What an AI Transformation Actually Looks Like, Week by Week

“AI transformation” has been claimed by so many slide decks that the words barely mean anything. So this piece is deliberately concrete: the actual sequence we run at Relote when a company hires us to rebuild its operations around AI, week by week, including the parts that are unglamorous and the places where things wobble. If you have read the rest of this track, you will recognize the ideas. This is what they look like on a calendar.

One framing note before the schedule: a transformation is not a software rollout. Tools are the easy part. The work is mapping how your business actually runs, deciding which parts of it AI should carry, and proving each change with numbers before anything old is switched off. The whole method can be compressed into one sentence: map, decide, set up, prove, hand over.

Weeks 1 and 2: the assessment

It starts with mapping the real work. Not the org chart and not the process wiki: the actual movement of decisions, handoffs and busywork. We sit inside your operations, watch how things truly get done, and write down what we see, including the workarounds nobody documented and the ex-employee’s spreadsheet everything secretly depends on.

Two weeks is enough because we are not mapping everything; we are mapping where the time and the context actually leak. The questions we keep asking: where does information get re-typed from one system into another? What does everyone re-explain weekly? Which meetings exist to move status around? What does only one person know? Which AI tools are people already using, quietly, and what are they pasting into them? The answers cluster fast, and they cluster in the same places at most companies, which is why this academy’s use cases look the way they do.

The output is a written assessment: where your time actually goes, which work AI should carry first, in what order, and what each is worth. It is specific to your company, and it is yours to keep whether or not you continue.

Month 1: triage and the first memory

With the map in hand, the first real decision: task by task, what stays human and what AI should carry. The split is rarely what people expect. Judgement, relationships and taste stay human. What moves to AI is coordination, reporting, routine drafting and the endless re-explaining, the connective tissue that eats your team’s week without anyone having chosen it. We write this split down explicitly, because “what AI will not do here” turns out to calm more anxieties than any all-hands presentation.

In parallel, your company gets its context layer: Context Engine, set up on your own account from day one. We help you decide how many engines you need, usually one for the company and one per major client or project, and write each engine’s purpose with you, because that purpose is what tells every AI tool what belongs where. Then each person connects the AI tool they already use: ChatGPT, Claude, Claude Code or Cursor. We seed the memory so it is useful in week one instead of month three: your key documents and folders go into Files, where text and PDFs become searchable, and your website is read to suggest your values and voice, which you keep or discard. Within days the engine holds your clients, your decisions and your commitments as typed, linked memories.

Month 1 also installs the one new habit the whole system runs on: ending meaningful work with “save what matters,” and a few minutes a week looking at what was saved, answering the duplicate questions on Home and correcting what is wrong. Small, boring, and the hinge of everything, because a memory nobody vouches for is a memory nobody will trust.

Months 2 and 3: first workflows, run in parallel

Now workflows go live, and this is where our method differs from the “deploy and hope” school. Nothing old is switched off on faith. Each new AI-carried workflow runs alongside the existing way until the numbers prove it: fewer hours, fewer errors, fewer times a human had to step in.

The first workflows are usually the ones this academy documents: pre-meeting briefs built from what the memory already knows, a decision log that fills in as decisions are made, action items that become tasks with an owner and a due date. In each, your AI tool does the work, using its own access to your calendar or transcripts where you have connected those to it, and the engine is where the result is kept. They are chosen deliberately: visible daily, low risk, individually useful within a week, and they build the habit that matters most, your team learning to trust, check and correct the system. Trust grows from watching the parallel run, not from a training session.

Expect one workflow to underperform. That is normal and it is the point of the parallel run: we tune it or kill it based on its numbers, not its demo. Expect, also, one or two people to be skeptics, and do not fight it: skeptics who watch the parallel run and then convert become the strongest internal advocates you will get, precisely because everyone knows they were not sold, they were shown.

What the middle months feel like from inside, honestly: week 6 is the wobble. The novelty has worn off, the answers still have gaps because the memory is young, and someone will ask whether this is worth it. This is where the numbers from the parallel runs earn their keep, and why we collect them from day one.

Months 4 and 5: deepen and connect

With the foundations trusted, the work reaches further. Account records stay current from calls and email, with your AI tool proposing updates and your team approving them instead of doing data entry. Weekly status assembles from what was saved instead of from a meeting. Sales calls become account memory you can ask about. Your brand voice becomes part of the memory, so every drafting tool writes with it. Where your industry has its own shapes, client handovers at an agency, matter files at a firm, those move onto the same layer now.

This is also when the compounding starts showing up in the numbers rather than the rhetoric. Every tool saves into the same memory that every other tool reads, so briefs get sharper because account records are cleaner, and account records get cleaner because calls are captured. Improvements stop being additive. This compounding is the actual product of a transformation; the individual workflows are just its visible surface.

Month 6: handover

The last month inverts the relationship: we teach your team to run and extend the system themselves. How to connect a new AI tool and choose its access, when a new engine is warranted and how to write its purpose, how to read Summaries and Memories to see what the business has been saving, how to add the next workflow without us. The test we hold ourselves to: your team ships one new workflow, end to end, with us watching rather than driving. Your people stop being users and become operators.

Then you own it, per the ownership tests this track laid out. The engines are on your account. Any connection we used during the engagement is cut off. You can export every engine as Markdown and JSON that opens without Relote and without Context Engine, at any time. Success for us is you needing us less every month. Not a dependency that renews forever.

What it costs, plainly

The transformation is $3,900 a month for six months. After that there is no consulting fee left to pay. What continues is your Context Engine plan, which you can see on the pricing page and change or leave whenever you like.

Worth stating what the transformation fee is not buying: seats, per-user licenses or a metered relationship with your own memory. It buys the map, the setup, the tuning and a team that can run it, after which the economics of every additional workflow are your own.

If the way your company runs today would not survive honest mapping, that is precisely the argument for those first two weeks: a written map of where your operation actually leaks time, and a concrete plan for what AI should carry first. Start with an assessment.