Sales Execution 56→1
One mobile platform replacing 56 legacy tools across 22 countries and roughly 60% of the client's global revenue — then the harder half: getting it actually used.
- Role
- Business Analyst → Product Owner
- Timeline
- 2018 – 2024
- Scope
- 22 countries · 30+ person delivery org
- Result
- 4 → 7 calls per rep per day
A 22-deal internal sales process
The consolidation
I joined as a Business Analyst in 2018 on requirements and process mapping for the MVP, was promoted to Product Owner in 2020, and owned the platform across 22 markets through 2024 — a 0→1 mobile build replacing 56 legacy tools and covering ~60% of the client's global revenue.
Prioritisation ran on RICE for cross-market features and MoSCoW for local requests, with quarterly reviews where all 22 market leads looked at the same backlog. Gap analysis surfaced 120+ workflow differences, each classified as a global standard, a regional variant, or a market-specific exception.
This wasn't a rollout — it was a 22-deal internal sales process. Each country head needed to believe the platform served their market, not just the global standard.
Stalled at 60%, and the dashboards couldn't say why
The adoption crisis
The platform shipped and adoption flattened at 60%. The analytics told me where usage dropped, not why. So I went to the field and shadowed reps in the most resistant market.
The global build had ignored local requirements. The fix was "global method, local tools": one methodology, with country-specific functionality handled as configuration rather than exception. Adoption moved from 60% to 70%.
Where it fit: computer vision on the shelf
AI judgment
A third-party computer vision capability detected SKU presence on shelves against contracted mix and promo targets. I owned the product integration end to end: detection → CRM opportunity → field action → verification loop → contract-compliance tracking, and treated precision/recall as an ongoing tradeoff with an override path and confidence tuning.
- ~30% more outlets hitting their SKU-mix target.
- ~20% more outlets hitting their promo-mix target.
The eval question was never "did the model detect correctly." It was "did acting on the detection produce the intended business result."
Where it didn't: killing the Einstein pilot
AI judgment
Salesforce pushed an Einstein pilot for route and visit planning. On inspection it optimised for rep proximity, not business need. I built a rules-based scoring engine instead and killed the pilot.
Reps went from 4 to 7 calls a day — a 12% productivity gain.
It taught me to be genuinely critical about whether AI is the right tool for a problem — not just whether we can say we have an AI feature.