Skip to content
Ioannis Ouzounis

Process automation

Manual work is a cost you're already paying.

Every team has tasks that are slow, repetitive and one typo away from an expensive mistake. I find them, automate them with tools your team will actually use, and measure the hours and errors you get back.

83–98% less time per task across my published projects

Illustrative
One daily task · one year
Hours that repeat
230 h
Errors that travel
2 a month
Knowledge in one head
1 person
Experts on clerical duty
5 h/wk
Budget line
none

The hidden bill of manual work

It never appears as a line in the budget. It shows up as overtime, late deliveries and the colleague who “just knows how it works”.
  • Hours that repeat

    A one-hour task done every working day adds up to about 230 hours a year, for one person. Most teams have several.

  • Errors that travel

    A broken placeholder or a wrong figure is cheap to catch before delivery and expensive after it: rework, rush fees, credits, and a client who starts double-checking your work.

  • Knowledge in one head

    When only one person knows the process, every vacation and every resignation puts a deadline at risk.

  • Experts on clerical duty

    Your best linguists, project managers and analysts spend hours copying, reformatting and re-checking instead of doing the work you hired them for.

Manual work fails quietly

A step that works 95% of the time sounds safe. Repeat it twelve times, say once per language in a release, and the chance that at least one goes wrong is close to one in two.

People get tired, skip a step under deadline or go on vacation with the know-how in their head. Automation runs the same way every time, and when it can't, it tells you.

IllustrativeTwelve runs of a step that works 95% of the time
One run goes right
0.95
All twelve go right
0.54
At least one goes wrong
0.46

1 − 0.95¹² ≈ 0.46

Is your task a candidate?

Three or more checks and it probably is.
What is true of your task?

0 of 6 checked

The work I remove most often

  • Messy client data

    Dates in three formats, merged cells, inconsistent columns: cleaned and converted into files your systems accept.

  • File preparation

    Glossaries, spreadsheets and exports turned into the exact format your tools need.

  • Quality checks

    Missing translations, broken placeholders, wrong terminology, number and date formats: checked in seconds, every time, and flagged where people already work.

  • Recurring reports

    Daily and weekly reports assembled from several sources with one click.

  • Project setup and handoffs

    Folders, files, tasks and notifications created from a single form.

  • Copy-paste between systems

    TMS, spreadsheets, CRM and invoicing kept in sync without retyping.

  • New software, when it's the right answer

    Internal tools and platforms in React/Next.js or Laravel, designed to grow with you.

Find. Build. Measure. Hand over.

  • AI agents
  • Human checkpoint
  1. 01 Find

    What happens
    I watch how the work is really done, not how the process document says it's done. I time each step and note where errors creep in, then we rank tasks by yearly cost, risk and how rule-based they are. AI agents map the processes from sample files in parallel, so this stays short.
    You get
    A ranked shortlist with estimated hours, risk and payback for each task.
  2. 02 Build

    What happens
    Agents explore competing designs and draft code and tests in parallel. I choose, build and review the smallest thing that removes the task: a one-click button in Excel or Word, a small desktop tool, a script in your pipeline, an integration between systems. Validation is built in, so problems are flagged before anything goes out.
    You get
    A working tool with validation built in and a manual fallback.
  3. 03 Measure

    What happens
    Before launch, we record the baseline: minutes per run, runs per month, errors per month. After a few weeks of real use, we measure again.
    You get
    Numbers, not adjectives.
  4. 04 Hand over

    What happens
    I document the tool and show the people who use it how it works. If an automation needs a developer on call, it isn't finished.
    You get
    Documentation, a short training session and tools your team runs without me.

Reviewed by IoannisIO

AI where it helps. Rules where you need certainty.

Many of the biggest savings come from plain, predictable automation with no AI at all. I use AI agents to understand your process faster, compare designs, and write and test the code.

Where a task needs no AI, the tool you receive is an ordinary, auditable script or app that runs in your environment, with no per-use AI fees. Where it does, AI only gets jobs whose output can be checked, with confidence thresholds and a human review queue.

Good jobs for AI

  • Sorting messy inputs
  • Pulling fields out of unstructured documents
  • Drafting summaries for a person to review
  • Suggesting terminology matches

Better left to rules

  • Validations
  • Calculations
  • File conversions
  • Anything where one mistake is expensive

Savings measured, not promised

Before I build anything, we write down the baseline. After go-live, we measure the same things again.

Measured before and after

  1. Minutes per run
  2. Runs per month
  3. People involved
  4. Errors per month
  5. Hours saved per year and their value
  6. Payback period

Go or no-go

The baseline also answers the first question: is this task worth automating at all?

Real before-and-after numbers

Time per task, before and after automation
TaskBeforeAfter
Glossary preparation1–2 h5–10 min
Translation file validation1.5 h/day2 min/day
QA report generation1 h/day1 min/day
Project setupPlatform built as part of a development team30 min2 min

What is one task costing you?

Minutes, frequency, people, hourly cost and mistakes. About a minute, in your browser, nothing stored.

Built with what fits, not what's fashionable

Tools follow the problem. If the best answer is a spreadsheet formula, that's what you get.
  • Python
  • VBA and custom Office ribbons
  • PowerShell
  • JavaScript and TypeScript
  • React and Next.js
  • PHP and Laravel
  • REST and GraphQL APIs
  • Database design
  • CAT tool and TMS integrations
  • CI/CD for localization

Questions people ask

Our process is a mess. Can it still be automated?

Usually, and that's often where the biggest savings are. Part of the work is making the process consistent enough to automate, without forcing your team to change how they think.

Will automation replace my team?

It replaces the copy-paste, not the people. The point is to free skilled people from clerical work and to stop errors before they reach a client.

Will my team need to learn new software?

Rarely. I put automation where people already work: a button in Excel or Word, a simple window, or a step that runs on its own.

Is my data safe if AI is involved?

Which data the agents may see is agreed up front, as part of the brief, and anything off-limits stays off-limits. More on how I handle your data.

Who maintains it afterward?

Many tools need no maintenance at all, and the handover includes documentation your team can follow without me. Code ownership and any support afterward are agreed for each project.

Is a 20-minute task worth automating?

Twenty minutes every working day is more than 75 hours a year, per person. Often, yes. The estimator will tell you.

Which task would you delete first?

Tell me about it. If it isn't worth automating, I'll say so.

I read and answer every message myself.