We do not work for a finished implementation. We work for a Success Case.

The difference is not semantic. Most projects die in the mirage of implementation: the system ends up configured, the vendor gets paid, and the value never arrives. Nautilius is the methodology that prevents that ending; Zherpa implements it.

Our commitment is measured with the Master Formula of Value: CE = IE × AS × KPIHQ × t. Effective Implementation, Sustained Adoption, High-Quality KPIs and Time. It is a multiplication, not a sum: if one variable is zero, the result is zero, regardless of how good the other three are. That is why victory and its indicators are defined before starting, not at closing.

Behind this there is no sales pitch; there is a published body of knowledge: the Nautilius Theory of Value and the book Humano al Mando, by Miguel Ángel Arce. The same criterion that governs your project is written down and available. You can read it before hiring us.

To that we add 30+ years of business consulting, 1,400+ projects and 22 years inside the Zoho ecosystem, with commercial independence: we are not an official partner and we have no license quota to defend. The technical recommendation is not conditioned by a vendor's margin.

Humano al Mando begins in Phase 0 with purpose, evidence and responsibility. When the project is authorized, it materializes into limits, COA, RID, RDA, observability and measurement.

Our success is measured by the productive autonomy that remains with the client, not by the dependence we generate.

Because before asking whether you are ready for an agent, we ask whether you need one.

Not every operational problem is solved with AI. Many are solved with a required field, an assignment rule, a change of sequence in the process, or a conversation nobody has had. If that is the answer, we tell you, and it costs you less. An agent that solves what a workflow already solved is cost without difference.

Then comes the second question: if the agent is indeed the solution, do the conditions to take advantage of it exist today? The risk is not only building the agent badly; it is also onboarding it into a company that cannot yet sustain it. Phase 0 verifies value, process, data, owner, adoption and governance.

The verdict is never a flat "no". The possible resolutions never leave you without a route:

1. No agent is needed, with the simplest alternative identified.
2. Ready to move forward.
3. Move forward with conditions, with the conditions named and their owner.
4. Prepare the foundation first, with the order and scope of that preparation.
5. Not yet, with the specific reason and what is missing for it to become a yes.

The difference between "no" and "not yet, and this is what is missing" is the difference between a closed door and a map.

There is also an effect our clients report: preparing the organization for an agent improves the company before the agent exists. An agent demands what the operation had postponed for years. A defined process. One owner per decision. Complete fields and reliable sources. Explicit criteria instead of tacit agreements. Discipline of record-keeping. When those conditions are met, the team decides better with the information it already had, without having deployed anything yet.

That benefit is not a consolation prize. It is a direct consequence of the agent's demands, and it arrives before the agent does.

That is why Phase 0 is mandatory. It does not filter clients: it orders the investment, and sometimes it prevents it.

USD 370 (MXN $6,264), VAT included. Verdict delivered within 3 business days after the diagnostic session.

It covers approximately three hours of specialized work by the Zherpa team across preparation, session, analysis and verdict. It is the smallest fraction of the budget, it resolves in under a week, and it is the one that decides whether the rest is worth spending.

You receive an executive verdict with six pieces:

• Problem validation. If what you need is solved with a process change, a rule or a configuration, we say so right there.
• Readiness Traffic Light. The honest reading of where your organization stands today on value, process, data, owner, adoption and governance.
• Detected gaps, specific and prioritized, each with its owner and its closing condition.
• Conditions to move forward. What has to be true before investing.
• Recommended profile or cell for your case, with its maturity state in plain sight.
• Executive resolution. One of the five possible resolutions.

It is mandatory for all clients, non-refundable and not deducted from the project. The reason is independence: the service is delivered in full even when the verdict recommends not moving forward or not using an agent. If Phase 0 were credited against the project, we would have an incentive to approve everyone. We do not have it, and that is why the verdict is worth something.

Three days and a fraction of the budget to know whether the rest deserves to be invested. A "not yet" is not a bad outcome: it is the outcome that saves you an entire misinvested project.

