What Maya does
Most small business owners do not lack data. They lack time to ask it questions. Maya is built to answer those questions in plain language, in seconds, from your own books: which products are selling faster than last month, which customers are overdue, how much cash will be left after Friday’s payroll, what you are about to run out of.
Maya can send a morning briefing by email or SMS so the day starts with what changed overnight. And she can draft work: purchase orders for items running low, suggested reorder points, and promotions for slow-moving stock. Today she works through 31 tools over your own data, and voice works in the browser.
Why she asks permission
An AI that can draft a purchase order could, in principle, send one. We decided early that Maya would not. Every action Maya proposes goes to an approval gate, where a person with the right permissions reviews it and approves or rejects it. The number we hold ourselves to is zero actions without approval.
This matters because the cost of a mistake in a business system is real. A wrong purchase order commits money. A wrong price reaches customers. An AI that is right most of the time is useful as a drafter and dangerous as an actor. Maya drafts; you decide.
The safety controls, in detail
Approval is the most visible control, but not the only one. Every action Maya takes is recorded in an append-only audit trail, so you can see what she proposed, who approved it and what happened. Approved actions can be undone. Value caps and rate limits stop her from proposing anything over a limit you set, and high-value actions can require step-up authentication or two-person approval.
Maya also respects roles. She only discloses what the person asking is allowed to see, so a store manager asking about sales does not get company-wide payroll figures. And there is one kill switch that turns her off for the whole business instantly.
- Approval gate on every action
- Append-only audit trail; 100% of actions audited
- Undo for approved actions
- Value caps, rate limits, step-up and two-person approval
- Role-aware disclosure and consent gating
- One kill switch
Forecasting with honest uncertainty
When Maya forecasts demand, she uses models trained in Azure Machine Learning and shows a confidence range, not a single number, along with plain-English drivers: seasonality, a recent promotion, a trend in a product category. A forecast that says “between 40 and 55 units, mostly because of last December” is more useful than one that pretends to know the answer is 47.
Your data stays yours. Customer data is not used to train AI models.
A morning with Maya
Here is what Maya is designed to do on an ordinary day. At 7 a.m., the owner of a two-location retail business gets a briefing by SMS: yesterday’s sales by location compared with the same weekday last month, three products that will run out within their supplier’s lead time, two customer invoices that became overdue overnight, and the projected bank balance after Friday’s supplier payments.
Over coffee, the owner asks Maya why sales at the second location are down. Maya answers from the books: fewer transactions rather than smaller baskets, concentrated in weekday afternoons, and starting about three weeks ago. She does not invent causes the data cannot support; the owner knows a nearby road closure began that week.
The owner asks Maya to prepare reorders for the three low-stock products. Maya drafts purchase orders using each supplier’s usual quantities and the forecast range, and sends them for approval. The owner adjusts one quantity, approves the other two, and the approvals and the change are recorded in the audit trail.
What Maya will not do
Maya will not send a purchase order, change a price, pay a bill or contact a customer without approval. She will not show a user information their role does not allow them to see. She will not use your business’s data to train AI models. And when an administrator uses the kill switch, she stops entirely.
Those limits are not temporary restrictions waiting to be removed as the technology improves. They are the design. We think every business should be able to use AI the way it would use a capable new hire: give them access to the right information, let them prepare work, and keep the decisions with the people who are accountable for them.
How to start with Maya
Maya is available in SeroBooks today, and the tools she can use keep growing as we learn from how businesses use her. The approval gate is there for the same reason a new hire’s first purchase orders get a second look: trust is earned one approved draft at a time.
If you want to try Maya, start with questions. Ask about your sales, your stock and your cash. When she drafts something, read it carefully before approving. That is how we expect every AI employee to work: useful, accountable and never in charge.