How to use it. Each section says what goes there and includes an example. The example is a fictional company — “Cobros Ejemplo S.A.S.”, a 60-person debt-collection firm — and its numbers are made up for illustration; they’re nobody’s data. Copy the file, delete the example and fill it in with yours.
1. Profile
What goes here: industry, size, business lines, channels, which AI they already use, and who pushes for this internally.
Example (fictional): third-party debt collection · 60 people · 3 lines (early-stage, late-stage, legal) · channels: phone, WhatsApp, email · they already use a generic chatbot on the website · who pushes: the head of operations.
2. Today — in their words
What goes here: the situation told the way the team tells it, not the way we would. If a sentence sounds like a brochure, it isn’t theirs.
Example (fictional): “Our agents spend the morning searching three systems for what each debtor was promised before they can call.”
3. The cost of waiting
What goes here: a number. You get there by asking “so what?” until the pain lands on money or hours.
Example (fictional): 40 agents × 1.5 hours a day searching for context ≈ 60 hours a day that aren’t calls.
4. Where AI moves the needle
What goes here: the processes ranked from highest to lowest return, each with the metric that measures it.
| Process | Metric | Expected return |
|---|---|---|
| Debtor summary before each call | prep minutes per call | high |
| Classifying WhatsApp replies | inbox hours per day | medium |
| Drafting payment agreements | drafting time | medium |
5. The future state
What goes here: how leadership would say it out loud 90 days from now. One or two sentences.
Example (fictional): “Every agent starts the call knowing everything the debtor was promised, without opening another system.”
6. Targets
What goes here: measurable, dated outcomes, using their metrics. Each target has a type: make money · save money · go faster · reduce risk.
Example (fictional):
- Cut prep per call from 4 minutes to 1 by day 45 — go faster.
- Recover 30 agent-hours a day by day 90 — save money.
7. The map
What goes here: the capabilities needed to get from today to the future state. Their shopping list, not our feature list.
Example (fictional): read access to the three systems · a per-debtor summary that builds itself · one person inside who owns the workflow.
8. The plan in three stretches
| Stretch | What happens | What’s running at the end |
|---|---|---|
| Days 1–15 | Diagnostic | This document, with processes ranked by return |
| Days 16–30 | Plan and first installs | Two workflows running in production |
| Days 31–90 | Install, measure, hand over | Everything measured against the targets, and a trained internal owner |
9. One automation already started
What goes here: a small piece working before anything gets signed. Every step that matters waits for a person’s approval.
Example (fictional): a message comes in → it’s classified → the procedure is looked up → the history is read → the customer record is cross-checked → a reply is drafted → a person approves it before it’s sent.
10. Three diagnostic questions
- What does your best person do today that nobody else knows how to do?
- Which task repeats every day and nobody wants?
- Which number do you look at on Monday to know whether the week went well?
11. How not to blow it
- No tools before you have the cost-of-waiting number.
- No automation without one person who approves it and another who owns it.
- Real customer data only with a legal basis and their consent; for testing, test data.
Template from Sin Filtro series 01 — “The framework for working with AI”. Use it, copy it, improve it.