Automating a broken process does not fix it. It speeds it up.
Before a single line of automation is written, we document the process as it is actually executed: the exceptions, the manual steps and the criteria that exist only in one person's head.
02 / Why automating before mapping the process gets expensive
Why automating before mapping the process gets expensive
- 01
The documented process is not the real one
The diagram in the quality manual describes the usual path. In practice someone skips two steps once an order passes a certain value, and that exception is written down nowhere until it gets automated away.
- 02
Exceptions dropped for simplicity
Automating only the general case and leaving exceptions for later means the automated process fails precisely on the cases that consumed the most time from the person who used to handle them.
- 03
No human review point
A process automated end to end, with no point where a person can check before the effect becomes irreversible, converts a two-minute mistake into a production incident with a customer attached to it.
- 04
Activity measured instead of outcome
Counting how many times the automation ran says nothing about whether the result was correct. Without an outcome metric, an automated process can fail systematically for months and look healthy on the dashboard.
- 05
Tacit knowledge never captured
The judgement the most experienced person applies to an odd case is rarely written down. Automating without extracting that judgement first leaves the process working below the standard it had when that person ran it.
03 / How we build it
How we build it
- 01
process archaeology
Observation and interviews to document the process as executed, not as written
- 02
exception mapping
Catalogue of the cases that deviate, and the criteria applied to each
- 03
flow design
Which steps the system automates and where a person stays in the loop
- 04
implementation
Built on your existing systems, without replacing the system of record
- 05
review point
Held for human approval on the highest-impact or lowest-confidence steps
- 06
outcome measurement
Automated result compared against what the manual process produced
- 07
tuning
Thresholds and exceptions revised from the real cases reaching the review point
04 / Integrations
Integrations
SAP
Salesforce
Dynamics 365
Microsoft 365
Google Workspace
PostgreSQL
Zapier
n8n
05 / Guarantees
Guarantees
- 1 week
- Spent documenting the real process before automating
- Signed off
- Exception catalogue, agreed with the team before first deployment
- Outcome
- Success metricNot execution volume
- Human
- Review point on the highest-impact steps
06 / Compliance
Compliance
Record of every automated decision and every human intervention
EU AI Act, traceability obligations, applicable from 2 August 2026
Human review point on decisions with a significant effect on individuals
EU AI Act art. 14, human oversight
Process data hosted inside the European Union
GDPR, chapter V
Technical documentation maintained in a form a market surveillance authority can inspect
EU AI Act art. 99(4), penalties of up to EUR 15M or 3 % of global annual turnover
07 / Process
Process
- 1-2
Process archaeology
Direct observation and interviews with the people who run the process today, to document the real path and its exceptions rather than the official diagram.
- 2-3
Exception and criteria catalogue
Every exception identified, and the criterion a person applies to resolve it, written down as the basis for the automation rules.
- 3-4
Automated flow design
Which steps the system automates, which require human approval, and which outcome metric will decide whether the process is working.
- 4-7
Implementation and pilot
Built on your existing systems and run in parallel with the manual process until the outcome metric holds, not until the demo looks convincing.
- ongoing
Operation
Thresholds and exceptions tuned from the real cases that reach the human review point.
08 / Frequently asked questions
Frequently asked questions
- Why spend a week just documenting the process?
- Because the written process almost never matches the executed one. Automating the wrong version means redoing the work when the first uncontemplated exception appears, which usually happens in the first week of production.
- What about exceptions only one person knows how to handle?
- They are captured during process archaeology, through interviews and direct observation, and turned into explicit rules before anything is automated.
- Does automation replace the people running the process?
- Not on the highest-impact or lowest-confidence steps, where the design keeps an explicit human review point before the effect becomes irreversible.
- How do you measure whether the automation works?
- By comparing the automated outcome against what the manual process produced, not by counting executions. A process that runs often and produces the wrong result is not working.
- Which tools do you automate with?
- The tool follows the process and the existing systems: a custom integration, an orchestrator such as n8n, or a combination. We do not start from a fixed template and bend the process to fit it.
- Which processes are poor candidates for AI automation?
- Those depending on judgement the organisation cannot or will not make explicit. If the criterion cannot be written as a rule or delegated to a human review point, automating it only moves the problem.
- How long until the first process runs in production?
- Four to seven weeks from the start of process archaeology to the parallel pilot, depending on how many exceptions the process carries and how many systems it touches.
09 / Related services
Related services
Wiring every agent to every system does not scale.Each direct integration is a promise of permanent maintenance. We build custom MCP servers that centralise how your agents reach your systems, with a tool contract and a schema validated on every call.
An agent with access to everything is a risk, not a gain.An agent with unrestricted access and no audit log is not a productivity tool; it is a security incident with no date on it yet. We build internal agents with role-scoped permissions and a log that survives a security review.
Services
10
Tell us which process to fix
Describe the process and the systems behind it. You get back a technical proposal — architecture, timeline and acceptance criteria — not a service catalogue.
Email hola@teledi.ai