What the 2026 Tariff Adjustments Did to Retail AI Roadmaps that were written in January

Quick answer

It is 8:50am on a Thursday. There is a 9am sourcing meeting on the calendar about the new landed-cost projections following the tariff adjustment. At 10am, the AI implementation vendor wants a go or no-go on phase two. The vendor does not know that the CFO put a soft hold on discretionary tech spend two weeks ago. The CFO does not know that pausing phase two now will cost a restart fee larger than the phase-two invoice itself. Both of those facts have been sitting in one person's head for nine days. And the calendar, already showing zero slack for the month, offered no moment that felt right to surface either one. That gap, held in one person's body while the implementation timeline slips one day at a time, is where retail AI projects go quiet. This brief looks at the mechanics of why, and at what the tariff environment specifically did to make the gap harder to close.

TL;DR

The statistics that frame this moment:

• Tariffs covered roughly half of U.S. goods imports by value in 2026 (Yale Budget Lab, April 2026)

  The Tax Foundation estimates new tariffs affect 54 percent of U.S. goods imports in 2026, with U.S. tariff policy changing more than 50 times since January 2025. Retail roadmaps written in January were built on cost structures that no longer exist (Tax Foundation, 2026).

• A majority of goods-sector companies reported higher cost of goods sold in 2026, with margin pressure following across consumer-facing categories

  A KPMG tariff survey conducted through Q1 2026 found 78 percent of organizations reporting higher cost of goods sold, 51 percent facing margin declines, and 68 percent having postponed major investments by one to twelve months (KPMG, 2026).

• Tariff-driven uncertainty pushed companies away from long-term AI and automation funding toward short-term cost reduction

  A PYMNTS survey of product leaders at goods firms found that tariff-driven uncertainty had constrained their companies' ability to fund AI and automation initiatives, and that disruption had pushed their organizations away from long-term initiatives toward cost-saving operational adjustments (PYMNTS, 2025).

• Nearly half of enterprises report running AI spend over budget in the second half of 2026

  Futurum Group's survey of 1,636 global enterprise technology decision makers found that 46.9 percent of organizations report actual AI spend exceeding their planned budgets, versus only 5.6 percent coming in below plan (Futurum Group, 2026).

• Annual AI run costs routinely reach 20 to 40 percent of the original build cost (Gartner, 2024)

  A paused implementation does not simply stop accruing cost. Annual run costs for an AI system land at 20 to 40 percent of the build cost, and the hidden layer of data prep, evaluation infrastructure, and change management typically equals or exceeds the build itself (Launch Day Advisors, 2026).

• Retail total tech budgets reach $113 billion in 2026 (Gartner, 2026)

  Forrester estimates total U.S. retail technology budgets will reach $113 billion in 2026, up 6.6 percent from the prior year. The budget is growing. The number of competing claims on it is growing faster (Forrester, 2026).

Before the numbers

I want to say something about the 8:50am before I say anything about the tariff numbers, because the tariff numbers come second. I have been in the position of holding two conversations in my head simultaneously, each one waiting for the other one to resolve first, while the calendar filled in around both of them. I know what that waiting feels like. It feels like discipline. Like reading the room. Like being responsible about timing. What I did not see clearly until I was on the other side of it: the decision to wait is made very, very early, often before the month starts, and then every day after that confirms it because the day is always too full. In August I was ramping up several AI agent projects and winding down another client engagement at the same time. That kind of month has its own energetic physics: everything legitimately needs to happen right now, and the natural response is to tighten up, narrow down, and push through. I had one Wednesday where I ran three one-hour training sessions, a production deployment, two kickoff calls, and was firefighting production bugs in parallel. My gym session dropped. I pushed through to Saturday. I got sick the following week. The cost showed up not in the push itself but in the nine days beforehand, when I was already bracing for a month I had not yet started. The leader sitting at their desk at 8:50am on a Thursday, holding the CFO's soft hold and the vendor's go or no-go at the same time, is in the same physics. They decided, very early, that raising either conversation would trigger a cascade they did not have the capacity to manage. So both stayed internal. And every day that passed made the restart-cost math worse, and made the vendor relationship a little colder, and moved the CFO one day further from the facts they needed to make a clear decision. The one thing I would say to that person, because I would have needed someone to say it to me: the moment you are waiting for is not coming. You are holding this. And you are paying for holding it right now, today, before either conversation has happened. What follows is the operational and financial picture underneath that moment.

