Innovation & Tech

Accio cut Parcel Air’s sourcing quote from 3 weeks to 12 hours

Accio went through 68 suppliers for Parcel Air in 3 minutes, 17 seconds. Adoption costs—not token fees—remain the hurdle for small firms, an Alibaba case study shows.

By Grace Kim

5 min read

Affordable AI takes more than a lower price
Affordable AI takes more than a lower pricePexels / pixabay

What's News

  • Elizabeth Field DiGiovanni said Accio went through 68 suppliers in 3 minutes and 17 seconds, turning a quote that took more than three weeks into a 12-hour result.
  • An August Goldman Sachs survey found 86% of alumni of its 10,000 Small Businesses program used AI, but only 15% fully integrated it into core operations.
  • Census Bureau data from December 2025 to May 2026 showed AI use below 20% among U.S. employer businesses with four or fewer employees, versus 37% among those with at least 250.
  • Stanford’s 2025 AI Index put GPT-3.5-level query costs at $20 per million tokens in November 2022 and $0.07 in October 2024.
  • Zhang said Accio’s token cost fell by half in four months while consumption grew tenfold.

By Grace Kim

Elizabeth Field DiGiovanni said Accio went through 68 suppliers in 3 minutes and 17 seconds on her first proper run for Parcel Air, turning a sourcing quote that had taken more than three weeks into a 12-hour result.

The case study was shared with Kuo Zhang, president of Global B2B Business at Alibaba Ecommerce Group. Zhang uses it to argue that AI affordability depends on the work of deploying the technology.

What did lower model prices leave unresolved?

Zhang’s case study describes Field DiGiovanni running Parcel Air from Los Angeles. The company makes travel activity boxes that keep young children occupied on flights. The boxes include toys, crayons and packaging sourced largely from Asian factories, where work starts around her bedtime.

In Zhang’s account, her sourcing chain runs in sequence: finding factories leads to quoting them; a minimum order quantity changes which price break to request, and a certification answer determines whether lead time matters. With the time difference, each link cost her a day. One vendor negotiation took more than three weeks to produce a quote.

Before using Accio, Field DiGiovanni expected a week of setup she could not spare: choosing among tools and connecting her catalog and supplier list. She told Zhang that running Parcel Air while minding her 4-year-old and 2-year-old left little free time.

Zhang’s point is that affordability includes the total cost of adoption, not just subscription or token prices. Someone must still configure tools, frame business questions and check results. Larger companies can split that work across teams; Parcel Air has no employees, so Field DiGiovanni does it all.

  • A Goldman Sachs survey conducted in August found that 86% of surveyed alumni of its 10,000 Small Businesses program used AI, but only 15% fully integrated it into core operations. Zhang notes that these owners completed a business education program and generally sit ahead of small firms.
  • Census Bureau data covering December 2025 through May 2026 showed AI use below 20% among U.S. employer businesses with four or fewer employees, compared with 37% among businesses with at least 250 employees.
  • Stanford’s 2025 AI Index tracked the cost of querying a GPT-3.5-level model from $20 per million tokens in November 2022 to $0.07 in October 2024. Zhang said Accio’s token cost fell by half in four months while consumption grew tenfold.

Zhang’s conclusion is straightforward: cheaper intelligence has not removed the labor required to deploy it.

What must vendors lower to make AI affordable?\n

Zhang lays out three cost reductions for vendors, including Alibaba Ecommerce Group.

Can technical setup become plug-and-play?

Zhang said owners should be able to start with a sourcing job without learning model selection, prompt engineering or software integration. Without dedicated IT, small businesses have little capacity to assemble and maintain AI systems.

Pre-configured expert agents and sourcing plugins do that work in advance, reducing configuration, custom development and integration for owners who would otherwise handle it.

For Zhang, owners should not have to become the IT department to launch a useful workflow.

Can agents close the expertise gap?

Zhang said e-commerce agents need technical simplicity and practical knowledge of sourcing, trade and operations. A tariff changes landed cost; a missing certification can block market entry; supplier screening and international logistics demand expertise that owners without procurement teams may lack.

Developers can embed that knowledge in the owner’s workflow. The system should show what needs checking, flag missing information and identify claims requiring documentation.

Zhang’s point is that the agent must know the trade, not merely complete a software task.

Can owners trust autonomous agents?

Zhang warned that constant supervision can create as much work as an agent removes, while unlimited authority creates risks a small owner may not be able to afford.

Owners need clear permissions, a visible record of actions and the ability to inspect recommendations and intervene, he said. Routine supplier research and quote organization can run independently once the workflow proves reliable; orders, payments and contractual commitments require explicit human approval.

Those controls belong inside the product. The goal is less routine supervision while owners retain control over consequential decisions.

What changed for Parcel Air?

In the case study, Field DiGiovanni sources through Alibaba.com and uses Accio, Alibaba’s AI agent for small businesses and solo founders. She briefs it like a colleague, specifying the product, target cost, safety certifications and quantity.

In her account, Accio’s agents work with suppliers through the business day in Asia while she sleeps. Field DiGiovanni reviews the results in the morning before her children wake and then makes the call.

She described it as “having an entire APAC team on the night shift.”

Field DiGiovanni still sets the target cost, chooses which supplier to trust and requests a document whenever a certification claim lacks evidence. The speed changed the waiting, not the limits of her judgment: she no longer loses a day to every question and gets her evenings back.

In her workflow, supplier research moved off her evenings while high-stakes decisions stayed with her.

How should vendors measure affordability?

Zhang says vendors should answer two questions before treating AI as affordable for small businesses.

  • How much time does an owner need to reach a useful result, including setup and learning?
  • How much time does the tool save on recurring tasks after setup, review and upkeep?

These measures show whether a product removes enough work to justify adoption. Field DiGiovanni’s experience favors the test because the tool must fit a busy workweek.

She put it plainly: “You don’t need a bigger team. You need the night shift covered.”

Zhang argues that small businesses should be able to afford to put AI to work. His forward test is straightforward: vendors must lower setup and review time without shifting hidden labor back to owners or weakening approval over orders, payments and contracts.

Original: goldmansachs.com

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,582 articles

Related articles

« Previous article