A 90-Day Roadmap: From a Manual Process to an Orchestrated One

Don’t start with a platform, and don’t start with a sweeping transformation programme. Start with one process, one outcome, and a baseline you can actually measure. The roadmap below is a planning framework for a focused pilot, not a promise that every process can reach production in 90 days. Timing depends on access to systems, process complexity, and the approvals required.

Days 1–15: Pick the right battle

Look for a process that runs frequently, crosses at least two systems, contains waiting time or exceptions, and has a business owner. A good pilot process hurts enough to be worth fixing, yet is contained enough to ship in weeks.

Record a baseline: cycle time, human handling time, completion rate, exceptions, and estimated cost per case. Without a baseline, even success remains a gut feeling rather than a number.

Days 16–30: Map what really happens

In a joint workshop with business and technology teams, map the process as it truly runs today — including the spreadsheet workarounds, the approvals handled over chat, and the cases that loop back. Define the start event, the desired outcome, the decisions, the data, the systems, and the owners.

This is also where you mark what follows fixed rules, where AI can help, and where a human decision is required. The goal is not a pretty diagram; it is a shared definition of how the process should run.

Days 31–60: Build an end-to-end MVP

• A BPMN model covering the standard path plus two or three common exceptions.
• Integration with systems through APIs or dedicated job workers.
• Human task screens with the context needed to decide.
• Metrics, logs, and a correlation ID for every instance.
• Time limits, retry rules, and compensating actions for work that needs to be reversed or corrected.

A complete process at small scale beats deep automation of a single step. That is how you test the central assumption: does orchestration improve the end-to-end outcome?

Days 61–75: Shadow mode and pilot

Run the new flow in shadow mode first: compare its proposed results with the existing process without duplicating customer messages, payments, or other live actions. Then move a limited group across, with a daily dashboard and a way back. Look not only at successes, but at where intervention was needed and what users failed to understand.

Days 76–90: Improve and decide on scale

Compare against the baseline, fix the highest-volume exceptions, and decide whether to increase volume, add an AI capability, or move on to the next process. In parallel, build a repeatable playbook: templates, security, metric naming, and ways of working.

An example target for day 90: one process running in production, five live metrics, an engaged business owner, and an improvement backlog grounded in real data.

Further reading (Hebrew sources)

• SmartFlows: Practical steps for building automations
• Calcalist: From an organization that uses AI to an AI-driven organization

Summary

• Choose a process based on value, scope, and business ownership.
• Build the full flow before going deep on a single automation.
• Scale only after comparing to the baseline and learning from exceptions.

Want help choosing the right pilot process? Book a discovery workshop with us