What the January roadmap assumed, and why those assumptions are gone

Retail AI roadmaps written in January 2026 assumed a cost environment that has since been revised on every material dimension. Landed costs for sourced merchandise shifted as tariffs continued to cascade through supply chains. The Tax Foundation estimates that in 2026, new tariffs amount to an average household tax increase of $820, down from $1,000 in 2025 only because the Supreme Court vacated certain IEEPA tariffs in February, not because underlying trade policy stabilized (Tax Foundation, 2026). The effective tariff rate reached 10.6 percent in January 2026, and tariff impacts on core PCE inflation peaked in Q1 2026 before partial easing (Dallas Federal Reserve, 2026). For retail operations, the practical consequence is that the cost-per-unit assumptions embedded in vendor contracts, margin models, and technology ROI calculations from six months ago are now materially wrong. Stout's analysis of sourcing practices notes that tariff volatility in 2025 and 2026 has made static unit-cost sourcing models unreliable for modern manufacturing and retail decisions, with leading organizations now required to model three to five discrete tariff scenarios across 12 to 24-month time horizons (Stout, 2026). A roadmap written in January against one set of assumptions is now, effectively, a document from a different company.

How a sourcing crisis reprioritizes which AI problems feel urgent

When landed costs shift suddenly, the process problems that become visible first are inventory and cost modeling problems. Demand forecasting built on last season's cost structures produces wrong margin signals. Replenishment logic optimized before the tariff adjustment may now be recommending buys that destroy margin. Sourcing teams that had weeks to negotiate now have days. This matters for AI roadmaps because the implementation that was prioritized in January, whether that was a customer personalization engine, an order management optimization, or a labor-scheduling tool, may no longer be the most urgent process problem in the building. When tariffs raise landed costs, carrying marginal SKUs becomes harder to justify, and the entire assortment and inventory logic comes under pressure (Kase Retail Dispatch, 2026). The question of which AI problem to solve first has been answered differently by the tariff event than it was in the January planning session. IHL Group's August 2026 retail AI research finds that retail AI budgets are funding the wrong layer first, stalling pilots before they reach operational impact, and that most AI projects that succeed in a clean cloud environment against historical data go quiet the moment they meet an actual store environment, a supply-chain disruption, or a data source that has changed since the pilot was designed (IHL Group, 2026). A tariff-driven sourcing crisis is precisely the kind of real-world condition that exposes that fragility.

What a soft hold on discretionary tech spend actually does to a phase-two decision

The CFO's soft hold on discretionary tech spend arrived as a conversation in a different meeting, communicated sideways, and now lives in the VP's head while the vendor waits for a go or no-go on phase two. Nothing is in writing. The financial math of that pause runs in one direction only. First-year enterprise AI production systems cost $500,000 to $1.5 million all-in, with integration and data preparation, not the model itself, dominating the bill (Teamvoy, 2026). Annual run costs land at 20 to 40 percent of the build cost and must be maintained whether the system is in active use or idle (Launch Day Advisors, 2026). Ongoing monitoring alone runs $30,000 to $100,000 per year, and retraining costs 15 to 25 percent of the initial build annually (Teamvoy, 2026). Pausing a mid-stage implementation crystallizes sunk costs, triggers re-ramp expenses when the work resumes, and may void the continuity pricing built into the current vendor agreement. The SaaS CFO's June 2026 analysis of enterprise AI budget dynamics notes that 2026 AI budgets, most of which were set last October, are already blown at many organizations, and that total IT budgets are growing around 3.5 percent while AI cost demands are growing several times faster (The SaaS CFO, 2026). A soft hold arrived into an environment where the math on the existing project was already tight.

The nine-day silence and what it costs operationally

Back to 8:50am. The sourcing meeting is in ten minutes. The vendor call is in seventy. The two facts that need to surface, the restart-fee math and the CFO's hold, have now been held for nine days. Depletion changes what a leader surfaces and when. Raising either conversation triggers a cascade they do not have the capacity to manage during the same month their sourcing team is in crisis. So both stay internal. The rationalization is that it is not the right moment. The actual mechanism is that every passing day makes both conversations feel heavier, and heavier things get scheduled later, and later means the implementation timeline slips without any formal record of why. I do not know exactly when the decision to wait gets made. I think it gets made before the month starts, not during it. The bracing happens early, and then the full calendar just confirms what was already decided. What I do know is that by day nine, the math is worse in every direction. The vendor may be reallocating team members. The CFO is operating without the restart-cost information that would change their framing. And the person holding both facts is running a sourcing meeting in ten minutes. Implementation timelines have a known failure mode: the 14-month median time from pilot approval to production shutdown for failed generative AI projects is almost always traceable to teams that built without a path-to-production funding commitment, with the project quietly bleeding until it finally stops (Uvik, 2026). Gartner's research finds that through 2026, organizations will abandon 60 percent of AI projects not backed by AI-ready data, and 63 percent of organizations admit they lack, or are unsure they have, the right data management practices for AI (Gartner, 2026). A tariff-induced budget uncertainty landing on top of an already fragile organizational foundation describes the common condition of most mid-stage retail AI implementations right now.