Almost nobody arrives ready. That is not the problem. The problem is deploying without knowing it.

An agent on top of broken data does not create intelligence. It automates errors and makes them faster, more numerous and harder to trace.

In that case the resolution will be prepare the foundation first or move forward with conditions, and in both you receive the route, not just the verdict. First the minimum foundation is corrected: process, owner, fields, sources, permissions, criteria and usage habit. Every gap leaves the verdict with its owner and its closing condition, so you know exactly what is missing, who resolves it and when it is closed.

That preparation is not wasted time, and it is the effect we describe above: ordering the process and cleaning the sources produces value with or without an agent.

Postponing is not a no. It is a "not yet, and this is what is missing".

You contract a professional service from Zherpa. Concretely: designing, implementing, transferring and governing an agentic productive capacity inside your operation.

You are not buying software: the software already exists and is licensed separately. You are not hiring an employee: the "Digital Candidate" is a position architecture, not a person.

We use the language of a position because it precisely describes what is designed: mission, processes, tools, permissions, metrics and limits. It is the same discipline used to define a human role, applied to an automated capacity. But there is no legal personhood, no labor relationship and no subject of responsibility other than the people already identified: the designated human owner and the Zherpa/client split.

The proposal separates the concepts so you can see what you pay and to whom:

That separation is not accounting formality. It is the reason we can recommend the most convenient architecture without defending a license margin.

Starting with one is not only possible: it is almost always the right call.

A candidate with a clear mission and limit produces evidence fast, is cheap to correct, and teaches the organization to operate with one agent before operating with several. That learning curve is real, and it pays to buy it cheap.

The criterion to know your case is simple: count decisions, not tasks.

• If the result closes within a single decision and a single owner, start with one candidate.
• If the result crosses several decisions, areas or owners, a cell of bounded roles is better, each with its own COA and its own risk class.

What is never advisable is a "super-agent" that does everything. When it fails, nobody knows which of its twelve functions failed, traceability is lost, and its risk class ends up being the most restrictive of all its tasks, which cancels the advantage of the rest.

The rule is ecosystem first: the smallest complete system that closes a value cycle, not the largest number of agents. One closed and measured cycle is worth more than five half-adopted capacities.

Every candidate in the catalog carries its maturity state in plain sight. We do not use "proven" as commercial decoration: evidence must be showable.

None of the three is bad. They are different risks and different prices of learning. A base profile can be the right decision if the case is yours and you want the advantage of going first, as long as you know you are going first.

What we do not do is present a base profile with production language. That is why the state appears next to each candidate, before the commercial conversation and before you commit budget.

The decision is not one of technological loyalty. We are no one's official partner and have no license quota to defend, so the architecture is chosen on technical and economic criteria, not on margin.

The criterion is where the work lives:

• Inside Zoho belongs the capacity triggered by events and tied to processes, permissions and records that are already there. The agent operates with the identity, roles and audit trail of the system your team already uses, without duplicating data or inventing a new perimeter.
• In an external model belongs complex reasoning, document reading and cross-system analysis that Zoho does not govern. There, cost per result and quality of reasoning weigh more than proximity to the record.

Most real cases use both layers. The right question is not which one wins, but which task lives in each.

Phase 0 identifies the constraints and the probable architecture. If moving forward is authorized, the project defines in detail which capacity lives in each layer, what data each may process, and with what identity, permissions and audit trail.

We do not assume the information "never leaves". We verify it and contract it. The data route, the provider that processes it, retention and training use are put in writing before the first record leaves your system.

We do not assign a single autonomy to the whole agent. We classify each action separately, by its impact and its reversibility. The same agent may execute one task on its own and be forbidden from executing the next.

The classification is not set by the vendor on its own. It is agreed with you in Phase 0, written into each agent's COA, and it is not permanent: when volume or context changes, the action is reclassified. A task can rise from A to B because its scope grew, and can go down from B to A when accumulated evidence justifies it. Expanding autonomy requires passing an evidence gate, never habit.

