Skip to content
02 · Service

SerSan · Workflow Automation

Workflow AutomationAutomate the work nobody should still do by hand.

Automate the work your team should not still be doing manually. We connect the systems you already use, build the small internal tools that are missing, and use AI only where judgement is genuinely required.

Days to weeks per workflow · fixed scope

What goes wrong with automation

In most businesses, the work worth automating is still done by hand.

Not because nobody thought of it — the obvious tools stop right where the work gets hard. So somebody re-types the same data into three systems, chases approvals by email, and rebuilds the same report every Monday. Or the opposite happened: years of Zaps and scenarios nobody can map, quietly dropping work when a field changes.

Typical build includes

What a well-built automation actually contains.

Most automations use three or four of these, not all five. We agree which before we start: one workflow should never inherit the process of a platform build.

  • 01

    Rule-based first, where software is enough

    If the step is deterministic, we write deterministic code: no model, no token cost, no uncertainty. Lightweight workflow code in the stack you already have, not a heavyweight platform. Most of a good automation is exactly this.

  • 02

    AI-assisted where judgement is required

    Classification, extraction, summarisation, reasoning over messy input: the steps rules can't do. Model calls are wrapped in retry, cost budgeting and a fallback to a human queue when confidence is low.

  • 03

    Human-controlled where mistakes matter

    Not every step needs review. We map which decisions can run autonomously and which need a human checkbox, by impact, by reversibility, by regulator.

  • 04

    Work never silently disappears

    When an automation fails, nothing vanishes. Every step has a defined failure behaviour, and anything that can't complete is captured, surfaced and safely retried — a dead-letter queue your team can inspect, replay or escalate.

  • 05

    Yours to run, and the costs are visible

    You own the code. We document how it works, train whoever will look after it, and show running cost per workflow and per step so the monthly invoice never surprises you. No internal engineering team required.

Where this lands

Three kinds of manual work we're usually called in to remove.

  • 01 · Use case

    Repetitive work eating your team's time

    Lead intake and qualification, CRM updates, onboarding steps, internal approvals, scheduling and notifications, the report someone rebuilds every week. We start with one workflow, prove the saving, then expand.

  • 02 · Use case

    Tools that don't talk to each other

    Your CRM, your inbox, your spreadsheets and your billing tool each hold part of the truth, and someone re-types between them. We connect them properly, or build the small internal tool that should sit in the middle.

  • 03 · Use case

    Documents, emails and human judgement

    Invoices, contracts, applications or inbound requests that need reading, routing or enriching. The model reads, rules route, a person approves anything that matters — and every step leaves an audit trail.

Deliverables

What you actually hold at the end.

Concrete artefacts, handed over at the end. The automation code, a map of what runs when, and what each run costs — yours to keep, extend or hand on.

  • Automation code in your repository, owned by you
  • Per-step retry / rollback / dead-letter routing
  • Cost dashboard + budget alerts
  • Human-review queue with audit log
  • Workflow map, documentation + owner training
  • Optional: 30-day stabilisation window post-launch

Selected work

Projects behind this service.

3 projects that exemplify this service — 1 built by SerSan, 2 from prior senior delivery.

Common questions

Answers from real projects.

Do you replace Zapier or work alongside it?

Both, depending on volume and reliability requirements. Low-volume, low-stakes flows can stay in Zapier. Anything tied to revenue, compliance or customers moves to owned code — and we migrate workflow-by-workflow, never big-bang.

How small can a first automation be?

Small. A useful automation doesn't need to become a six-month transformation programme. Many start with one workflow, prove the saving, then expand. Even a screen recording of how you do it today is enough to start.

Can you work with the tools we already use?

Yes, wherever it's technically sensible. Most of this work is connecting things you already pay for — your CRM, your accounting software, your website, your spreadsheets — through their APIs. Where a tool has no way in, we say so early and design around it instead of pretending.

What if AI isn't the right answer for our process?

Then we say so and build the ordinary version. Most workflows are fixed with rules, integrations and a small internal tool, with no model involved at all. AI earns its place on the steps that genuinely need judgement, and nowhere else.

Start with the problem

Name the job that still runs by hand.

Two or three sentences, or a screen recording of the job as you do it today. A founder reads it and comes back with the smallest version worth building first.