Firmly AI Shopping Funding Backs Merchant Control Over Automated Checkout

Firmly raised $15 million from investors including J.P. Morgan Payments and Mastercard, backing merchant-authorized AI shopping rather than automated checkout without retailer cooperation. The Firmly AI shopping funding supports infrastructure that connects retailers with shopping assistants, plus a new assistant for merchants’ websites. Its central proposition is straightforward: retailers should participate in conversational shopping while retaining control over their stores.

The Seattle-area startup announced the Series A on November 5, 2026. Squarepoint Capital led the round, with Samsung NEXT also participating. JPMorgan and Mastercard backed the investment, but neither led it. Mastercard also has a commercial relationship with Firmly, making its involvement more consequential than a name on an investor list.

The underlying contest concerns how shopping agents reach merchants. Browser agents navigate websites on a shopper’s behalf; authorized integrations connect shopping services with participating retailers’ systems. Firmly favors the latter approach. That puts merchant cooperation at the center of its strategy, while leaving a harder commercial question open: whether connected shopping experiences generate enough completed purchases.

What the Firmly AI shopping funding actually backs

The investment backs merchant connectivity, alongside a new shopping interface for retailers’ own websites.

According to GeekWire’s coverage of the Series A financing, Firmly enables purchases inside external shopping services while retailers retain their payment providers and customer relationships. That arrangement distinguishes its business from simply supplying another chatbot.

The company also launched Ask Agent, an assistant intended to handle product discovery and purchasing conversations on merchant websites. Firmly therefore addresses shopping both outside and inside a retailer’s own storefront.

Those products serve related needs, but their commercial tests differ. An external integration must make a retailer’s products available somewhere customers already shop. An on-site assistant must help existing visitors find suitable products and finish purchases.

The distinction matters when assessing the announcement. Funding a connection platform does not establish demand for every destination it supports. Launching an assistant does not establish that customers prefer conversation over familiar product pages.

For retailers, the appeal is avoiding repeated integration work. A merchant considering another shopping channel must evaluate product information, availability, checkout behavior, support responsibilities, and its existing technology commitments.

An intermediary offers to absorb some of that complexity. The merchant then evaluates the intermediary’s reliability instead of independently maintaining every connection.

That exchange creates a recognizable infrastructure business. Firmly does not need to own the consumer’s preferred assistant to have a role in the transaction. It needs participating merchants and destinations to find its connection useful.

However, integration convenience and commercial importance are different claims. A connection becomes valuable when it supports transactions a merchant wants, under conditions the merchant accepts.

Ask Agent adds another route to proving that value. Instead of waiting for shoppers to arrive through external assistants, a retailer can introduce conversation within its existing website.

That also creates a measurement problem. Results from an on-site assistant should not automatically validate external agent distribution. Each product needs evidence showing where it improves the shopping journey.

Firmly AI shopping funding provides resources to develop both routes. The announcement establishes investor backing and product direction; it does not establish which route will become the larger business.

The immediate change is therefore concrete but bounded. Firmly has fresh capital, named financial industry participants, and an expanded product offering. Its next chapter depends on turning those assets into repeatable merchant outcomes.

Retailers face another distribution decision

The pressure falls first on merchants deciding which shopping channels deserve access to their businesses.

A retailer can welcome additional discovery without welcoming every proposed transaction mechanism. Being recommended by an assistant and allowing that assistant to place orders involve different operational decisions.

For example, a shopper might ask an assistant to find furniture that fits a particular room. A useful recommendation still needs accurate dimensions, available inventory, delivery conditions, and a valid final order.

That example illustrates the integration challenge rather than a verified Firmly customer transaction. The assistant’s fluent answer is only the visible beginning of a purchasing process.

If a retailer builds every connection separately, its teams must maintain channel-specific behavior. If it relies on an intermediary, it must understand what that intermediary handles and what remains its responsibility.

This is where Firmly merchant integration becomes a business decision rather than a technical checkbox. Retailers need to judge how participation affects merchandising, operations, customer service, and their ability to correct mistakes.

