Every commerce organization runs on questions that begin with “why.” Why did this promotion apply to that order? Why did this shipment route to a distribution center three states away when a closer store showed stock? Why did a loyal customer’s order arrive in two boxes, two days apart?
The answers exist. But they are scattered across engine logs, admin screens, and the memory of whoever configured the rule six months ago. Getting a single answer typically costs a support ticket, a screen-share session, or an email to your implementation partner. By the time it arrives, the moment to act on it has often passed.
This is the problem the Explain function of KIBO AI was built to solve. It gives every person who touches your commerce operations (marketers, fulfillment managers, customer service reps, developers) a direct, plain-language answer to “why did this happen?”, grounded in what your platform actually did. And because Explain never changes anything, it is the most practical first step a cautious organization can take into agentic commerce.
What Is the Explain Function in KIBO AI?
Explain is one of five agentic functions in KIBO AI, each designed around a distinct kind of work:
- Engage: conversational AI for shoppers, B2B buyers, and service reps, anchored to live commerce data.
- Configure: stages changes to promotions, content, and settings, always left for a human to review and save.
- Explain: answers “why did this happen?” by investigating decisions the platform has already made.
- Analyze: conversational reporting: ask a question, get a chart or table.
- Optimize: models the impact of possible changes and recommends a better state.
Explain’s job is accountability. Instead of interpreting logs or hunting through configuration screens, you ask a question in natural language inside the Kibo Admin experience (with developer-facing capabilities in the Developer Center), and the agent traces what happened, then explains it in plain terms. It is the function that turns an opaque system into one you can interrogate.
How Explain Works: Five Design Principles
Explain was engineered so that operations teams, support engineers, partners, and client developers can trust what it tells them. Five principles define it:
- Read-only by design
Explain never creates, updates, or deletes data. It retrieves and explains, full stop. There is no approval workflow to build and no damage it can do. - Grounded in what actually happened
Explanations for specific transactions (a promotion applying, an order routing, a return routing) are drawn from the engines’ real runtime logs and decision traces. The agent does not re-execute business logic or simulate what should have happened. It reports what did. - Permission-inherited
Each user sees exactly what their own role already allows in the Kibo Admin or Developer Center. The agent opens no new path to data. - Honest about its limits
When logs fall outside the retention window, the agent says so clearly and suggests an alternative, such as querying summary-level transactional data, instead of guessing. - Conversational
Context carries across a session. Ask “why was order 100503 routed to DC-EAST?” and then “what else was considered?”, and the agent remembers the order, the locations, and the thread of the investigation.
Where Explain Gets Its Answers
An explanation is only as good as its sources. Explain reads across the same ground your best operations analyst would cover:
- Decision logs from the promotion, order routing, and returns routing engines: step-by-step traces of which conditions passed, which locations were scored, and which path was taken.
- Transactional commerce data: orders, items, shipments, inventory, and locations.
- Your tenant’s configuration: routing strategies, promotion rules, fulfillment and inventory settings.
- Reporting and analytics: standard and custom reports across orders, discounts, fulfillment SLAs, payments, and returns.
- Kibo product and API documentation: “how does this feature work?” questions are answered from official documentation, not folklore.
- Platform request logs: API calls, webhook deliveries, workflow transitions, and errors, traceable by correlation ID.
What Explain Looks Like in Practice
The fastest way to understand Explain is to see the questions real roles ask it.
The marketing manager: “Why was promotion BOGO-SHOES applied to order 55123?” The agent reads the promotion engine’s evaluation trace: the cart contained three shoe items, satisfying the “buy 2+ shoes” condition, the promotion was active, and the lowest-priced item (SKU-2290, $49.99) received the 100% discount. Mystery solved in one prompt, with no ticket to the promotions team.
The fulfillment operations manager: “Why was order 100503 routed to DC-EAST, and what else was considered?” The routing log shows DC-EAST won on inventory availability and distance. The Chicago DC scored lower on distance; Store-042 was missing one of three items. The follow-up question (“what else was considered?”) is where Explain earns its keep: alternatives and their scores are part of the trace, not tribal knowledge.
The customer service rep: “A customer is asking why order 100880 arrived in two separate boxes.” Explain walks the split: item A shipped from the Dallas DC because it held the only stock; item B came from Chicago. Two boxes, two tracking numbers. The CSR answers on the first contact instead of opening an investigation.
The returns manager: “Which items on order 100234 are still eligible to be returned?” Two of three items remain within the return window; the gift card is non-returnable. And because Explain is read-only, looking up eligibility never accidentally creates a return.
The inventory manager: “Why is SKU-1234 showing zero available-to-sell at the Austin store?” Eight units on hand, but five are reserved as safety stock and three are allocated to open orders, so net ATS is zero. Now the manager knows exactly which lever (safety stock) would free units.
The developer: “My order.created webhook isn’t firing. What’s wrong?” The subscription is active and posting, but 47 deliveries failed in the last 24 hours, all HTTP 502s from the merchant’s own endpoint. Kibo’s side is fine; the fix is on the receiving server. A correlation ID can trace an entire transaction the same way: cart add (200), fulfillment set (200), submit failed (500, gateway timeout), payment.failed emitted with no retry.
Notice the pattern across all six: the answer is specific, evidence-backed, and immediate. That is what explainable AI order management looks like inside a working order management system: not a dashboard you interpret, but an answer you can act on.
Why Explain Is the Right First Step Into Agentic Commerce
For executives weighing AI in ecommerce, the hardest question is not whether agents are useful; it is where to start without taking on operational risk. Explain resolves that.
There is nothing to break
Because Explain is read-only, adoption requires no guardrail debates, no UAT cycles for agent actions, and no change-management program. The worst-case outcome of any prompt is a well-sourced answer.
You pay only for the function you use
KIBO AI is licensed by function, so a merchant can adopt Explain on its own, without committing to the full agentic suite, and expand when the organization is ready.
It returns hours to your team
Every “why” question that used to become a ticket, a log search, or a partner engagement now takes seconds. Multiply that by every CSR, marketer, and ops manager asking these questions weekly, and the productivity case writes itself.
It builds the trust agentic commerce runs on
Explain produces the evidence trail that compliance and audit conversations demand: which promotions fired, which locations were scored, which deliveries failed. Teams that learn to trust explanations today are the teams ready to trust Configure’s staged changes and Optimize’s recommendations tomorrow.
Frequently Asked Questions
Does Explain ever change data in my tenant?
No. Explain is strictly read-only. To act on what you learn, you use the corresponding Configure capabilities, which stage changes for human review.
What happens if the logs are older than the retention window?
The agent tells you the data is outside the retention window and offers an alternative, such as summary-level transactional data, rather than filling the gap with a guess.
Who can use Explain?
Anyone working in the Kibo Admin experience, with developer-facing capabilities in the Developer Center. The agent inherits each user’s existing permissions, so it only surfaces data that user is already authorized to see.
Do I need to adopt all of KIBO AI to use Explain?
No. KIBO AI is licensed by function, so you can start with Explain alone and add Engage, Configure, Analyze, or Optimize when it makes sense.
The question “why did this happen?” is where trust in automation begins. Explain answers it with evidence drawn from real decision logs, your own configuration, and your transactional data, without ever touching a record. For organizations that want the productivity of agentic commerce without the perceived risk, explainability is not a compromise. It is the on-ramp.
Ready to see it on your own data? Request a demo of KIBO AI, or read how Explain fits alongside the other five agentic functions that transform commerce operations. For the full capability list, see the Explain with Kibo AI documentation.
