One repeatable job
Systems, actions, exclusions and output measures are agreed before build.
Human-controlled AI agents for UK service businesses
A human-controlled AI agent for document chasing.
We assess repeatable document-chasing work and, where it is safe and practical, scope an AI-assisted, human-controlled workflow that sorts replies, attempts agreed filing and tracker updates, flags exceptions, and prepares weekly progress summaries.
For document-heavy service teams that spend too much time chasing clients, suppliers, applicants, landlords, tenants, brokers, contractors, or internal teams for the same missing information.
Human approval at launchSpecified action recordsExisting tools firstExceptions go to humansNo regulated advice
Simulation · sample data · not proof
Front-end simulation with sample data; not a customer case study, live integration, operating test or proof of delivered results.
Mach Lilies' primary implementation focus is Quiet Quarter for accountancy and bookkeeping practices. We also consider selected document-heavy workflows in other UK service businesses where the process is repeatable, safe, and measurable.
Most document chasing breaks in the same places.
Whether the workflow involves onboarding, renewals, compliance packs, client evidence, certificates, forms, signatures, or missing attachments, the pattern is usually the same: someone has to check what is missing, send another reminder, find the reply, save the attachment, update the tracker, and flag anything unusual.
- Replies arrive across messy inbox threads.
- Attachments are saved inconsistently.
- Trackers fall behind.
- Different staff chase in different tones.
- Exceptions are spotted late.
- Managers only get visibility when a deadline is close.
- The team repeats the same follow-up loops every week.
What the helper actually does
A working cycle, in order:
- Reads the agreed tracker — checks who is missing what from a spreadsheet, task board, export, or agreed workflow source.
- Drafts or sends approved chasers — uses approved templates and starts in human approval mode.
- Classifies replies — spots received documents, questions, wrong attachments, unclear responses, and exceptions.
- Files replies and attachments — puts replies and files into agreed folders or locations where technically feasible.
- Updates the tracker — marks chased, received, pending, exception, or needs-human-review according to agreed rules.
- Flags exceptions — escalates anything sensitive, unclear, angry, unusual, risky, or outside scope.
- Sends a weekly summary — gives your team visibility over progress, blockers, and next actions.
Useful where document chasing is repeated, measurable, and painful.
Estate and letting agencies
Tenant onboarding, landlord certificates, compliance documents, maintenance evidence, renewal paperwork.
Insurance and mortgage brokers
Renewal packs, client evidence, application documents, missing forms, policy or lender information requests.
Recruitment and staffing firms
Candidate documents, right-to-work checks, references, onboarding packs, timesheet evidence, compliance records.
Professional services
Client onboarding documents, engagement letters, supporting evidence, recurring information requests, signed forms.
Trades and construction services
Supplier documents, contractor evidence, certificates, job packs, compliance checks, missing attachments.
Healthcare and care administration
Operational admin workflows, forms, internal follow-ups, appointment paperwork, non-clinical document chasing.
Mach Lilies supports operational admin workflows only. We do not provide legal, financial, healthcare, tax, accounting, clinical, or regulated advice.
From scattered chasing to a controlled weekly workflow.
Before
The manual chase
- Staff check what is missing manually.
- Follow-ups are copied and pasted.
- Replies land in multiple inboxes.
- Attachments are saved wherever someone remembers.
- Trackers are updated late.
- Exceptions depend on whoever spots them first.
After a monitored workflow
A controlled weekly workflow
- The agreed tracker drives the chase list.
- Approved templates keep the tone consistent.
- Replies are classified.
- Files are saved where technically feasible.
- Trackers are updated against agreed rules.
- Exceptions are escalated to humans.
- Weekly summaries show progress and blockers.
Start with one repeatable workflow.
This is an enquiry pathway, not a second published package. The chase-loop assessment checks whether one document-chasing workflow is safe, measurable, and worth scoping before any wider work is considered.
Agree the first workflow
Define one repeatable document-chasing process and its evidence before build.
What is agreed
- Workflow definition and output measures agreed in writing
- Inbox routes, trackers and document locations fixed in the signed scope
- Approved chase and escalation templates
- Human approval mode
- Reply classification
- Filing where technically feasible
- Tracker updates against agreed rules
- Exception flagging
- Weekly summary
Exact quantities, support and commercial terms are proposal-specific and agreed in writing before anything starts.
Widen only after stability
Additional workflows are considered only after the preceding workflow is stable.
What is agreed
- Additional workflows only when separately included in the signed scope
- Each inbox, tracker source and document location agreed explicitly
- Proposal-specific rollout dates and service levels
- Approved templates and exception routes for each workflow
- Staff handover
- Review cadence agreed in the signed scope
- Output review at the checkpoint agreed for the scoped engagement
No generic workflow, inbox, contact or template maximum is published on this route.
Quiet Quarter is the only published service and tariff, for accountancy and bookkeeping practices. Other document-chasing requests are assessed by enquiry and must not be treated as that offer or as a second package.
Where this works best
The same workflow suits some teams far better than others. Here is where it fits — and where it does not.
A strong fit
- Your team repeats the same document-chasing workflow every week or month.
- There is an agreed list, tracker, spreadsheet, board, or export showing what is missing.
- Replies usually arrive by email or another trackable route.
- There are standard follow-up messages or templates.
- A human can approve drafts and exceptions at launch.
- The workflow is operational admin, not professional judgement.
- Success can be measured through workflow outputs.
Not a fit (yet)
- You want AI to make regulated, legal, financial, clinical, tax, or accounting judgements.
- There is no clear tracker, owner, or repeatable process.
- The workflow changes completely every time.
- You need heavy bespoke integrations before proving the process.
- You want autonomous client sending from day one.
- You expect every client to reply.
- You want to replace staff judgement rather than reduce repetitive admin.
A controlled way to decide whether to start.
Approval at launch
Client-facing messages begin held for a person; unclear work has a named route.
Test the outputs
Review the available action record and the agreed outputs before widening the workflow.
Before go-live, we agree measurable workflow outputs in writing: chases prepared or sent, replies classified, attachments filed where technically feasible, tracker updates completed, exceptions flagged, and weekly summaries produced.
The promise depends on agreed scope, timely access, usable source data, and your team using the approval process as agreed.
Quick workflow fit check
A quick self-check before a fit call — no form, no sign-up.
- Is the chase repeated weekly or monthly?
- Is there a tracker or list of missing items?
- Do replies arrive somewhere trackable?
- Can a human approve drafts at launch?
- Can success be measured through workflow outputs?
- Is the workflow operational admin rather than regulated advice?
If you answered yes to most of these, book an ai workflow assessment.
Have a repeatable document-chasing workflow?
Book an AI workflow assessment. We will tell you whether the workflow is safe, practical, and worth automating — and what should stay human-led.