Why the C-suite's risk appetite changed, and what the leader in the middle is reading wrong

KPMG's Q1 2026 tariff survey found that 82 percent of organizations report a decline in foreign sales and 61 percent report a decline in domestic sales, with 68 percent having postponed major investments (KPMG, 2026). When a C-suite is absorbing those numbers, its posture toward discretionary technology investment shifts. The shift runs on sentiment as much as analysis. The VP-level leader holding the implementation reads that shift and interprets it as a signal about their specific project. They are probably right. A C-suite in reactive mode, however, often has not yet thought clearly about what pausing costs versus continuing. Those are two meaningfully different situations. KPMG's research also indicates that the share of businesses passing on more than half of tariff costs to consumers has risen to 34 percent, more than doubling from 13 percent in May 2025, with consumer pushback a growing consequence of that pass-through (KPMG, 2026). The C-suite is managing a revenue problem and a cost problem simultaneously. The AI implementation sits at the intersection of both, whether or not the room has framed it that way. The organizations managing the tariff environment most effectively are the ones using AI to simulate scenarios before committing capital, not after, with KPMG's 2026 supply chain trends research pointing to AI-powered scenario simulators as essential for testing what-if situations before policies take effect (KPMG, via Pull Logic, 2026). In that context, the argument for continuing the implementation is an operations argument about what the next sourcing disruption will cost without better modeling capability.

The conversation that has not been scheduled, and how to schedule it

The CFO conversation and the vendor conversation are not the same conversation, and treating them as one problem is part of what keeps both from happening. Here is how to separate them. First, be honest about why neither has been scheduled yet. The calendar is not actually what prevented it. The calendar is full because that is what calendars do in a month like this. The conversations have not been scheduled because scheduling them means having them, and having them means managing whatever comes next before the sourcing crisis is resolved. That is the real sequence. Naming it matters, because the fix for a timing problem is different from the fix for a capacity problem, and what is actually happening here is closer to the second one. The CFO conversation requires three numbers: the restart fee or re-ramp cost if phase two is delayed beyond the current contract window, the annualized cost of the data and monitoring infrastructure already running, and the projected process cost of the problem the implementation was built to solve if it stays unsolved through Q4. Those three numbers, stated plainly in a five-minute slot, are a materially different conversation than a request for budget approval. The vendor conversation is a timeline and terms conversation. Most vendors in a mid-stage engagement have more flexibility on phasing than they surface in a standard go or no-go request. They want the project to succeed; a client who pauses cleanly and resumes is more valuable than a client who goes quiet. Ask specifically: what does a 60-day phase two delay cost in restart fees or timeline, and what does a scope reduction to a partial phase two look like financially? Get that in writing before the CFO meeting. Forbes notes that tariff pass-through to consumer prices is real, with corporate profit margins at all-time highs, suggesting the costs are being absorbed partly by consumers (Forbes, 2026). The C-suite knows this. A well-framed AI implementation, one that improves cost modeling, demand sensing, or margin management at the SKU level, is an easier approval in that environment than a generic technology investment. The framing matters more than the number. The one ask here: before the sourcing meeting ends today, open a blank document and write down the restart fee number. Just that one number. Everything else can come later. But nine days without that number on paper is nine days of the math running against you invisibly, and that is a cost you are choosing even if it does not feel like a choice.

The under-appreciated point: tariff volatility is the use case, not the obstacle

The category conversation about tariffs and AI treats the tariff environment as a reason to pause technology investment. The evidence points in a different direction. Deloitte's 2026 Manufacturing Outlook finds that leading organizations are deploying AI-driven trade analytics and autonomous agents to continuously assess risk, run scenario plans, and rebalance sourcing networks in near real time (Deloitte, via Pull Logic, 2026). Over 46 percent of supply chains are actively using AI or machine learning today, and only 16 percent report no plans to adopt it in any form (Pull Logic, 2026). Stout's sourcing analysis notes that companies with dynamic sourcing models can respond faster to tariff escalations and reduce operational disruption, with the shift toward tariff-adjusted landed cost reshaping supply networks to prioritize flexibility and geopolitical resilience (Stout, 2026). The tariff environment has not made retail AI roadmaps less relevant. It has made the wrong ones wrong faster, and made the right ones more urgent. An AI implementation built to improve demand forecasting or inventory cost modeling in a stable sourcing environment becomes a different order of value when landed costs can change faster than purchase orders can catch up. The leaders who make that argument to their C-suites are the ones who convert a soft hold into a phased continuation.