Mastercard’s September 9 Agent Connect launch directly addresses that coordination problem. Its announcement describes a single integration connecting participating merchants, agents, commerce services, and payment providers.

Mastercard also identifies Firmly as a partner helping businesses extend product catalogs and shopping experiences. The announcement says merchants can choose participating AI experiences while retaining control over pricing, fulfillment, and business rules.

That establishes a commercial context for Mastercard’s investment. It does not establish a guaranteed flow of customers or transactions to Firmly.

The strategic interpretation is that merchant access has become valuable enough for payment companies to help organize it. Payment authorization alone cannot supply a suitable product, reliable availability, or an acceptable delivery promise.

For merchants, this creates pressure to prepare their information before committing broadly. A poorly specified product remains poorly specified when an assistant reads it.

Conversation can expose those weaknesses quickly. Shoppers ask about compatibility, intended use, delivery timing, and policy exceptions in ways that exceed a simple keyword search.

Retailers evaluating participation should therefore distinguish distribution readiness from catalog readiness. An integration can expose information, but the business must still ensure that information supports purchasing decisions.

There is also an organizational consequence. Responsibility extends beyond the team installing software. Merchandising staff, customer support, payment operations, and fulfillment teams all have reasons to examine the resulting experience.

Firmly’s proposition is attractive if it reduces repeated work across those functions. It becomes less attractive if convenience during installation produces additional investigation after orders arrive.

The funding announcement increases attention on that tradeoff. Merchants gain another credible option for entering AI shopping channels, while facing the same obligation to protect their operating standards.

The near-term response need not be universal adoption. A narrower pilot with clear product scope and measurable outcomes can reveal whether a new channel deserves broader access.

Authorized connections change the checkout bargain

Firmly’s approach exchanges dependence on website navigation for dependence on merchant participation and integration quality.

An AI shopping agent is software that helps select products or act on purchasing instructions. A browser-based version can operate through website interfaces, following a path similar to a human shopper.

Firmly instead describes connections to the systems behind participating storefronts. Its approach relies on merchant agreement rather than treating every accessible page as an available transaction channel.

For readers seeking Firmly shopping agents explained, that architectural distinction is more useful than the image of a chatbot choosing products. The company’s central role concerns access and execution.

Consider a hypothetical clothing purchase. A browser agent might locate a product page, select a size, add the item, and proceed through checkout screens.

An authorized connection can potentially exchange the relevant product and order information through a more structured path. That can reduce dependence on how a storefront arranges its buttons and pages.

However, structure does not eliminate judgment. The system still needs to select the requested size, recognize unavailable variants, and avoid substituting an unsuitable item.

Permission also has limits. A merchant agreeing to a connection does not imply approval for every possible action. A shopper approving a purchase does not authorize every future purchase.

The operational task is keeping those permissions aligned. The assistant must understand the request, the merchant must accept the proposed transaction, and the resulting order must reflect both.

This explains why the competition cannot be reduced to Firmly versus a particular assistant. A consumer-facing service can use different purchasing mechanisms across different merchants.

Nor should browser automation be treated as inherently unauthorized. The relevant distinction is whether a particular interaction respects the permissions and requirements of the parties involved.

Visa provides useful competitive context through its Trusted Agent Protocol. Its documentation describes signed requests that help merchants recognize approved agents and distinguish them from other automated traffic.

Visa’s work addresses agent recognition, while Firmly emphasizes merchant connectivity. These functions occupy related parts of the purchasing journey, rather than proving identical products or interchangeable capabilities.

The comparison highlights a broader engineering point. Identifying an agent, interpreting a customer’s instructions, exposing a catalog, and completing an order are separate tasks.

A successful shopping experience must coordinate them. Completing one task does not establish that the entire transaction is correct.

Firmly merchant integration aims to make that coordination easier for participating retailers. Its business case strengthens if merchants can add destinations without repeatedly rebuilding their operations.

The corresponding constraint is coverage. Authorized connectivity reaches the businesses and experiences that participate. Shoppers’ expectations may extend beyond that available network.

