AI delivery · Workflow automation

One repetitive workflow, automated with approvals, monitoring and an owner

If one repetitive workflow moves information between documents, email, a CRM and internal tools, you can commission me to implement it as a bounded delivery. I start by writing the input/output contract and measuring today's baseline, then build the workflow with human review points, error handling and duplicate-update protection, and hand it to a named owner with a runbook.

This is a good fit if…

  • You have one specific, repetitive workflow (intake, triage, enrichment, record updates) with clear inputs and a checkable output.
  • The work crosses systems, such as a shared inbox, a CRM, a ticketing tool and a spreadsheet, and people currently copy data between them.
  • You need approvals for some steps and cannot accept an automation that silently creates duplicate or wrong records.
  • There is someone in operations or product who will own the workflow after launch.

Look elsewhere if…

  • You want staff to use AI tools better in their own work rather than an implemented system. Use AI enablement for business teams instead.
  • You have a list of ideas and no chosen workflow yet. Start with the AI readiness and implementation roadmap.
  • The workflow depends on data nobody can locate or access. Start with AI data readiness.
  • You need the automation to make regulated or high-stakes decisions without a person approving them. I do not build that.

What you get

Workflow implementation and handover

  • A written input/output contract for the workflow, agreed with the person who owns it.
  • A measured baseline of today's volume, handling time and error rate, so the change can be judged.
  • A working workflow in your environment with approval points, retries that do not duplicate updates, and an exception queue.
  • Monitoring that tells the owner when the workflow is failing, slowing or producing more exceptions than usual.
  • A runbook and handover session so your team can operate and change it.

How it runs

  1. 01

    Brief and fit check

    You describe the workflow, the systems involved and who owns it. I reply with questions and an honest view on whether it is a good candidate before anything is signed.

  2. 02

    Starting assessment

    I map the workflow with the people who do it, write the input/output contract, sample real cases and measure the baseline. If it is not automatable as described, we rescope or stop here.

  3. 03

    Build with review points

    I implement the workflow against your systems' APIs, with permissions scoped to the task, human approval where agreed, idempotent writes and an exception queue.

  4. 04

    Parallel run and acceptance

    The workflow runs alongside the manual process on real cases until the agreed acceptance measures are met and the owner signs off.

  5. 05

    Handover

    Runbook, monitoring, credentials and change process transferred to the named owner, with a walkthrough on a live exception.

What needs to be in place

  • One chosen workflow and a named business owner who can make decisions about it.
  • API or service access to the systems involved, provisioned through your normal process.
  • A sample of real past cases, including awkward ones, handled under your data policy.
  • Agreement on which steps need human approval before anything is written or sent.

Not included

  • Automating several unrelated workflows under one scope. Each is scoped and accepted separately.
  • Decisions that affect people's employment, credit, health or legal position without a human decision-maker.
  • Replacing or re-platforming your CRM, ERP or ticketing system.
  • Out-of-hours support after handover unless agreed separately in writing.

Which workflow-automation route fits your problem

Start here whenSignature outputMain acceptance measure
Workflow automation (this page) One repetitive workflow crosses several systems and needs approvalsImplemented workflow, exception queue, runbook and handoverParallel-run accuracy and exception rate against the baseline
Document and reporting workflows The work is mainly extracting from documents and preparing reviewed reportsDocument pipeline with source traceability and an exception queueField-level accuracy on an evaluated sample set, with every value traceable
Voice AI workflows The workflow happens on calls and needs escalation and consent handling (taken on selectively)Latency budget, consent dependencies, evaluation calls and handoff designLatency and escalation behaviour on scripted evaluation calls
AI data readiness The demo works but real data is fragmented and access is unclearSource inventory, permission map, freshness checks and ingestion planEvery source has an owner, an access route and a freshness check

The situation

Most operations teams have one workflow that everyone knows is wasteful. A request arrives by email or form, someone reads it, looks up the customer in the CRM, checks a spreadsheet, updates a ticket and sends a reply. It happens dozens or hundreds of times a week, the inputs are fairly predictable, and the person doing it would rather be doing something else.

AI makes this kind of workflow newly automatable, because the messy part (reading free text, classifying a request, extracting the fields that matter) is now something a model can do reasonably well. The difficulty is everything around the model: permissions across several systems, what happens when a step fails halfway, how to stop a retry from updating a record twice, and who notices when the workflow quietly starts getting things wrong.

