IDT in Practice

Practical Rules

1. Ask Which Link in the Value Chain the Initiative Falls On

And how much value it moves there. If it falls on none, it can almost always wait.

2. Decide What Will Not Be Transformed

It is as strategic as the scope.

3. Do Not Automate Waste

Simplify first. In Layer 2, the order Lean → Automation → Agents is not negotiable.

4. A Quick Win Only Counts If It Comes from an Already Simplified Process

And if it is compatible with the future structure.

5. The 50% Rule

An automation only pays off when there are people who spend more than half their time on a repetitive, daily process. For example, automating order entry for a team that does it full time transforms operations; automating what twenty store managers do once in a while ends up as a tool nobody opens.

The question is not “can it be automated?” but “is there someone whose day changes if I automate it?”
— Mauricio Peschard

6. Climb the Analytics Ladder Without Skipping Rungs

Each one rests on the previous one:

1. Descriptive: what happened? A reliable view of sales, operations, and profitability. It is the floor.
2. Diagnostic: why did it happen? Which category dropped in which store, which supplier is eating the margin, where the bottleneck is.
3. Predictive: what could happen? Demand, risks, cash flow, stockouts.
4. Prescriptive: what should we do? Recommended actions and priorities.
5. AI automation: how do we scale it? Agents and automations that turn decisions into execution, in a measurable and governed way.

The mistake MIT documented [1] looks like this: jumping from the floor (raw data in an ERP) to the top rung (an agent).

An AI agent cannot be asked to decide the future of a business that still cannot explain its present.
— Mauricio Peschard

7. Context, Not the Model, Makes an Agent Reliable

That context includes:

written business rules: the dictionary for every term, the exact formula for every figure, what is included and what is excluded (always flagging it), and how to compare fairly;
locks against making up data: if something is not there, it says so and does not estimate it;
in addition to the rules of the present, the company’s future objectives. When the agent knows where the business is heading, it stops describing and starts answering with judgment.

8. Validate Every Answer Against an Independent Source

The same figure calculated two ways.

An AI answer that cannot be checked against anything is not data: it is an opinion formatted as a report.
— Mauricio Peschard

Who Leads the Transformation

It Is Not Delegated to IT

For 20 years, initiatives of this kind were delegated to IT. That made sense when transforming meant installing systems; it stopped making sense when transforming came to mean redefining how the company operates and decides.

IT can install a platform, but it cannot decide which value streams are priorities or how much autonomy is acceptable for an agent that approves credit. Those are business decisions. When leadership delegates them, it does not delegate execution: it delegates strategy. That is why IDT places Strategic Direction in the first layer.

The Roles

Sponsor: every initiative needs one with real authority, able to unblock decisions that cut across areas.
Middle management: this is where the real quick wins are, because it knows the processes first-hand and knows where the friction is that never shows up in any report. Leadership sets the course; middle management finds the ground.
The orchestrator: the profile that brings together people, processes, and information so that the company works as a system.

In 2040 the advantage will not be having the best technology, but being the only one that got theirs to work as a single system.
— Mauricio Peschard

The Market Signal

In April 2026, Microsoft refocused its strategy to drive AI transformation through its partner network, with the goal of taking companies beyond pilots [2].
A month later, Microsoft and EY announced a joint investment of more than one billion dollars over five years with the same purpose [3].
Those who sell the technology are betting on the layer that turns it into value: the trusted advisor.

Even those who sell AI have understood that selling it is not enough; that is why they now also build the bridge.
— Mauricio Peschard

SMBs and Large Companies: Two Speeds, One Destination

Large companies: they have standardized processes and investment capacity, but they carry complexity and inertia. Their constraint is not getting started, but keeping coherence. For them, architecture and governance are the condition for transformation not to break into a hundred parallel initiatives.
SMBs: knowledge lives in the heads of specific people, and processes exist in practice but not in any document. Their constraint is that dependence; their advantage, speed of decision. Their biggest opportunity is the most basic one: turn what is now one person’s memory into a process, and then automate it.
The common destination: no company will operate in 2040 without AI built into the way it works, for the same reason that no company operates today without the internet.

Cases

The Home Depot®: Structure First, Capability Later

The author worked for almost 20 years at The Home Depot®; the case uses only public information.

