You are looking at two numbers on the same page. One is the tariff on goods you move across the border. The other is the wage for the same work done in Ontario or Ohio. The first number is pushing you to move the work home. The second number is what makes that move hard to afford.
Reshoring means bringing manufacturing, logistics, customer operations or technical support back to your own country. Nearshoring means moving it to a neighbouring, allied market. The labels matter less than the decision underneath them. You are not just changing where the work sits. You are changing what the work costs to do well.
The move fails when leaders copy the offshore headcount model into a high-wage market and expect the same unit cost. That is the trap. The lease gets signed, the software gets bought, the vendors get rerouted, and the people side is left to sort itself out. It does not.
Trade pressure is real, and it is not abstract. The United States imposed an additional 50 percent duty under section 338 on a range of Canadian goods, effective August 22, 2026. Goods qualifying under CUSMA are not exempt from that duty. Ottawa extended its tariff remission to June 30, 2028 on November 9, 2026, while withdrawing relief for some trailers, aluminum wire and paperboard where domestic supply exists. The agreement itself was not renewed at the July 1, 2026 joint review, so reviews are now annual, and the United States Trade Representative opened consultation for the 2027 review on October 5, 2026. None of that tells you whether your own move will pay. It tells you the pressure to move will not lift soon.
The constraint is not space or equipment. In the 2026 USA Reshoring Survey from the Reshoring Initiative and Regions Recruiting, covering 249 manufacturers, 66 percent rated hiring skilled technicians as very difficult or at crisis levels. That is the bottleneck to name before you hire. If you replicate the offshore team locally, you pay more per person and you still cannot fill the roles.
I have seen the same trap in a different move. A prior learning platform migration took about three years. When we moved again, we moved staff first, in small role-customised waves, instead of moving everyone at once. We moved 60,000 users in less than 14 months. First-year turnover fell to about half, with no increase in compensation. Staff only moved to Workday Learning. External clients, career college learners and FNIM learners stayed on Lambda. The lesson was not that moving is easy. It was that sequence is the intervention. We did not copy the old plan into a new system. We redesigned the move, then redeployed the people who already knew the work.
Start with work allocation, not headcount. Break the incoming work into what software should do, what AI should do first, and what judgment must stay human. Routine data entry, compliance checks and basic routing go to software. Drafting, anomaly detection, customer triage and initial analysis go to AI first, with humans stepping in for exceptions. Complex problem solving, client escalations, supplier negotiations and final quality sign-off stay with your local team. If you hire before you make those calls, you hire for a process you are about to replace.
Look inside before you recruit outside. Hiring people who know both your industry and AI tools is slow and expensive. You likely have mid-career people in shrinking legacy roles who already know your customers, your products and your compliance rules. Retraining and redeploying those people into redesigned roles is faster than starting from scratch on a job board, and it keeps the knowledge that never makes it into a manual.
Do not put redeployed staff through six weeks of classroom training while tariffs shift in real time. Put guidance where the work happens. Build checklists, vetted AI prompts and decision trees inside the software screens they use each day. Use AI sandbox simulations so staff can practise tricky exceptions before they touch live customers or live inventory. Pair departing offshore leads with incoming local staff for a focused overlap to capture the unwritten knowledge that manuals miss.
Measure business output, not training activity. Course completions do not tell you whether the move pays. Track time to autonomy, the days until a local worker hits target productivity without hand-holding. Track exception accuracy, how well workers catch AI errors and handle edge cases. Track cost per transaction, whether redesigned local workflows match or beat old offshore unit costs. If those measures do not move, the redesign has not worked, and hiring more people will not fix it.
Globalisation is not ending, but frictionless lowest-cost operations are no longer a safe plan. Technology alone will not save your margins. The winners will be leaders who name the bottleneck before they spend on the wrong fix, redesign the work, redeploy the people who already know the business, and prove the change in output, not activity.
If you want to name the bottleneck in your own function before you hire, the AI-Ready L&D Function Assessment applies five questions to your L&D function to show where you are genuinely prepared and where a small change will make the biggest difference: https://ravindertulsiani.com/ai-ready-ld-function-assessment/


