AI is changing your org. Don't make these calls alone.
You're not asking "should we use AI." You've already made that call. You've run the pilots. You've redesigned a workflow. You've watched at least one team genuinely transform around AI capability. Now you're asking the questions that actually matter: what does this company look like in two years if we get this right — and what if we don't?
The decisions in this phase have weight. Some of them are about people you've worked with for ten years. Some of them are about brand and customer trust you've spent your career building. Some of them are about org design choices that will shape what the company does for the next decade. The cost of getting them wrong is not abstract.
Stage your decisions by reversibility. Make the reversible ones first; learn from them; only then make the harder-to-reverse ones.
This page is what we'd think about if we were sitting at your offsite. The decisions to stage. The frameworks for ordering them. The cautionary tales worth learning from for free. The work below isn't a "this week" list — it's a "this quarter and next" list, ordered for reversibility.
You might be here if…
- You've changed roles or org structure because of AI. Maybe one team got smaller. Maybe a role got rewritten. Maybe you didn't backfill a position because the work has been absorbed elsewhere. The change is real and it's already happened.
- You have written AI policies and governance. An acceptable use policy. Maybe a customer-facing AI disclosure. Maybe a vendor risk framework specific to AI tools. Real artifacts, not just intentions.
- You're measuring AI's contribution to revenue or cost — not just usage. Your dashboards have moved past "how many people are using Copilot" to "what's the dollar impact of the workflow we redesigned in Q2." The conversation with your CFO has changed.
If most of these don't sound like you, the page you actually want is probably another phase — see the four phases →
Why Phase 4 is structurally harder than the previous three
Phases 1 through 3 are mostly about tools and capability. You can buy your way into Phase 1, train your way into Phase 2, and champion your way through Phase 3. The hardest decisions are about resource allocation — which is what executives are trained to do.
Phase 4 is about structure and governance. The decisions are about people, accountability, brand risk, and the shape of work itself. Executives are also trained to do this — but it's slower, more political, and the failure modes are more public.
The two patterns to learn from at this stage are well-documented and worth studying carefully.
Klarna — the cautionary tale on speed
In February 2024, Klarna announced that its AI customer-service assistant was doing the work of 700 human agents and that it would replace its customer service org. The numbers looked airtight. The CEO declared victory publicly. Industry coverage was breathless.
By mid-2025 — fifteen months later — Klarna had publicly reversed course. Customer satisfaction had collapsed. The CEO was on record saying "we focused too much on efficiency and cost." The company restarted human hiring under what it called an "Uber-style" hybrid model.
The lesson isn't "don't use AI in customer service." Plenty of companies are doing that well. The lesson is about speed and reversibility: Klarna made a hard-to-reverse structural change at high speed, on a customer-facing function where errors damage brand directly. The reversal cost more than the original transition would have cost if they'd staged it carefully.
Shopify — the structural reset, done deliberately
On April 7, 2025, Shopify CEO Tobi Lütke published an internal memo (originally dated March 20) titled "Reflexive AI usage is now a baseline expectation at Shopify." The most-quoted line: "Before asking for more headcount and resources, teams must demonstrate why they cannot get what they want done using AI." AI usage became a factor in performance reviews. (Lütke's published memo)
We're not recommending you copy Shopify's policy — it's aggressive and not right for every culture. But the move is instructive in two ways. First, Lütke didn't replace any specific role; he changed the bar for new headcount requests. That's a reversible structural decision (you can change the policy) on a forward-looking lever (future hiring), not a backward-looking one (past employment). Second, he made the decision public deliberately — getting ahead of a leak, owning the framing, removing ambiguity. That combination is what staged structural change looks like at this phase.
Klarna and Shopify are bookends. The lessons aren't "do this, don't do that." The lessons are about how to think about staging structural decisions when the stakes are real.
Decisions to stage now
Five decisions, ordered by reversibility. The first two are work to do this quarter — they're diagnostic and policy work, both reversible if you change your mind. The next two are decisions to make next quarter, after you've done the diagnostic. The last is ongoing — orchestration that holds the whole thing together.
If your leadership team has a quarterly offsite, this list is the agenda. We've structured each decision so the work translates directly to a working session — bring the relevant people, bring the data, run the conversation, write the call.
- 01 this quarter
Audit your org chart against the work that actually exists today
The work has shifted. The org chart hasn't. Most Phase 4 companies are running structures designed for a pre-AI distribution of tasks — too many handoffs, too many approval layers, too much middle-management coordination of work that AI is now coordinating. You can't redesign what you haven't named.
Get your leadership team in a room. Map every function against three questions: What work has AI absorbed? What work is now redundant because AI absorbed something upstream? What new work has emerged that nobody owns yet? The output is a one-page snapshot of structural drift. That snapshot is the agenda for the harder decisions below.
Free resource: Org-Chart Drift Audit Template (coming soon) - 02 this quarter
Write the brand-risk and customer-facing AI policies before you need them
In February 2024, the British Columbia Civil Resolution Tribunal ruled that Air Canada was liable for misinformation given to a customer by its chatbot. The chatbot incorrectly told the customer he could apply retroactively for a bereavement fare. Air Canada argued the chatbot was a "separate entity" responsible for its own outputs. The tribunal rejected that argument and awarded $650 in damages plus fees.
The dollar amount is trivial. The precedent isn't. Companies are legally responsible for what their customer-facing AI says. Most Phase 4 SMBs don't have an AI brand-voice policy, an AI escalation policy, or a clear answer to "what happens when our customer-facing AI gets it wrong." Write them now. The cost of writing them is hours; the cost of not having them shows up as headline risk.