The human does not approve every click. The human is in command of the system, which is a different and higher position: defining purpose, setting limits, reviewing evidence and deciding where autonomy expands and where it does not.

A person with a first and last name. Never the agent.

An agent has no legal, ethical or executive responsibility. It cannot have it. That is why, before it enters operation, every agent has a human owner designated in writing, with name, position and scope. No owner, no agent in production.

The split is explicit from Phase 0. Zherpa answers for the design, the implementation and the agreed controls. You answer for your policies, your data, adoption and the decisions that belong to you. Neither party can transfer its share to the system.

And when something fails, nobody investigates blind. The COA defines what the agent may do, what it never does, when it stops, whom it obeys, what it records and how it is shut down. The RDA preserves why each relevant decision was made, with what evidence and who authorized it. Traceability is designed before the incident, not after.

The procedure is the same every time: stop the agent, contain the effect, reconstruct the decision with the RDA, correct the cause, reclassify the risk and document it. Resumption requires evidence, not trust.

No. AI expands your team's capabilities. It does not substitute them.

That is the Humano al Mando position, and it has an operational reason, not a rhetorical one: an agent executes tasks, it does not exercise judgment. It classifies, captures, reminds, searches and follows up with a constancy no person sustains eight hours a day. It does not read an upset client, it does not negotiate an exception, it does not take responsibility before the board. Your people do that, and they do it better when they stop spending the day on administrative work.

That is why design starts with the task, not the position. In the first cases we free hours of capture, classification, reminders, search and follow-up, and your team concentrates that capacity on judgment, relationships and decision. The same person performs better because they operate with better information and less friction.

What to do with the freed capacity is a leadership decision: serve more clients with the same headcount, shorten response times, open a line that is not viable today for lack of hands.

Any labor decision remains human and the client's, and must consider strategy, culture, performance and legal obligations. The agent neither makes it nor recommends it.

Technology acts; the human being directs.

They are two different clocks, and confusing them is the most common cause of disappointment in these projects. A fast demo does not equal demonstrated value.

The pilot: 2 to 8 weeks. The catalog estimates that range starting from an approved Phase 0 and the availability of data, access and owners. If those three conditions are delayed, the clock is delayed with them. At the end of this period you see the capacity working in your real operation, not a demonstration.

The Success Case: validated over a runway of 8 to 12 weeks afterwards. Here the four variables of the Master Formula run, and all four must be non-zero:

• Effective Implementation. The capacity operates at the agreed quality.
• Sustained Adoption. The team uses it in the real flow, consistently.
• High-Quality KPIs. The indicators are reliable and allow decisions.
• Time. The result holds without Zherpa being present.

We deliver the pilot. The Success Case is built by both parties, and that is why the second clock cannot be compressed with more budget or more consultants. Adoption takes what a habit takes.

Day one delivers the system. Value arrives afterwards, and it does not arrive on its own.

That is where the variable that decides the Success Case begins: Sustained Adoption over time. A dashboard turned on is not adoption. Adoption is real work happening inside the system, consistently, without us present.

We operate in 90-day cadence with a fixed ritual: review → decide → execute → measure. No artifact, no meeting; and narratives do not substitute numbers.

It is measured on three fronts:

• Adoption. Use in the real flow, by role and by process stage, read by cohorts and not by a global average that hides where it breaks.
• Result. The high-quality KPIs agreed from the start, cost per result, and the value effectively captured.
• Agent behavior. Quality sampling, errors, abstentions, escalations, drift, versions and incidents.

Risk is reclassified when volume or context changes. Class A, B or C is not a launch label; it is a living state. Capacity expands only by passing evidence gates: Explore → Evidence → Scale. No evidence, no scale; and a front with zero adoption is corrected or cancelled before it keeps consuming budget.

The goal of the accompaniment is to stop being necessary.