That creates a product design obligation for assistants. They should make clear when recommendations reflect a connected selection rather than an exhaustive view of the market.

The bargain is therefore specific. Merchants gain a more deliberate route into external shopping experiences, while agents gain access conditioned on merchant participation.

For the Firmly AI shopping funding thesis to hold, that controlled access must prove sufficiently useful to both sides.

Ask Agent moves the experiment onto the merchant’s website

Ask Agent gives Firmly an on-site adoption path with a different source of customer traffic.

Firmly’s Ask Agent announcement describes natural-language product discovery, current catalog information, in-chat checkout, and customer support functionality. Those are company-described capabilities, rather than independent findings about performance.

The product changes where the conversation begins. With an external assistant, the retailer seeks access to shoppers using someone else’s service. With an on-site assistant, the visitor has already reached the merchant.

That difference affects how success should be measured. External distribution concerns additional reach and purchasing opportunities. An on-site assistant concerns what existing visitors accomplish after arrival.

A hypothetical shopper looking for a compatible replacement part makes the distinction clear. On a retailer’s website, the assistant could help narrow the catalog by model requirements.

The useful outcome would be an appropriate selection with understandable evidence. A confident answer alone would provide little assurance if the product did not fit.

This creates an evaluation opportunity. Merchants can examine conversations that lead to purchases, questions that remain unresolved, and cases requiring staff intervention.

They should also examine abandoned conversations. An assistant may help some customers while making straightforward purchases feel unnecessarily complicated for others.

The correct comparison depends on the task. A shopper with a precise product identifier may prefer direct navigation. Someone describing a need in ordinary language may benefit from conversational assistance.

Firmly shopping agents explained through this lens become less mysterious. Their value rests on whether they reduce the work between a customer’s request and an acceptable order.

Ask Agent also puts product information under pressure. If shoppers repeatedly ask questions the catalog cannot answer, the retailer learns something about its own information gaps.

That feedback can be valuable independently of automation. It can identify missing compatibility details, ambiguous policy language, or product descriptions that fail to address common purchasing concerns.

However, merchants should avoid confusing learning value with sales value. A useful research channel does not automatically justify unrestricted purchasing authority.

The funding release reports early performance improvements, but those claims require context before supporting general conclusions. Selected outcomes cannot establish what a typical merchant should expect.

A credible evaluation would explain the comparison period, participating visitors, product mix, and treatment of canceled or returned orders. Those details determine whether an apparent improvement survives scrutiny.

The same principle applies to support functionality. Deflecting a question helps only if the customer receives an accurate answer and can escalate when necessary.

For Firmly, the opportunity is to show that shared infrastructure can support distinct shopping surfaces. The challenge is reporting each surface’s results clearly enough that merchants can judge them separately.

Connectivity does not settle intent, reliability, or responsibility

A valid connection and an accepted payment can still produce an unwanted purchase.

That is the strongest limitation on the investment narrative. Firmly addresses an identifiable integration problem, but the quality of delegated shopping depends on more than connectivity.

The first issue is intent. A customer’s request can combine requirements that must all hold together, such as compatibility, delivery timing, material preferences, and an acceptable return policy.

An assistant that satisfies most requirements can still fail the task. Retailers therefore need a way to distinguish firm constraints from preferences that allow alternatives.

The second issue is changing information. A recommendation reflects conditions at a particular moment. Availability, delivery options, or the final order details can change before completion.

An acceptable design must account for those changes. If the proposed purchase materially differs from the approved one, the workflow needs a clear response.

The third issue is responsibility after execution. Customers need to know where to seek help when an agent selects incorrectly, an order changes, or a refund becomes necessary.

These are evaluation concerns, not allegations about documented Firmly failures. The funding announcement does not provide enough evidence to resolve them across all supported shopping experiences.

Even participants in Mastercard’s ecosystem emphasize the gap between completing a transaction and fulfilling a request. Its shopping agent trust discussion highlights authorization, accurate product information, limits, and ways to resolve mistakes.

