Most automation programs fail not because the technology is wrong, but because the sequencing is wrong. They start with the biggest, most visible thing, get stuck on integrations, and lose executive patience by month four. Here is a sequence that lands wins early, builds the data plane that subsequent wins depend on, and produces a defensible cost reduction by day 90.
Days 0–10 — the audit
- List every system in the operating stack — NMS, OLT controllers, ACS, CRM, ERP, ticketing, ACS, GIS.
- Get API documentation for each. Mark which have working APIs vs which are UI-only.
- Sit with the NOC, customer care, field. Build a heat-map of recurring manual work.
- Capture handle-time, ticket volumes, truck-roll rates by category.
- Pick 3 candidates from the heat-map that are: high volume, low complexity, single-tool to automate, measurable in money saved.
Days 10–30 — the data plane
Almost every later win depends on having one place where network state, subscriber state and field activity are queryable together. This is the most thankless 20 days of the project — you will not show a demo at the end of it — but skipping it doubles the timeline of everything that follows.
- Stand up the unified data plane (NetXol or build-your-own).
- Onboard the OLTs and ACS — the network side first because the CRM/ERP side is more stable.
- Verify LLDP/CDP topology discovery against your existing manual topology. Fix the deltas.
- Start streaming alarms, events and counters into the unified store.
Days 30–60 — quick wins land
With the data plane in place, the quick-wins shortlist becomes implementable. The right three for almost everyone:
- 1Auto-provisioning for new subscribers — eliminate the 22 NOC minutes per activation and the 7% first-week error rate.
- 2AI RCA on customer-reported ONU outages — cut diagnostic time from 12 minutes to seconds, reduce truck rolls by 30–40%.
- 3Alarm suppression / topology-aware grouping — bring visible alarm volume down 80–90% so the NOC can actually read the queue.
What you will demo at day 60
A new-subscriber activation flow from CRM click to "online" in under 10 minutes, end-to-end. A live RCA on a real outage with confidence score and recommended action. An alarms dashboard that finally shows you the real signal.
Days 60–90 — close the loop
The last 30 days are where automation becomes autonomy. You let the platform act, not just recommend. Begin with the smallest, lowest-risk action class.
- Auto-Fix policy for one category — e.g., "if CPE high temperature persists 15 min, push QoS limit and notify owner."
- Predictive maintenance for SFPs and ONUs — surface candidates for replacement before they fail.
- Executive auto-reporting — replace the analyst's three-day monthly report with an auto-generated daily one.
- Capacity-planning pipeline — first set of 60-day saturation forecasts on the desks of finance and engineering.
What good looks like by day 90
−45%
NOC + care handle-time
Diagnostic queries collapse from 4 tabs to 1.
−35%
Avoidable truck rolls
RCA closes more cases before dispatch.
+95%
New activations zero-touch
Exceptions go to a single, well-defined queue.
−80%
Visible alarm volume
Topology-aware suppression of duplicates and aftershocks.
Why we sequence this way
Three reasons. First, the data plane is the bottleneck for everything else, so it goes first even though it is unglamorous. Second, the quick wins create executive and operator confidence — they unlock the rest of the budget and the rest of the cultural change. Third, autonomous action goes last because it is the most consequential step and you want it land on a foundation that has been stress-tested.
