The problem
Backend operations were mostly manual and repetitive, so customer requests took longer than they needed to. Customers noticed: requests that had been automated came back noticeably faster than the ones that hadn’t.
My role
I ran the roadmap end to end: identifying candidate processes, prioritizing them, getting approval from the Operations Head, planning delivery, and implementing. Each project had its own delivery team:
How I prioritized
I scored every candidate process on impact, cost, and time to implement. The first wave was high-impact, low-cost projects that could ship fast and act as pilots. Once those proved the value, we used that credibility to take on high-impact, high-cost projects.
Gained
Early, visible wins that built trust with executives and operations teams.
Gave up
Some of the biggest-value projects waited until the pilots had landed.
Impact →
Wave 1 · pilots
High impact, low cost
Wave 2 · scale
High impact, high cost
Fill-ins
Low impact, low cost
Not now
Low impact, high cost
Cost →
The pushback
Some department heads resisted at first. They expected automation to dissolve their teams. I reframed it as an accelerator that takes the repetitive work off their teams rather than replacing them.
Automation as an accelerator, not a replacement.
To make that concrete, I ran AI literacy workshops showing how AI would augment their teams’ work, and that the shift was coming either way, so learning it beat resisting it. Redeploying freed-up people, instead of cutting them, backed that promise with action.
Inside one project: customer feedback analysis
Before. The customer service team analyzed feedback from 15 channels by hand. Each cycle took 5 days. Analysts could only read a random sample of the 300K+ data points, so most feedback was never seen, and the repetitive work invited errors.
What I led. I took a feedback-analysis product from concept to deployment. AI reads feedback from all 15 channels in multiple languages, sorts it into topics and subtopics, and publishes it to a dashboard. Channel heads for the mobile app, retail website, corporate website, and branches use it to see what’s broken and what’s working, and act on it.
After. Every data point gets analyzed instead of a sample. The first full read surfaced slow email response as the top complaint driver. A cycle now takes 2 days instead of 5, and the work needs 6 people instead of 10, a 40% cost reduction. The other 4 moved to a different team within the bank.
Top complaint drivers
Channels
Mobile appRetail websiteCorporate websiteBranchesResults
| Metric | Result |
|---|---|
| Backend tasks covered by the roadmap | 20% |
| Projected annual savings | $10M, net of implementation cost |
| Capacity freed and redeployed | ~1,500 roles |
| Customer feedback analyzed | 100% of 300K+ inputs (previously a sample) |
| Feedback analysis cycle | 5 days → 2 days |
| Feedback analysis cost | 40% lower (team of 10 → 6) |
What I’d do differently
I’d build AI literacy first and automate second. I ran the workshops in response to pushback; running them before the first project would have built that buy-in from day one, instead of earning it back after resistance had already formed.