Method · How we work
The TRAZO Method
Before putting AI into a process, we map how it works today, measure its starting point and check whether it really needs AI. Only then do we design the agent, with clear rules and a person in the loop, and judge it by its numbers.
Method created by Alejandro Romero Conde, AI and Automation Specialist at SmartNexo.
- TStep 01 / 05
Trace the workflow
We gather how the process works today and measure its starting point.
- Who takes part: customer, advisor, back office, supplier.
- What happens step by step and what triggers it: a WhatsApp message, a call, a form or an email.
- Where it gets stuck: waiting times, data copied by hand, repeated questions, lost leads.
- Volume and hours: how many cases arrive, when, and how many go unanswered.
- Baseline: we measure today the indicators we will later improve. No baseline, no project.
DeliverableWorkflow map and indicator baseline. - RStep 02 / 05
Rules or AI?
We review the workflow step by step to decide whether AI is really worth it.
- Deterministic (same input, same answer: confirm an appointment, send a catalog, log a record): solved with rules, which is cheaper, faster and more predictable.
- Needs judgment or language (understand an open question, classify intent, summarize a conversation): this is where AI helps.
- Not worth it: if volume is low or mistakes are costly, we say so and don't automate for the sake of it.
DeliverableStep matrix (rules, AI or person) with a justified decision.Inbox · workflow steps
Example classification. Every project has its own matrix.
- AStep 03 / 05
Agent and human
We design the agent and the exact place of the person in the loop.
- What the agent does: channels (WhatsApp, voice), tone, knowledge base and the questions it must ask.
- What it connects to: CRM, calendar, catalog or order system.
- When it hands over to a person: upset customer, out-of-scope case, high-value sale or a question it cannot answer with certainty, always with the conversation context.
DeliverableAgent design and escalation rules. - ZStep 04 / 05
Safe zone
Lightweight governance, no bureaucracy: the minimum to operate with peace of mind.
- Which data the agent sees and stores, and which it never asks for.
- What it never answers on its own: unconfirmed prices, promises, legal or medical topics.
- Customers know they are talking to an AI assistant and can ask for a person.
- Who approves changes and how to switch it off in a minute if something fails.
- Compliance with the personal data rules of the customer's country.
DeliverableOne-page checklist agreed with the client. - OStep 05 / 05
Objectives measured
What isn't measured can't be managed: the indicators decide whether the project succeeds or fails.
- 3 to 5 indicators per project, for example: first response time, conversations handled after hours, cases solved without a person, qualified leads in the CRM, escalation rate or complaints.
- Success criterion defined before launch and compared with the baseline from step T.
- Weekly review in the first month and monthly afterwards: the data decides whether to scale, adjust or switch off.
DeliverableIndicator dashboard versus baseline and a documented decision.BaselineTargetReview- First response time
- After-hours coverage
- Cases solved without a person
- Qualified leads in the CRM
- Escalation rate
Example indicators: 3 to 5 are chosen per project.
What makes TRAZO different
The «rules or AI?» question comes first
Not every process needs AI. If a step is deterministic, we solve it with rules and save cost and risk.
Baseline before building
We measure the starting point from day one, so success or failure is shown with numbers, not impressions.
A person always in the loop
The agent knows when to hand the case over, and does it with full context so nobody has to repeat their story.
Frequently asked questions
- What is the TRAZO Method?
- It is how SmartNexo decides, designs and measures an AI agent in five steps: Trace the workflow, Rules or AI?, Agent and human, Safe zone and Objectives measured.
- Does TRAZO always end in an AI agent?
- No. If the «Rules or AI?» step shows that a step is deterministic or not worth it, we recommend automating it with rules or leaving it as it is.
- Why measure before starting?
- Because without a baseline there is no way to know whether the agent improved anything. Indicators and the success criterion are agreed before launch.
- What if the agent can't answer?
- It hands the conversation to a person on the team with full context, following the escalation rules defined in step A.
Shall we trace your first workflow?
Tell us which process takes up most of your time and we will review it with TRAZO.