Alisa Ops is the operational brain of an e-commerce business — it knows every rule, applies them to live orders, and handles routine decisions so that people only deal with the ones that actually need a person.
For operators, the system removes the guesswork. Instead of messaging the manager to ask how to handle a COD order or a return request, the operator sees what to do, why, and what the system considered. Training time drops. Mistakes from misremembered rules stop happening.
For the manager, the system captures knowledge once and applies it consistently. No more answering the same question from different people. When a rule changes, it changes everywhere — not in one person's head while others keep following the old version.
For the CEO, the business has an auditable, browsable set of policies that actually govern operations. Growth no longer requires linearly hiring more people. The ratio of orders processed per human hour improves continuously.
For other e-commerce businesses (eventually), Alisa Ops is a product they adopt to solve the same problem: structuring operational knowledge, automating the routine decisions, and keeping humans on the genuinely hard ones.
E-commerce operations are full of decisions that are too complex for simple automation but too routine for skilled humans. When to request prepayment, how to verify a COD order, which returns to accept, how to handle an edge case with a repeat customer — these decisions follow patterns, but the patterns live in people's heads.
At Alisa.ua, this knowledge is scattered across 1,100+ Planfix items — half outdated, unstructured, mixing hard rules with vague guidelines and tool instructions. Operators ask the manager. The manager answers from memory. The answer doesn't get recorded. The CEO can't see what policies actually govern the business. The business scales by hiring more people and hoping they absorb enough tribal knowledge fast enough.
Until now, there was no way to automate these decisions because they required judgment — reading context, weighing exceptions, interpreting messy situations. Simple if-then rules couldn't handle it, and the decision-making layer that could didn't exist.
LLMs change what's possible. They can be the decision-making layer that didn't exist before — reading an order, checking customer history, applying a rule with exceptions, and recommending an action. Not perfectly, but well enough to handle the routine 80% and surface the interesting 20% for humans.
We're building the system that makes this real for Alisa.ua, then for other e-commerce businesses.
Import the existing 548 knowledge base items from Planfix. Use an AI agent in Slack to interview the manager, clarify vague instructions, classify each item, and turn them into structured, approved rules. Build a web UI where the manager and CEO can browse the result.
This is the foundation. Nothing else works if the rules aren't clean.
Connect to Odoo, Shopify, and other systems. When orders come in, the system reads the context and applies the rules — recommending actions to operators. Operators see what to do, why, and what the system considered. They approve, override, or escalate.
The agent in Slack stays as the conversational interface for anyone to ask about policy.
Two feedback loops. Operators flag bad recommendations with structured reasons. The system also detects when operators consistently ignore recommendations — and surfaces those patterns to the manager. Rules get updated. The system gets better.
Gradually let the system act on its own for decisions that are clear, reversible, and low-risk. Add notes, apply tags, send template messages, move orders through safe states. Keep humans on anything ambiguous, high-stakes, or irreversible.
As more processes get automated, operators handle fewer routine tasks. The remaining human work is the genuinely complex stuff — the exceptions, the judgment calls, the conversations that actually need a person.
With less routine work left for humans, the operator UI gets smaller, not bigger. Instead of operators navigating Odoo's full interface for every order, they work in a focused, purpose-built UI that only shows what still needs a human — connected to Odoo underneath, but stripped down to the decisions that matter.
The end state: fewer people doing less repetitive work through simpler interfaces, with the system handling the predictable operations in the background.
At this point, we've proven the concept on a real e-commerce business. The structured rules, the agent-based decision layer, the automation, the simplified operator UI — all of it works on Alisa.ua's actual operations.
That becomes the product: an e-commerce operations brain — a system that any e-commerce business can adopt to structure their operational knowledge, automate routine decisions, and simplify the human side of their operations.
-
Structured knowledge base. All operational instructions are imported, classified, and maintained as versioned rules with clear ownership, status, and scope. The web UI makes them browsable and searchable by anyone in the company.
-
AI-assisted rule creation. A Slack-based agent interviews the domain expert, extracts structured rules from conversation, and submits them for approval. New knowledge enters the system through conversation, not form-filling.
-
Decision recommendations. The system reads order context from connected platforms (Odoo, Shopify), matches applicable rules, and recommends actions to operators. Each recommendation shows its reasoning and the rules it applied.
-
Feedback loops. Operators flag bad recommendations with structured reasons. The system detects patterns where recommendations are consistently overridden. Both signals surface to the manager as rule improvement candidates.
-
Autonomous execution. For decisions that are clear, reversible, and low-risk, the system acts on its own — adding notes, applying tags, sending template messages, advancing orders through safe states.
-
Focused operator interface. A purpose-built UI shows only what still needs a human decision. Connected to Odoo underneath, but stripped down to the work that matters.
These are limitations that exist today and no longer exist in the envisioned future:
-
Knowledge is trapped in people's heads. Today, operational rules live in the manager's memory and in 1,100 unstructured Planfix items. In the future, they live in a single structured system that anyone can browse and search.
-
Every decision requires a human. Today, even routine, predictable decisions (verify COD, accept a standard return) go through an operator. In the future, the system handles the predictable 80% autonomously.
-
Scaling means hiring. Today, more orders = more operators. In the future, more orders = more automated decisions, with human effort growing sublinearly.
-
Rules drift silently. Today, when a policy changes, some people hear about it and some don't. In the future, a rule change propagates instantly to every decision the system makes.
-
No visibility into what governs the business. Today, the CEO can't see the full set of operational policies. In the future, the rules UI is the single source of truth.
These are new limitations we expect to face in the envisioned future — ideally ones that are easier to manage than the current ones:
-
Rule quality becomes the bottleneck. When the system applies rules automatically, a bad rule causes damage at scale instead of one-off mistakes. Rule review and approval processes carry more weight.
-
Edge cases concentrate. As routine decisions get automated, the remaining human work is disproportionately hard. Operators need better judgment, not less — but for fewer situations.
-
Integration maintenance. Each connected platform (Odoo, Shopify, payment providers) is a dependency. API changes, schema migrations, and platform updates require ongoing attention.
-
Trust calibration. The business needs to continuously decide which decisions the system can handle alone and which still need a human. Getting this boundary wrong in either direction has costs — too cautious wastes human time, too aggressive causes errors.
-
Multi-tenant complexity. When Alisa Ops becomes a product for other businesses, each customer brings different platforms, rule structures, and operational patterns. The system needs to be flexible without becoming unmanageably configurable.
- PRD v1 — current implementation plan, schema, modules, user stories
- Roadmap — deferred features and future work
- Planfix data analysis — source data structure and counts
- GitHub Issues — active implementation tracking
- Deployed app:
alisa-ops-stanvoos-projects.vercel.app