Saved IDR 676M annually by building a custom HRIS for 1,000+ staff across 165+ outlets
I led the end to end design and product management of Pantry, an internally built HRIS that replaced Hangry's third party subscription. By building a system tailored to our operations across 165+ outlets and 1,000+ staff, we eliminated IDR 676.2M in annual vendor costs while gaining full control over our workforce processes.
The organization faced three compounding issues that made continuing with the existing vendor unsustainable.
We built Pantry, a custom HRIS designed specifically for Hangry's operational structure. Rather than adapting to a generic platform, Pantry was shaped around the workflows our staff and admins already knew, giving us direct control over workforce processes across all outlets.
One platform, three surfaces
Before Pantry, Hangry ran on a third party HRIS (Mekari Talenta) that fanned out across three separate surfaces and two very different audiences. Understanding its full shape came first, so the scope of what we would rebuild in house was clear.
Dozens of features across the stack
The platform spanned three layers of HR, from day to day core operations to strategic tooling. Rebuilding all of it in house was never realistic in the time we had, which made auditing the full surface area essential.
40+ features across 3 HR layers, 3 surfaces, and 2 audiences.
Where the subscription cost actually sat
The clearest signal for what to prioritize was the third party subscription bill. Operational accounts and the outlet portal dominated the annual cost, while HQ was a rounding error by comparison.
Operational cost ~8× more than HQ, and carried the heavier day to day reliance.
Operational first
With cost and reliance both pointing the same way, I made the call: build the operational surfaces and the admin back office first, and defer only the HQ user app to after launch.
From requirement to live, then improve
With scope set to operational first, this was the P0 plan to reach go live before April. Everything here shipped for launch. The HQ app and deferred features became P1 and P2 after live.
Every step above was P0, delivered before and around go live. The HQ app and the features we deferred became the P1 and P2 roadmap after launch.
The Portal App runs on one shared device per outlet. Any staff member finds their name, confirms their shift, and clocks in with a face scan. It is the most safeguarded surface in the system, because attendance ties directly to payroll.
Operational staff use their own phone to request overtime and leave, check payslips, and keep their profile current. The design leaned into the two things they do most, and cut every step we did not have time to build.
A limited selection of screens is shown. The full app is under NDA.
Feb 2023 to Nov 2025
Before rolling out to 165 outlets, we trialed live attendance at 5 outlets over 5 days, 6 to 10 March. Three ran with geo-fence restrictions, and two warehouses ran without, since warehouse staff are highly mobile.
174 of 227 · first two trial days (6 to 7 March)
- The Operations team had initially defined both the range and the stored coordinates for every outlet.
- Some ranges were set as low as 1.9 to 2.5m, below what phone GPS can achieve.
- Several outlet coordinates were themselves inaccurate.
- With natural GPS drift of 15 to 30m in malls and shophouses, staff standing at the counter were rejected.
Scale risk: with 165 outlets going live end of March, the same misconfiguration would not stay in the trial. It would multiply platform-wide.
I dug into GPS accuracy and found the error depends heavily on the environment, so a single tiny threshold could never work everywhere.
The rejections had two root causes, so the fix had two parts.
- With the coordinates corrected, no further complaints about inaccurate locations, and staff could check in successfully 100% of the time.
- With the 50m range applied, every clock-in fell within range. Of 22 records on 9 March (15:40 to 23:19), the mean was 7.20m, with a minimum of 1.22m and a maximum of 15.19m, all well under the 50m boundary.
Pantry went live on April 21st. These are the results captured from launch to date.