Starting point (late 2017): it was the world’s largest home improvement retailer, with more than two thousand stores and record sales. But the physical and digital worlds ran on separate tracks: e-commerce was less than 7% of sales, and the store and the website did not share information.
The decision: One Home Depot (December 2017), 11 billion dollars over three years, invested in the foundations rather than a digital storefront [4]. That meant more than one billion for a network of about 170 distribution centers able to deliver to 90% of the U.S. population the same or next day [5]; stores turned into fulfillment centers; and a common data platform.
The proof (2020): digital sales nearly doubled in one year, and the infrastructure absorbed them without breaking [6].

The Home Depot® did not win the 2020 digital wave in 2020. It won it in 2017, when it decided to build the foundations before the storefront.
— Mauricio Peschard

The Notary Office: A Failure That Produced the 50% Rule

First case, successful: for a notary office, the author automated data capture for the parties to each deed. Two lawyers spent almost their entire day on that task. An AI assistant read the documents, extracted the information, and left it pre-filled for review. Capture time dropped to one third from the first week.
The replica, failed: the same solution, in another organization where the process was spread across 10 lawyers and took less than 10% of each one’s day, failed. Not because of the technology, which was identical: there was nobody whose time it could free up. The tool was abandoned because it had no owner who needed it.
The lesson: this is where the 50% rule comes from.

Automating a task nobody does full time is solving a problem nobody has.
— Mauricio Peschard

The Wholesaler: From Data to an AI Agent

Starting point: a chain with several stores and a solid ERP, but no analytics. It had all the data and could not see it: every question required exporting to Excel and building the analysis by hand.
What was not done (2025): the temptation was to connect an agent to the ERP. It was not done, and that decision is what made the project work.
The floor first: a data platform that consolidated the ERP into an analytical model (by supplier, store, category, item, sales, and profitability), usable in Power BI.
Then, the agent: once the analytics were validated, the Sales and Profitability Agent was added, built on the Model Context Protocol (MCP) [7], with about 600 lines of business rules and locks against making up data.
The result: leadership asked how sales from January to May had evolved over recent years. The agent showed two readings. Across the whole chain, sales were growing better each year. In same stores, the established business was growing below target: almost all the growth came from new stores.
Why it could be trusted: every number was checked against Power BI.

Reading the two cases together gives the complete pincer: first the slow arm (the data platform); on top of that structure, the quick win (the agent), fast precisely because the structure already existed.

Self-Assessment: 6 + 3 Questions

One question per dimension, answered honestly and without the benefit of the doubt:

1. Strategic direction. Can the company describe precisely how it wants to operate in the future, or does it only have an approved technology shopping list?
2. Process redesign. In the last three automations, was the process redesigned first, or digitized as it was?
3. Automation. Are the automations connected to each other and to a common structure, or are they loose solutions that depend on whoever built them?
4. Digital agents. If an automated process made a wrong decision today that affects a customer, is there a defined owner, a log, and a human review point?
5. Architecture. When the company needs a new capability, does it reuse what it already has or build again from scratch?
6. Governance. Are there clear, enforced rules on AI use, limits, logging, and validation, or does each area decide on its own?

How to Read It

It is not scored or averaged. What matters is which one got the worst answer, because in a system progress is set by the weakest piece. That is where the work starts, even though it is rarely the most impressive dimension to present to the board.
The quick read: if more than two of the six questions are uncomfortable, the company is currently in the 95% profile.

A company’s digital maturity is not defined by its best capability, but by its weakest.
— Mauricio Peschard

Three Questions of a Different Kind

1. If tomorrow the company lost its technology and kept only its processes, its data, and its people, how far would it be from operating again?
2. Of the approved AI initiatives, how many can be tied to a specific business number that should move, and in what time frame?
3. What is being done today, concretely, about the weakest dimension just identified?

References

MIT NANDA. (2025). The GenAI Divide: State of AI in Business 2025. Massachusetts Institute of Technology.
Microsoft. (2026, April 21). Accelerating frontier transformation with Microsoft partners [Blog post]. Microsoft.
EY, & Microsoft. (2026, May 21). Global initiative to help clients scale AI enterprisewide value creation. Microsoft Source.
The Home Depot. (2017, December). One Home Depot: Plan de inversión de 11,000 millones de dólares a tres años [Corporate announcement].
The Home Depot. (2018). Red de distribución: Inversión aproximada de 1,200 millones de dólares y cerca de 170 centros de distribución. Industrial Distribution.
Danziger, P. N. (2020). Home Depot’s transformation to a fully interconnected retailer shows record-breaking results. Forbes.
Anthropic. (2024). Model Context Protocol (MCP). https://modelcontextprotocol.io