common questions

The questions that come before a scoping call.

What is AI automation?
It reads unstructured input, such as an email, a screenshot or a scanned form, decides against rules that are written down and agreed during the build, acts inside the systems already in use, and escalates anything it cannot resolve to a named person. The escalation step is what separates it from a rules tool: nothing is silently guessed and nothing is silently dropped.
What does an AI automation agency actually do?
It maps one workflow at a time, builds a worker against the rules of that specific process, and runs it as a managed service rather than handing over a platform. MSD proved the mechanism on its own back office first: Michael Szalaty, MSD's technical director, runs a staffing agency, and MSD is the supplier and the case study at once. No AI worker has run for a client yet: we scope it and discuss it, and do not claim otherwise.
What is the difference between AI automation and RPA?
RPA repeats a fixed sequence of clicks and keystrokes, and stops the moment the input changes shape: a rescanned form, a renamed field, a message that arrives as a photograph instead of a spreadsheet. AI automation reads the input regardless of its shape, decides against rules agreed during the build, and sends anything it cannot resolve to a named person rather than failing silently. The difference shows up exactly there: on the exception.
Which process should we automate first?
A scoping conversation maps every candidate workflow against how repetitive it is and how clear its rules are, with the people doing the job in the room. The workflow with the most volume and the clearest rules goes first, and the rest follow in a written, ranked recommendation.
How do you measure whether AI automation worked?
We record that one workflow's volume and errors before anything is built, then check the same record again afterwards, not against an industry average. MSD has no published measured outcome to quote yet: the mechanism first ran on Michael's own staffing agency, and MSD is the supplier and the case study at once, which is exactly why a number from it needs the strictest labelling once one exists, not the loosest. For a client engagement, the same before-and-after check runs against that client's own record.
What is the difference between agentic AI and generative AI?
Generative AI produces text or an image when it is asked; agentic AI is given a job, acts inside real systems under its own credentials, and leaves an audit trail of every read, decision and write it makes. One responds to a prompt. The other runs a process. No AI worker has run for a client yet: the mechanism above is how one is built and scoped, not a claim that one is live today.

Related pages

AI workers Defined operational tasks, agreed boundaries and named escalation. Automation and AI Find the work, map it, automate it, check it. Recruitment and staffing Where MSD's own back-office mechanism first ran. Case studies The own-operation case study, with the disclosure. Contact Book a scoping call or send the details of your workflow.

A conversation with a director

Where does the repetitive work in your operation actually sit?

That is the scoping question. We map the candidate workflows with the people who do them, and come back with a written recommendation in priority order.

sales@msdsystems.co.ukBook a scoping call

Your operation is the starting point.

Optional
The more specific you can be, the more useful the first conversation will be.
Read the privacy notice