Yes, for low-risk merchants within a risk appetite the payments company has already written down. An AI agent can verify the business, review its website, assign a category and starting limits, and activate the account at depth 3. Higher-risk categories, future-delivery businesses, ownership gaps and anything outside the written appetite go to a named person on the risk team.
Why can a payments company let an agent approve, when a bank account often cannot?
The difference is exposure. A bank that opens a deposit account takes on a relationship with few built-in limits. A payments company that onboards a merchant can cap its exposure from day one: low processing limits, delayed payouts, a rolling reserve, and the right to hold settlement if things go wrong. If the agent approves a merchant that turns out badly, the loss is bounded by those settings.
That changes what can safely be automated. Where the policy says "small online retailers of everyday goods, with this profile, start at this monthly limit with payouts after this delay", applying it is execution of a written rule. The judgement was made when the policy was approved. For comparison, see whether an AI agent can onboard a business customer, where acceptance stays with a person.
What conditions decide it?
The agent may activate a merchant when:
- identity and ownership checks pass, with every owner and director identified and screened;
- the business category is on the low-risk list, and the website, products and pricing match it;
- the merchant does not sell anything on the prohibited or restricted lists;
- delivery is immediate or near-immediate, so customers are not paying today for something that arrives months later;
- expected volumes sit within the starting limits for the category;
- the refund policy, terms and contact details are published and plain.
Future delivery deserves its own mention. Airlines, event organisers, furniture makers and subscription businesses taking annual payments create exposure that grows quietly. If the merchant fails before delivering, the customers charge back, and the payments company pays.
What may the agent do, and what stays with a person?
| Step | The agent | Stays with a person |
|---|---|---|
| Identity, ownership and screening | Runs and records checks | Possible matches and unclear ownership |
| Website and product review | Compares the site with the declared category | Borderline or restricted products |
| Assigning a merchant category | Proposes from the business evidence | Disputed or mixed categories |
| Setting starting limits and payout delay | Applies the policy table, depth 3 | Anything above the table |
| Activating a low-risk merchant | Activates at depth 3 | Nothing routine |
| Future-delivery or high-volume merchants | Prepares the file | Approval and reserve |
| Declining | Prepares the reasons | A person decides |
| Raising limits later | Prepares the review | Approving the increase |
An example: two applications on a Tuesday
Take a payments company that receives two applications in an hour. The first is a sole trader selling handmade candles online, with a simple shop, clear prices, a returns policy, and expected sales of a few thousand a month. The agent verifies the trader, reviews the site, assigns the category, applies the starting limit and the standard payout delay, and activates the account. The merchant can take payments that afternoon.
The second is a small company selling "wellness products" with a declared category of health and beauty. The website includes supplements making medical claims and a subscription that bills annually in advance. The agent records both concerns, sets no limits, and sends the file to the risk team. A person decides whether to accept the business, with what reserve, and whether some products must come off the site first.
What controls make it safe?
- The risk appetite as sealed policy. Low-risk categories, prohibited lists and starting limits come from the policy library, not from the agent's judgement.
- Website evidence captured. The agent stores what the site showed at onboarding, so later changes can be compared.
- Limits as a bound. Activation always comes with the policy's starting limits; the agent cannot activate without them.
- Instant payment exposure considered. Where merchants accept instant payments, refunds and recalls are harder; see instant payments and irreversibility controls for AI agents.
- Post-activation monitoring. Early volume and chargeback patterns are watched, and the agent flags merchants that drift from their profile.
Merchants accepting account-to-account payments face different risks from card merchants, because instant payments settle in seconds and carry no card-style chargeback. The general rules for regulated payments work are in AI agents in banking: what they may and may not do.
Where Pryme fits
In Pryme, activating a merchant is a capability certified on the payments company's own past applications, including the ones its risk team declined. The scope gate tests that every future-delivery or restricted-product file is held. The starting limits are the customer's own bound in the minimum chain, so the agent cannot activate a merchant above them.
Questions people ask
Can the agent onboard sub-merchants for a platform?
Yes, under the same rules, as long as the platform's own agreement allows it and each sub-merchant is checked individually. Platforms often hide higher-risk sellers among many small ones.
What if a merchant's website changes after onboarding?
The agent can re-review sites on a schedule and flag changes in products or pricing. Changes into restricted categories go to a person.
Can the agent set a merchant's processing limits?
It applies starting limits from the policy table. Raising them later is a separate decision with its own evidence.