What could go wrong

• The vendor contract has a hard go or no-go date

  If the phase-two contract window closes without a decision, the vendor may rightfully treat the project as paused and begin redeploying their team. Restart costs, if not contractually specified, are negotiated from a weaker position than the original terms.

• The CFO meeting happens too late to change the framing

  I have watched this one happen. The formal stop arrives before the leader has surfaced the restart-cost math, and the window to present the business case for continuation closes with it. A soft hold reversed with documentation is a different thing entirely from a formal stop that needs to be undone. The first is a budget conversation. The second is a credibility conversation.

• The implementation timeline slippage goes undocumented

  Each day of informal delay, without a written change to the project timeline or budget, creates a record gap. When the project is reviewed later, the slippage will look like mismanagement rather than a deliberate hold, because no one formally documented the hold.

• The sourcing crisis consumes all available attention before the AI conversation happens

  Sourcing crises are immediate and visible. AI implementation timelines slip quietly. The natural pull is to address what is screaming first and let the quiet thing wait. The quiet thing's cost accumulates on a different ledger, and nobody reads that ledger until the project is already in trouble.

• A soft hold treated as a full stop produces the worst outcome

  Continuing to run infrastructure costs on a system no one is actively advancing, without a clear timeline to resume or a decision to end the contract, is the most expensive version of both choices. A soft hold requires an active decision, in writing, within a defined window.

• The C-suite frames the AI investment as discretionary when the sourcing problem makes it operational

  If the leader does not reframe the implementation from a technology investment into an operations capability for navigating the next tariff event, the C-suite will categorize it alongside other discretionary holds. The reframing is the leader's job. It will not happen on its own.

The J.Caresse point of view

The nine-day silence is something I understand from the inside. I have held information that needed to move in two directions at once, and found that the longer I held it, the heavier it got, and the harder it became to schedule the twenty-minute conversation that would have cleared it. I used to tell myself I was waiting for the right moment. I do not know if I fully believed that. Maybe I was just waiting for the month to feel less full. It never did. What the tariff environment has done is remove even the illusion that a better moment is coming. A sourcing team in crisis and a vendor waiting for a go or no-go are not conditions that resolve themselves while you finish Q3 planning. They run in parallel, and every day they run in parallel without both conversations happening, the options narrow. I have seen the leaders who navigate this version of a month clearly, and what they have in common is a decision they made before the month started: that holding competing information without surfacing it costs more than the uncomfortable ten minutes of putting both conversations on the calendar. Not because they are braver or more organized. Because they have paid the other cost enough times to know what it actually is. I am still working on making that decision earlier. Some months I do. Some months I find myself at 8:50am on a Thursday realizing I have been bracing for nine days instead. I do not have this fully figured out. What I do know is that the bracing starts before the month does, and that is where it has to be interrupted.

Key takeaways

What this brief argues, stated plainly:

• January roadmaps are working from the wrong cost base

  New tariffs have materially affected the cost base underlying most retail AI roadmaps written in January, with KPMG's 2026 research showing 68 percent of organizations having postponed major investments in response to the tariff environment and widespread reporting of higher COGS across retail and consumer goods (KPMG, 2026). The roadmap needs a cost-base update before it can be credibly defended.

• Pausing costs more than most soft holds account for

  Annual AI run costs equal 20 to 40 percent of the original build, and the hidden layer of data prep, monitoring, and change management typically equals or exceeds the build itself (Launch Day Advisors, 2026). A pause without a written timeline and a documented restart cost is a decision with an unknown price tag.

• The CFO conversation and the vendor conversation are two separate meetings

  Treating them as one problem that requires a single moment of courage is what keeps both from happening. They require different information, different framing, and different outcomes. Separate them, schedule them within the same week, and document what was said.

• The tariff environment is a stronger argument for the implementation, framed correctly

  Organizations using AI for scenario-based sourcing and landed-cost modeling are responding faster to tariff escalations than those reacting after costs change (Stout, 2026; Deloitte, via Pull Logic, 2026). The implementation is an operations capability for a volatile sourcing environment. That is a C-suite argument, not a technology argument.

• Tariff uncertainty has constrained AI funding for product and technology leaders across retail

  Sixty percent of product leaders say tariff uncertainty has already constrained AI funding (PYMNTS, 2026). That number means the soft hold is not unique to one organization. It also means the leaders who navigate it clearly, with documented cost comparisons and a phased continuation plan, are operating in a market where most of their peers are standing still.

• The implementation timeline slipping one day at a time is a documented risk, not a background condition

  Gartner finds that 60 percent of AI projects without adequate data foundations will be abandoned through 2026 (Gartner, 2026). Daily timeline slippage without a formal hold or a formal decision to continue is how a project moves from the 40 percent that survive to the 60 percent that do not.