The problem
People copy data between a form, a spreadsheet, an inbox and a CRM. The work is not hard, but it consumes hours and introduces typos.
Approvals sit in someone’s inbox. The process is waiting, not working, and nobody can see where it is stuck.
An automation exists in a no-code tool, but it fails silently when a field changes or an API returns an error.
The team has no audit trail for what the automation decided, so exceptions are handled by asking the person who built it.
What you get
Outcomes
- Hours removed from a measured workflow, not a generic productivity claim.
- Fewer manual keying errors because validation happens before the write-back.
- A visible exception queue instead of silent failures.
- A documented integration and a runbook your team can own.
- A before/after report with the baseline, the new cycle time and the remaining manual work.
Deliverables
- A workflow design with trigger, steps, systems and owners.
- The AI step: classification, extraction or a structured decision with a defined output schema.
- Integration to the tools you already use, through an API, webhook or file exchange.
- Validation rules and confidence thresholds.
- A human approval or review step for the cases that need one.
- Error handling, retries and a dead-letter path for failed items.
- Structured logs so every automated action can be traced.
- A runbook and a handover session with the team that operates it.
- A before/after measurement on the same samples used for the baseline.
Evidence
Fit and boundaries
This is a good fit when
- One workflow with at least five hours a week of repetitive work.
- A process owner who can approve the rules and the exceptions.
- At least one system with an API, export or structured file.
- A team that wants to keep its existing tools.
This is not a fit when
- No process owner, or a process that changes every week.
- A decision that requires a regulated professional judgement with no human review.
- No access to the systems involved. A screenshot is not an integration.
- An attempt to replace a whole team with an agent. This is one workflow, not a reorganisation.
How it works
- 01
Baseline the current workflow
We agree the samples, the time measurement and the definition of a successful outcome. The baseline is frozen before the build starts.
- 02
Build the happy path
Trigger, AI step, validation and write-back. The common case is automated end to end before the exceptions are added.
- 03
Design the failure path
What happens when the model is unsure, the API is down or the data is invalid? Every workflow needs a visible place for the item to land.
- 04
Measure and hand over
The same samples are run through the new workflow. You get the before/after numbers, the runbook and a session with the people who own it.
Pricing and timing
Fixed-price engagement
€3,000 to €6,000 2 to 4 weeks · EUR, VAT not applicable at present
Fixed price for one workflow, two to four weeks. A paid audit can come first and is credited against the build.
- Written scope and acceptance criteria before work starts.
- 40 to 50 percent deposit reserves the calendar slot.
- Milestone invoices and 14-day payment terms.
- A checkpoint where either side can stop with paid work settled.
Questions
Do we have to replace our current tools?
No. The workflow is built around the tools you already use. If a tool has an API, a webhook or a reliable export, it can usually be part of the automation.
Where does the human stay in the loop?
Wherever the cost of a wrong action is high, or the model is not confident. The approval step is designed into the workflow, not added after an incident.
What happens when the automation fails?
Every item has a visible failure path: a review queue, a retry, or an alert. Silent failure is the thing that turns an automation into a liability.
How is the time saving measured?
On the same samples used for the baseline: time per item, manual touches, cycle time and exception rate. The report shows the range, not a single best case.
Can you maintain it afterwards?
Yes, through the monthly automation retainer. It covers monitoring, exception review and a small improvement block. It is optional; the handover is designed so your team can own the workflow.
Page last updated: October 2026