This page is for commissioning that workflow as a bounded delivery: one workflow, implemented in your environment, accepted against agreed measures and handed to an owner.

What the work involves

I start with the people who do the work today, not with a model. Together we write the input/output contract: what arrives, in what forms, what the finished output looks like, which systems are read and written, and which decisions need a person. Then I sample real past cases, including the awkward ones, and measure the baseline: volume, handling time, error and rework rate.

The build follows a few rules that come from running AI systems in production:

  • Scoped credentials per workflow. The automation gets the minimum access needed for its steps, not a shared admin account.
  • Approval before consequential writes. Anything that sends an external message, changes money or changes a customer-facing record can be held for review. The model proposes; the workflow checks; a person approves where agreed.
  • Idempotent writes and safe retries. Each case carries a stable key, so a timeout and retry cannot create duplicate updates.
  • An exception queue, not silent failure. Cases the workflow is unsure about, or that fail validation, go to a named queue with the reason attached.
  • Monitoring the owner can read. Volume, exception rate, approval backlog and failure alerts, in a tool your team already watches.

Where agents need to call several internal tools, I may expose them through a controlled tool layer such as the MCP integration pattern, with each tool scoped and logged.

The signature deliverable

You receive the workflow implementation and handover: working code or configuration in your environment, the input/output contract, the exception queue, monitoring, and a runbook. Illustrative example of an input/output contract extract:

FieldSourceRuleOn failure
Customer IDCRM lookup by sender domainExactly one match requiredException queue: “ambiguous customer”
Request typeModel classification of email bodyOne of seven agreed types; confidence threshold agreed with ownerBelow threshold: human triage
Ticket updateTicketing APIIdempotency key = message IDRetry up to agreed limit, then alert
Customer replyDrafted by modelHeld for approval for refund and cancellation typesNever sent without approval

Illustrative example showing the format, not a client deliverable.

How acceptance is judged

Acceptance criteria are agreed in writing after the starting assessment, before the build. Typically: the workflow runs in parallel with the manual process on real cases; outputs are compared field by field; the exception rate and error rate are measured against the baseline; and retry and failure tests show no duplicate writes. The business owner signs off. If the measures are not met, the gap and its cause are written up rather than smoothed over.

Ownership and handover

The workflow is yours from the first commit: your repositories, your accounts, your credentials. At handover, the named owner receives the runbook (how to pause the workflow, clear the exception queue, rotate credentials and change approval rules), and we walk through a live exception together. I do the work personally; any specialist help is disclosed and approved by you first.

Choosing the narrower route

If your workflow is mostly reading documents and preparing reviewed reports, the document processing and reviewed reporting workflows page is more specific. If it happens on calls, see voice AI workflow engineering, which is taken on selectively. If the real blocker is that data is scattered and access is unclear, start with AI data readiness. If you want staff to change how they work with AI tools rather than commission a system, business AI enablement is the better fit.

Questions buyers ask

How do you stop the automation creating duplicate records or sending things twice?

Every write carries an idempotency key derived from the source case, so a retry after a timeout updates the same record instead of creating another. Steps that send messages or change customer-facing records can sit behind an approval. During the parallel run we specifically test retries, partial failures and replays before acceptance.

What if the starting assessment shows the workflow should not be automated?

Then that is the finding, and we stop or rescope before the build. Some workflows have too many judgement calls, too little volume or inputs that are not stable enough. You still get the input/output contract and baseline, which are useful whatever you decide.

Do you use a particular automation platform?

No. I build in whatever your team can own: code in your repositories, your existing integration platform, or a small service with clear interfaces. The choice depends on who will maintain it. I avoid introducing a new platform your team has no one to operate.

Who owns the workflow after handover?

A named person on your side, agreed at the start. They hold the runbook, the monitoring alerts and the authority to change approval rules. If you want help after handover, ongoing maintenance is a separate, bounded agreement.

What data do you need to see?

A representative sample of real cases, ideally including failures and edge cases, accessed inside your environment under your data policy. I do not copy data out. Where cases contain personal data, we agree minimisation or redaction before I start.

Bring a repetitive workflow

A short, non-confidential description is enough to start. I read every brief personally and reply within two business days, including when the answer is that I am not the right fit.

Step 1 of 2 · The basics