Step 3 · Build & Validate

Pilot, PoC, and verification methods of Robo Claw in retail and e-commerce startups

Once the requirements are finalized, we will build a pilot with one sales channel, one product category, one operation, and one system connection, verify both normal and abnormal scenarios, and then decide whether to proceed to production.

Conclusion

It is important to narrow the scope of the pilot. Limit it to one sales channel, one product category, one operation, and one system connection, and be sure to verify scenarios such as misclassification, incorrect summarization, incorrect recognition of product/price/inventory information, incorrect recognition of order/shipping information, incorrect transmission, incorrect updates, double execution, duplicate transmission, inclusion of personal information, Prompt Injection, and SaaS API failure/specification changes. You can only proceed to the production migration decision after confirming that human approval, escalation, stopping, rollback, and switching to manual operation function as designed.

Who This Is For

Who This Is For

Targeted at business managers, EC operation managers, and CS managers who have finalized the requirements in the Refine STEP.

What You'll Decide

What You'll Decide in This Step

In the Build & Validate STEP, define the scope of the pilot, build Agent, Skill, and Tool Policy, conduct normal and abnormal tests, and decide whether to move to production.

Pilot Scope

Scope of the Pilot

It is recommended to conduct the Pilot with a limited scope according to the following units.

01

One sales channel

Even if multiple channels are operated, initially limit verification to one channel.

02

One product category

Target one product category with different characteristics, not all products.

03

One business process

Limit the verification to the one business process decided in Discover; do not verify multiple business processes simultaneously.

04

1 system connection

Limit the SaaS to connect to a minimum and make it easier to verify the approval flow.

Method

Implementation Steps

1. Build the Pilot Environment

Do not use production customer or order data; build Agents, Skills, and Tools with test data.

2. Implement Tool Policy

Implement the Allow/Deny scope defined in Refine as Tool Policy. Implement production final updates for price and inventory as Deny.

3. Test normal scenario

Verify whether expected classification, summarization, and draft are generated for the assumed input.

4. Test abnormal scenarios

We will verify misclassification, incorrect summarization, misrecognition of product, price, and inventory information, misrecognition of order and shipping information, incorrect transmission, incorrect updates, duplicate execution, duplicate transmission, personal information leakage, prompt injection, and SaaS API failures or specification changes.

5. Confirm human approval flow

Check whether human verification and approval function properly for operations that require approval.

6. Check the escalation route

We will check whether the escalation route to the business manager or CS manager functions in cases where escalation is necessary.

7. Confirm stop, rollback, and manual operation switching

Verify whether procedures to stop processing, roll back, and switch to manual operation in case of abnormalities function properly.

8. Decide in accordance with production migration criteria

Determine whether migration to production is possible according to the pre-established Go/No-Go criteria.

Data & Systems

Data and Systems Used

Verification data (test data simulating real data) Product, price, and inventory information Order information (anonymized) EC platform (test environment) OMS (test environment)

Human-in-the-loop

Where Human Approval Is Required

  • The test personnel confirm and approve whether the product description drafts and response drafts generated in the pilot can be reflected.
  • The business manager and EC operation manager review the settings of the Tool Policy.
  • The responsible person decides on the feasibility of moving to production based on pre-established judgment criteria

Measurement

KPI

Pass rate of test scenarios

The proportion of passed normal and abnormal scenarios

Number of occurrences of incorrect updates or incorrect transmissions

Number of incorrect operations or duplicate processes detected during the Pilot period

Evaluation of acceptance by the person in charge.

Evaluation of usability and business suitability by actual users

Pitfalls

Common Pitfalls

01

Judged as pass with normal cases only

If abnormal case verification is omitted, the behavior during SaaS failures may be unexpected after actual operation.

02

Testing using real customer and order data in production.

Using actual customer information and order data from the verification stage increases the risk of incorrect transmission and information leakage.

03

Not preparing rollback procedures

Without procedures to revert when problems occur, recovery takes time and impacts customer response

Go / No-Go Criteria

Checklist for production migration decision

  • All normal scenario tests have passed
  • Abnormal scenario testing (misclassification, incorrect transmission, personal information leakage, duplicate execution, prompt injection, SaaS API failure) has been completed.
  • Human approval flow functions as designed
  • Rollback procedures are established and actual rollback has been confirmed
  • It has been confirmed that unauthorized operations (including actual updates of price and inventory in production) are rejected.
  • Practical issues are resolved in acceptance testing by the person in charge.
  • Procedures for switching to manual operation in case of failure are established.

FAQ

Frequently Asked Questions

What is the approximate duration of the Pilot period?

It varies depending on the complexity of the target business and connected SaaS. Since a uniform period cannot be presented, please consult individually considering the scope.

Can the verification be done without using actual customer and order data in production?

It is possible. We recommend preparing test data that imitates real data and conducting it in a verification environment that does not include personal information.

What happens if a Pilot fails?

Review the Refine requirements and Tool Policy, and conduct the Pilot again. It is recommended to continue verification until the standards are met without forcing the transition to production.

Shall we organize the Pilot design and verification together?

The scope, test scenarios, and Go/No-Go criteria can be concretized through discussions in the formal LP.

Consult on pilot design and verification