That discussion is useful evidence of industry participants’ concerns. Because Mastercard has commercial interests in the category, it should not be treated as an independent endorsement.

Independent technical work also separates transaction states that a shopping interface can blur. A commerce architecture study examines authority, checkout validity, changing dependencies, and evidence supporting generated claims.

The study reports results from controlled scenarios and explicitly limits their implications. It does not establish production security, legal compliance, or population-wide error rates.

The practical implication is narrower: merchants should ask how a system responds when the facts supporting an action change. They should also ask what evidence explains its final decision.

Those questions apply regardless of the assistant’s presentation. A friendly conversation can make a complicated purchase feel simple while leaving the underlying responsibilities unresolved.

Firmly’s merchant-authorized model provides a clearer starting point for cooperation. It does not, by itself, answer every question about authorization scope or customer expectations.

There is a commercial uncertainty as well. Retailers need useful demand from connected destinations; those destinations need sufficient merchant participation to satisfy shoppers.

If either side remains thin, integration quality alone cannot create a compelling network. The company must demonstrate that connections support recurring activity rather than isolated demonstrations.

Investor participation helps fund that effort. It should not be substituted for evidence of completed orders, manageable support requirements, or customers returning to use the experience.

The central risk is therefore a mismatch between infrastructure progress and purchasing adoption. Firmly can make connection easier while shoppers and merchants remain selective about delegating the transaction itself.

Watch deployment breadth, completed orders, and recovery from mistakes

The next evidence should show repeatable use across merchants, followed by reliable transactions and workable exception handling.

Over the coming months, the first signal is deployment breadth. Look for clearly identified live merchant experiences and shopping destinations, with enough detail to understand what each supports.

A partner announcement is useful context, but a functioning purchasing experience carries more information. It reveals available products, supported actions, and the boundaries customers encounter.

This signal would strengthen the investment thesis if participation expands through repeatable implementations. It would weaken the thesis if most visible activity remains concentrated in announcements and demonstrations.

The second signal is completed purchasing behavior. Retailers need evidence that customers move beyond recommendations and finish orders they actually want to keep.

Useful reporting would distinguish discovery activity from checkout attempts and completed orders. It would also explain whether observed improvements persist beyond an initial launch period.

Repeat use matters because novelty can generate attention without establishing a habit. A shopping channel becomes more credible when customers return after experiencing its full purchasing process.

For Ask Agent, the relevant comparison is the merchant’s existing customer journey. For external destinations, the question is whether the connection brings valuable purchasing activity.

Those results should remain separate. Combining them into a broad engagement figure would obscure which part of the business is working.

The third signal is recovery from mistakes. Look for clear handling of unavailable products, changed order details, unsuccessful payments, cancellations, and customer disputes.

These events expose operational quality that a successful demonstration cannot show. A system earns confidence when it recognizes a problem and gives the customer an understandable next step.

Strong exception handling would reinforce Firmly’s merchant-control argument. Confusing handoffs or unclear responsibility would weaken it, even if the initial recommendation experience looks polished.

None of these signals requires assuming that consumers will delegate every purchase. Adoption can remain selective while still creating a useful business.

A customer might welcome help finding a replacement item while retaining control over an unfamiliar purchase. Merchants can likewise support particular destinations without treating all agent traffic equally.

Firmly AI shopping funding supports an effort to make those choices practical. The important question is whether its infrastructure makes participation easier without making outcomes harder to understand.

If you evaluate AI shopping software, follow a complete purchasing journey before accepting a broad adoption claim. Examine how the assistant interprets requirements, confirms the final order, and handles a change after approval. Ask which merchant and destination supplied each piece of information. Keep product demonstrations, company performance claims, and observed results distinct in your assessment. For teams comparing vendors, a shared searchable knowledge base can preserve those findings alongside technical documentation. Start with a bounded use case and review the resulting orders with operations and support colleagues. Then ask the decision that matters: did the connection produce a better purchasing outcome that customers and merchants both want to repeat?