Social Dolphin Services
SDS · Field notes

How to keep AI from becoming lift-and-shift

The playbook for the conversation I lost a decade ago.

Type
Field note · part 2
Date
31 May 2026
Audience
Operators managing up

After I wrote about watching cloud lift-and-shift tank projects because the business moved fast on an operating model it refused to change, the same question came back from a dozen people, all of them some version of the engineer or ops lead I used to be: "Fine. So how do I not get overruled this time?"

That is the right question, and it is the one the first piece did not answer. Naming the trap is easy. Being the person in the room who keeps your company out of it, when the board wants AI in every workflow by next quarter, is the hard part. I lost that argument once. This is the playbook I wish I had then.

The goal is not to slow leadership down. That instinct is reasonable and it is never going away. The goal is to give leadership the speed they actually want, on a foundation that can hold it. Here is how you steer the conversation there.

Stop arguing against speed

The first move is the one that changes everything, and it is a reframe, not a tactic. Do not argue against moving fast. The moment you do, you are the brake, and the board stops listening.

Say yes to fast. Then make the redesign the thing that makes fast safe. "Yes, we can move quickly, and the way we move quickly without setting money on fire is to change the workflow around the AI, not just bolt it on." Now you are not the obstacle to speed. You are the person who knows how to actually deliver it. That single shift is most of the battle, because it puts you and the board on the same side of the table.

Six moves that keep the conversation honest

1. Say yes to fast, then define what fast lands on

Speed on an unchanged process compounds debt, not value. So the deal you offer is explicit: we move fast, and "fast" means we redesign the one workflow this touches before we wire the model into it. You are not asking for more time. You are spending the same time on the part that pays off.

2. No use case without a bottleneck and a P&L line

"AI everywhere" is not a strategy, it is a way to spend a budget. For every proposed use case, ask the room one question: what specific bottleneck does this remove, and which line on the P&L does that show up in? The use cases that cannot answer die on the slide, which is exactly where you want them to die, not in production six months and a quarter-million dollars later.

3. Put the unit economics on the table before you scale, not after

The fastest way to lose a bolt-on argument after the fact is the bill nobody modeled. So model it first. Cost per task, cost per interaction, cost per resolved ticket. "Scale it everywhere" meets arithmetic before it meets a budget, and the conversation moves from enthusiasm to economics while the cost of being wrong is still small.

4. Every pilot gets a baseline and a kill-or-continue line

The cloud pilots that bled money did not fail loudly. They drifted. "We are learning a lot" became a permanent state with a permanent invoice. So before a pilot starts, write down the baseline it has to beat and the threshold that ends it. "If it does not move this number by this date, we stop." That one sentence turns a pilot from an open-ended expense into a decision.

5. Name the foundation in risk language

The part the bolt-on skips is data governance, security, and the actual redesign of the work. To an engineer that is hygiene. To a board it sounds like delay, unless you name it as risk. Google Cloud's own security team makes this point: "just add a chatbot" skips the same foundation cloud lift-and-shift skipped. Put it in the board's language: this is the difference between an AI feature and an AI incident.

6. Bring the one slide

The room wants a yes in ten minutes. Give it a slide it can say yes to: the use case, the bottleneck it removes, the number it moves, the unit cost, the kill threshold, and the one workflow you will redesign first. If you cannot fit it on one slide, you do not understand it well enough to ship it yet, and that is useful to know before the board commits.

If you get overruled anyway

Sometimes you will lose the argument. I did. When that happens, do not go quiet and do not say "I told you so" later. Write down the decision, the assumption it rests on, and the one indicator that would prove it wrong, then send it to the room. "Proceeding with the bolt-on. The thing I will be watching is cost per interaction at scale. If it crosses this line, we revisit." You are not being difficult. You are creating the trigger that gets you the second conversation, on evidence, with your credibility intact. That memo is the difference between being right and being heard.

What this is not

  • Not a script for blocking AI. We ship AI in production every day, and we are aggressive about it.
  • Not a claim that every use case needs a six-month foundation project. Plenty are genuinely small, and you should move on them today.
  • Not a guarantee you will win the room. It is a way to be the person worth listening to when the bolt-on starts to wobble, which it will.

One-sentence takeaway

You do not keep AI from becoming lift-and-shift by slowing the business down, you keep it from becoming lift-and-shift by giving leadership the speed they want on a workflow you actually redesigned, and by writing down the trigger that gets you a second conversation if they overrule you.

Talk to us

If you are the person who can see the bolt-on coming and you are about to be in that room, send us the use case your board is most excited about. We will help you build the one slide: the bottleneck, the number, the unit economics, and the kill threshold. And if it turns out to be a real operating-model change rather than a bolt-on, that is the work we do.

We do not take every engagement, and we will tell you on the call whether we are the right partner.

Sources