How we decide, build and look after a system.

We start with the work people need to do. Then we decide what to build, what to test and who stays responsible.

An alert sent is not a task completed.

The question is whether the operation moved forward. That shapes both the system and the way we assess it.

Understand the process.

Who does the work, what starts it and what counts as done?

We map the actual steps, exceptions and systems with the people involved. A process owner and an observable starting point come first.

The deliverable

A process map and acceptance criteria.

Choose the intervention.

Where does AI help, and where is a rule enough?

We separate interpretation from execution. We define permissions, the information required and the decisions that stay with a person.

The deliverable

A solution design with explicit boundaries.

Build and test with the team.

What happens when the expected input does not arrive?

We connect the required tools and test normal work, missing information and exceptions. We review what actually happens, including how to recover from a failure.

The deliverable

An integrated system and recorded test results.

Put it to work. Keep reviewing.

Is the system being used, and is the process moving?

We agree on the rollout, responsibilities and follow-up. We review usage and incidents within the contracted scope, then prioritize changes.

The deliverable

A rollout and follow-up plan with named owners.

Four questions before we automate an action

  1. What event starts the work?
  2. Who owns the next step?
  3. What happens when information is missing?
  4. What evidence tells us it is complete?

The exact controls depend on the system and the agreed scope. We document them before treating them as a promise.

A checklist with ticked boxes and a pencil on a desk.
Illustrative setting

Controls that exist, with their scope

We do not promise universal controls. These are the ones in production today, and where.

  • What is not in the catalog does not get classified: the system asks.

    In Grupo ILT’s system, a message that does not match the catalog changes no status. A person is asked.

  • Every change carries an author and a time; nothing is rewritten.

    In two of our production systems, corrections are appended to the log. Delete and rewrite are revoked for every role, ours included.

  • Before a client is notified, there is time to cancel.

    In one of those systems, every client notice waits ten minutes before it goes out. If something is wrong, it stops.

  • Our own brake, on record.

    On August 4, 2026, an outreach engine we built for a client measured a 5.8% bounce rate against a 4% limit. It stopped itself, wrote down why, and sent nothing for six days, until a person fixed the cause and restarted it.

Records H-083 · H-115

What we will never tell you

AI has no shortage of promises. These are the ones you will not hear from us:

  • A line we don’t use: “You will save this much.”

    Not before we measure your starting point. Without a baseline, any saving is an invention.

  • A line we don’t use: “It runs by itself.”

    It runs with rules, with an owner on your side and someone on ours who looks after it.

  • A line we don’t use: “It is one hundred percent secure.”

    We tell you which controls it has, what we tested and what we did not. Nobody serious promises zero risk.

  • A line we don’t use: “Other clients achieved this figure.”

    We do not publish any client’s data without written permission, and we will not publish yours.

  • A line we don’t use: “We are faster than.”

    Comparing times that were not measured the same way is comparing apples and oranges. We give you a calendar with dates and with what depends on whom.

  • A line we don’t use: “We are the only ones.”

    We tell you what we are good for and what we are not.

If you ever read one of these lines with our name on it, tell us: it is a mistake and we will correct it.

What an executive asks before starting

Where do I start applying AI in my company?

With one bounded process that has an owner, not with the technology. The first moment with Teckel is defining that process: mapping the current work, deciding where to intervene and agreeing on how to check it improved.

How much does it cost?

Each moment is quoted by scope, after we understand the process. We do not price an implementation without knowing the work, the tools and the data involved.

Which tasks are worth automating?

The repetitive, high-volume ones with clear rules: typing the same thing into two systems, following up, notifying, classifying. Decisions that need judgment stay with a person, and the system asks.

Is it safe to use AI with my customers’ data?

Security depends on the design, the providers, the contracts and the type of data. Before implementing we assess controls, responsibilities and the legal requirements that apply. We do not assume compliance.

How soon do I see results?

It depends on scope, integrations and adoption. Before promising a timeline we define deliverables and acceptance criteria. A previous case does not guarantee the timing of another.

What is the difference between an agency and an applied AI firm?

An agency usually delivers campaigns. Teckel designs and implements systems inside the operation and reviews their use against agreed criteria. We do not take adoption for granted: we instrument it and check it.

What needs to work better?

Bring a process, a pilot or a question. You talk directly with the people who will build it.

Book a conversation

Opens Calendly in a new tab.

george@teckel-ai.com
Prepare your call with AVA

AVA asks a few questions so you arrive with your case in order.