A path for every challenge. One platform to connect the company.

1. 01

   Prepare the environment

   Agree on the pilot tenant and access with your administrator. Separate test credentials and data from production. SMB is managed by Zild; dedicated deployment, Docker and Kubernetes are Enterprise options to scope in your proposal.

2. 02

   Connect only what you need

   Register the provider credential in Admin, configure an agent and connect one data source. Start with read-only queries and restricted permissions. Never place API keys in the browser or a repository.

3. 03

   Test before expanding

   Build cases for success, missing data, outages and access denial. Compare expected output, cost, latency and human handoff. Promote the workflow only with owner approval.

Go/no-go criterion: critical cases pass and cost per correct task stays within the agreed limit.

1. 01

   Map the handoffs

   Record the input, source system, business rule, owner and output. For inventory and WMS, start with discrepancy alerts and stock-position queries. ERP/WMS remains the authoritative inventory source.

2. 02

   Design exception handling

   In production, track orders and delays before automating changes. For import documents, extract fields and compare versions; route discrepancies to a specialist. Do not delegate customs clearance, payments or inventory adjustments without defined approval.

3. 03

   Measure the whole workflow

   Compare completion time, rework, pending cases and cost per resolved process. Use a representative case set and record exceptions instead of measuring only how many messages the agent answered.

Go/no-go criterion: shorter cycle time without more errors or reduced traceability.

1. 01

   Define what to extract

   Choose a field list: project ID, revision, room, element, material or quantity. Record units, source and calculation rules. Obtain a representative sample of the files your team uses.

2. 02

   Validate the integration path

   The path may require table exports, an API, a plugin or a custom integration. Validate format parsing and metadata availability in the pilot. Mentioning AutoCAD and SketchUp does not imply a native connector ready for every file.

3. 03

   Keep technical review

   Compare extraction against an approved takeoff. Preserve the project revision and element reference. Quantities, estimates and engineering decisions require review by the responsible professional.

Go/no-go criterion: correct fields and units, traceable sources and shorter review time.

1. 01

   Build a comparable baseline

   Add licenses, infrastructure, support and operations for the same activities. Identify contracts that can actually be ended, termination fees, migration and apps that must remain.

2. 02

   Include AI costs

   Include the Zild subscription, provider APIs, channels, human review, implementation and governance. Freed hours represent capacity; they become cash savings only when actual expenses fall.

3. 03

   Track value and control

   Track cost per correct outcome, adoption and rework. To reduce Shadow AI, offer an approved path to models, process owners and a data policy. Review the budget against actual usage.

Go/no-go criterion: validated total cost, measurable returns and owners for budget and quality.

1. 01

   Choose operational evidence

   Pilot example: prepare the evidence package for a maintenance order or material return. Link order, item, location, date, receipt and owner. Identify missing documents or conflicting revisions. The agent organizes the review; the accountable owner validates the conclusion and applicable obligations.

2. 02

   Define who reads, reviews and approves

   Map systems, connected accounts and AI tools already used by the team. Separate public, internal and restricted documents; limit contractors to authorized work. Test allowed and denied access using demonstration data. Define who handles exceptions and how to correct or suspend a workflow producing incorrect results.

3. 03

   Measure the process without inventing impact

   Track evidence completeness, review time, rework and unauthorized access identified during testing. For environmental indicators, record source, period, unit, boundary and approved method. Lower costs or fewer tokens do not establish lower emissions. Also record training, accessibility and adoption difficulties; before-and-after comparisons require equivalent scope.

Go/no-go criterion: traceable evidence, verified access, accountable owners and explicit indicator boundaries.