Small importers should automate the sourcing decisions they can undo, and keep human control of the ones they cannot.
If a small importer’s AI ranks suppliers badly, they rerun the analysis with better inputs. No harm done. But if it clears a shipment that turns out to be bad, with wrong labels or failed compliance for example, they’ve potentially lost the customer. Whether a decision can be undone is the line small importers need to draw before letting an AI agent near their sourcing.
Agentic procurement tools reach small businesses with a simple pitch. Connect an account, set a budget threshold, and let the software source for you. But that pitch skips the real question: For a company that can hardly afford one bad order, which parts of sourcing should run unsupervised?
Sourcing has always been an operational risk. The dividing line should be the “reversibility rule;” the test of whether a decision can be taken back.
These AI systems are a huge leap in autonomous software. They read context, plan multi-step processes, and execute without a human triggering each step, moving from flagging a shortfall straight to issuing a purchase order. Task-specific agents are expected to show up in 40% of enterprise applications by 2026, up from under 5% the year before.
Automating that layer can make procurement functions 25 to 40% more efficient, freeing teams for decisions that need human judgment. To a small importer, that gain matters more. Since there’s no team to free, it’s the owner who quotes freight and reconciles invoices while automation gives them that time back to be used for pricing, product, and the calls only they can make.
But these tools have a limit. They handle speed, price and availability well, but they cannot confirm factory conditions, product quality, packaging integrity or shipment compliance. Those are realities only a person can verify (at least for now). For a small importer, a rejected shipment or a lost account can cause irreversible damage to the business.
Picture a small U.S. retailer placing its first large order for children’s furniture. An AI agent finds the cheapest supplier, checks lead times, confirms the production run is done, and recommends releasing the balance payment. Then the goods arrive with the wrong warning labels, missing paperwork, or packaging that fails in transit, and customs holds the container while the savings stop mattering. The importer faces delays, rework and inventory it cannot sell. Redoing that ranking would have taken only an afternoon, but releasing the goods could not.
Without accountability built into the system, autonomy adds fragility instead of advantage. Systems that decide at machine speed can weigh price too heavily against quality, or miss the contextual read an experienced sourcing manager catches on instinct. By Gartner’s estimate, over 40% of agentic AI projects will be scrapped before 2028 arrives, largely due to ballooning costs, murky ROI, and risk controls that never got built out. The reversibility rule gives small importers a working answer: Hand the agent whatever analysis can be redone, and keep a human gate on whatever can’t be.
An AI agent is a good fit for finding and ranking suppliers. It can score vendors on reliability, lead times, certifications, and risk exposure, and update those scores as conditions change. Cost and tariff modelling is another strong fit, testing order sizes against cash position and duty costs across countries, which matters most when tariffs are squeezing margins. Round out the list with bid collection, invoice checks and reorders from approved suppliers.
But verification must stay with a person. An AI agent can confirm a factory finished a production run, but only a person can confirm the goods meet specification, packaging and labelling rules, which is why factory vetting and pre-shipment sign-off should stay mandatory. The AI’s supplier scores have the same blind spot. They’re only as reliable as the data behind them, so pair them with an independent inspection instead of trusting the score alone.
The last piece is procedural: Define which decisions the AI agent executes on its own, which trigger a human review, and which stay human-owned, then log every action it takes. Without that, no one owns the call when an automated decision goes wrong. Start agents on structured, low-stakes tasks, and widen the boundaries as confidence builds. Never hand a first-time order from an unvalidated supplier to a machine.
A procurement bot that buys faster is not the same as a system that sources smarter. Importers who win with these tools are the ones who combine AI speed with human judgment at the few moments that decide if an order goes right or wrong.
The winners will be the operators who know which decisions to hand off to AI, which to keep for themselves, and those who never let AI make an irreversible call on its own.
Ran Leitman is chief revenue officer of Ship4wd.