When companies fire thousands while praising AI productivity, the math reveals a deeper truth: you cannot automate your way out of broken operations.
The productivity paradox in plain sight
Salesforce reportedly fired 4,000 people and praised AI for making the company more productive. The CEO line was blunt: “I need less heads with AI.” Revenue growth is reportedly at a low. The story reads like a win for efficiency. The numbers argue otherwise. When productivity climbs, value should follow. If it does not, you did not improve productivity, you hid the cost somewhere else.
As Regis Haegler reports:
This is the pattern I keep seeing. AI is treated like a cure-all instead of what it is: an amplifier. It scales whatever it touches, clarity or confusion. If your operating rhythm is off, adding automation just helps you make mistakes faster. If your process is sound, AI can multiply the win. Structured thinking before smart tools. That is the actual leverage.
Automating broken systems multiplies loss
Regis Haegler continues:
Last month, a $200M healthcare CEO called me: “We spent $3M on AI tools. Revenue is flat. What went wrong?” We examined their billing flow. The process was already broken, no clean handoffs, mismatched codes, zero feedback loop. They were dropping money in the cracks and then asking AI to sweep faster. Ferrari engine, no wheels.
We killed the AI project. Fixed the billing process manually. Cleaned inputs. Simplified exceptions. Rebuilt the handoff. Put in a human-in-the-loop for the edge cases. Six weeks later: $8M recovered revenue. No new AI. Just disciplined work and a clear line of sight from input to cash.
When you automate a flaw, you institutionalize the flaw.
I call this dysfunctional automation. It looks like progress on dashboards and feels like slippage in the bank. The lesson is simple, if you cannot map the steps and measure the outcomes, you are not ready for AI. Make the process visible. Name the friction. Fix the basics.
Augment winners, not replace laggards
Two patterns emerges from the field:
- Company A spent $5M on AI chatbots. Customer satisfaction fell 40%. Why? The bot covered for a support process that did not have the authority or context to resolve real issues. The brand paid the tax.
- Company B used AI to help top reps find and frame leads. Revenue rose 35%. Same tool, different target. They gave their best people leverage, not excuses.
Across 10 companies I advised, 8 tried to “AI their way” to growth and burned over $400M collectively. Two succeeded, only after fixing operations, then adding AI to amplify what already worked. Winner augmentation beats loser replacement. Give your top 20% tools that extend their reach and judgment. Do not try to make the bottom 20% less terrible with automation; you will calcify mediocrity.
Here is a practical way to think about it: AI extends cognition; it does not supply it. The best results come when you apply AI to a clear model of work. That is cognitive design, building a simple, shared operating system for thought around how value actually flows in your business.
An AI readiness checklist you can run this week
John Deacon comments:
Before you spend on models, spend on clarity. Use this quick pass to test if you are ready. If you fail any step, stop and fix it manually first.
1) Fix your broken stuff first
- Map the worst process end-to-end: inputs, handoffs, decisions, outputs, feedback loop. One page, no fluff.
- Prove it works manually for a full cycle. Measure throughput, error rates, time to value. If you cannot track it by hand, you will not track it with AI.
- Target: recover value before automation. We found $12M in one org by fixing manual processes first.
2) Make winners win bigger
- Identify your top 20% performers. Ask what slows them down and what patterns they already trust.
- Deploy AI to remove their bottlenecks: prospecting triage, research summaries, drafting first passes, next-best-action cues.
- Guardrail: keep the human judgment on the loop. AI proposes; your experts dispose.
3) If customers hate it, kill it
- Watch satisfaction on any AI touchpoint. If your chatbot has a 12% satisfaction rate, that is not innovation, that is customer abuse.
- Escalation must be simple and fast. Bad AI should fail open to a human, not trap customers in loops.
- Measure: time-to-resolution, first-contact resolution, and churn after bot contact. If the numbers turn south, roll it back.
4) Right-size the ambition
- Start with one process, one team, one KPI. Ship a thin slice in weeks, not a platform in quarters.
- Tie spend to recovered value or defensible learning. No vanity pilots.
- Keep model complexity low until the data proves you need more.
5) Build a thinking architecture
- Create a lightweight decision frame, purpose, process, people, proof.
- Purpose: What problem are we solving and why now?
- Process: Is the flow stable and measurable?
- People: Who benefits, who decides, who is on the hook?
- Proof: What evidence shows it worked? What breaks if we are wrong?
- Write this once and reuse it. That is your operating system for thought. It keeps the team aligned and your spend honest.
Conscious leadership beats headcount theater
Layoffs framed as “AI productivity” are often a symptom, not a solution. If you cut talent without fixing the work, you shrink capacity and keep the drag. If you ship AI to hide bad process, you pay twice, first in cash, then in trust.
AI is an amplifier. It multiplies what already lives in your system, discipline or drift.
Counterpoints worth holding:
- Not every company fits cleanly into winner/loser buckets. AI can upskill the middle, and often should.
- In complex systems, AI can help surface the broken process itself. Use it as a flashlight, not a crutch.
- Sometimes the right move is to design a new operating model, not automate the old one. But even then, prove the logic in a small loop before you scale the bet.
If you want leverage, earn it with clarity first. Fix the inputs. Tighten the loops. Put the tool where the work is already strong. Which broken process are you trying to AI your way out of? Map it today. Run it clean for a cycle. Then, and only then, add the horsepower.
To translate this into action, here's a prompt you can run with an AI assistant or in your own journal.
Try this…
Map your worst process end-to-end on one page: inputs, handoffs, decisions, outputs, feedback loop. Run it manually for one full cycle before adding any AI.