Free resource: AI Policy Starter Pack (coming soon) - 03 next quarter
Stage the structural decisions in order of reversibility
Not all org changes are equal. Restructuring a customer-service team is reversible (you can rehire). Restructuring an engineering function around an AI-native build pattern is harder to reverse (the architecture choices compound). Eliminating a senior role is the hardest to reverse — the institutional knowledge leaves with them and doesn't come back.
Stage your decisions by reversibility. Make the reversible ones first; learn from them; only then make the harder-to-reverse ones. The company that gets this right looks methodical from the outside. The company that gets it wrong looks like Klarna — they replaced 700-agent equivalent customer service with AI in 2024, declared victory, then publicly reversed in 2025 because customer satisfaction collapsed. "We focused too much on efficiency and cost," the CEO said. The reversal cost them more than the original transition.
Free resource: Restructuring Scenarios Planner (coming soon) - 04 next quarter
Identify the managers who are accelerators vs. the ones who are bottlenecks
The AI restructuring conversation eventually hits middle management. A 2025 Harvard Business School study found that 6 in 10 managers spend more than half their time on administrative tasks AI can now automate. McKinsey Global Institute estimates up to 30% of current managerial tasks could be automated within five years. The Gallup workplace data shows the average span of control has already increased from 10.9 to 12.1 direct reports per manager between 2024 and 2025.
The structural change isn't "do we have fewer managers" — it's "which managers are accelerators and which are bottlenecks." Accelerators champion bottom-up AI ideas, give their teams space to experiment, and make decisions faster than they used to. Bottlenecks gate-keep, slow approvals, and stall AI initiatives because they feel threatened. This is one of the harder calls in Phase 4. Long-tenured managers who've earned trust over years are the ones most likely to be threatened by AI-amplified ICs surfacing better ideas. Some of them grow into accelerators with coaching and a clear new mandate. Some don't. Knowing the difference is part of the work.
Free resource: Manager Accelerator vs. Bottleneck Diagnostic (coming soon) - 05 ongoing
Build orchestration before you need it
AI capability without coordination is just faster ways to go in different directions. The companies winning Phase 4 have what one of our clients called a "rudder" — a small executive function whose job is to ensure the AI-amplified work is pointed at strategic outcomes, not just faster outputs. Sometimes this is a Chief Transformation Officer. Sometimes it's the CEO themselves spending 20% of their time on it. Sometimes it's a fractional advisor.
Whatever shape it takes, the function does three things: it keeps the four-phase frame alive across the org (different teams are at different phases at any given time), it sets governance norms before incidents force them, and it adjudicates the structural decisions before they cascade. Without it, you'll do good Phase 4 work in three places and bad Phase 4 work in three others, and the bad will set the brand impression.
Free resource: AI Orchestration Function Charter (coming soon)
A 95-person logistics CEO at the inflection point
Here's the pattern, drawn from the Phase 4 SMBs we work with most often. A 95-person regional logistics company. Founder-CEO with 22 years in. Hired a Chief Transformation Officer 14 months ago specifically to drive AI and process change. Restructured one team — customer service — eliminating four positions and redistributing the work between AI tools and three cross-trained agents. The transition was hard. One employee left publicly upset; one of the kept agents has been stressed.
Now the CEO is facing a harder decision. His sales operations function (12 people) could probably run with five or six if he restructured. But these are loyal people who've been with him for years. He's also worried about brand risk — a competitor had an AI-driven customer service incident last quarter that made the local paper. He doesn't want that to be him.
He's thought about whether he needs an outside advisor for the next set of decisions. He's hesitated because most "AI consultants" he's talked to are either too junior or too theoretical.
The pattern that works for this company isn't a single big decision — it's three months of structured offsite work. Audit the org chart against the work that actually exists today. Stage the decisions by reversibility. Run a paired diagnostic on the sales-ops function: which roles are about judgment (keep), which are about coordination (AI absorbs), which are about execution (consolidate). Don't make the cut decisions until the structural map is on paper. When the cut decisions come, communicate them with care and respect — these are people the company has served, and they deserve a graceful exit.
A year from now, the company has fewer employees than it did and revenue is up double-digits. The CEO has answered the harder question — what does this company look like in two years — with a coherent plan that he believes in. The decisions weren't easier; they were just made deliberately.
If you'd rather not work through these decisions alone — Phase 4 is the highest-stakes engagement Sillewa runs. Most Phase 4 work is fractional advisory over a quarter or two, sometimes embedded inside a transformation office. Here's what working with us looks like →
What changes after you've done the work
When the staged decisions land — when your org chart actually reflects the work that exists, when your AI governance is written and tested, when your managers are sorted, when your orchestration function is real — the pressure changes. You stop reacting to AI; you start using it. The company shape becomes a deliberate output, not an inherited one.
That's where most of the Phase 4 work ends. There isn't a clean "Phase 5" — there's just running a company that's structurally fluent in AI, where the questions become normal strategic questions again instead of existential ones. We'll write more about what that looks like as we work with companies that get there.
For now: stage the decisions carefully. Keep the orchestration function alive. Watch for the new failure modes that emerge at this scale — they're different from the ones you've already seen. The biggest three to watch: model lock-in (when your operations become dependent on a specific AI vendor's capabilities), AI-amplified key-person risk (when the workflows your power users built become brittle if those people leave), and the slow erosion of organizational learning (when AI does the synthesis work your team used to do, the team can lose the skill of making sense of complexity). None of these have clean solutions yet. They're worth watching for, naming when they appear, and addressing before they compound.
If you found this useful, also read:
- Credits & sources — the Klarna, Shopify, Air Canada, McKinsey, and Gallup sources backing the claims on this page
- Working with Sillewa — how Phase 4 engagements typically run
- More about Sillewa — who we are and how we work
The other phases
Different teams in your org may be at different phases. Here's the rest: