Impact: what auto-select earns today, and what the change would add
This page answers two questions. How much more does the platform earn with auto-select than before auto-select existed? How much more would the proposed rule earn than today's runs? It is the fourth page of the auto-select timing initiative. Every number comes from the nightly copy of production, jodgig_ai, pulled on 17 September 2026 and complete through 13 September 2026. Every number carries one of three tags: fact from data, replay estimate, or assumption.
Terms used on this page:
| term | meaning |
|---|---|
| auto-selected job | a jod_jobs row whose auto_select_user_id the system set. A repost copies its parent's auto_select_user_id, and those copies are not counted. |
| manager-selected job | a jod_jobs row with hired_by_manager_id set that is not auto-selected. |
| job with applicants | a jod_jobs row with no_users_apply_job above 0. |
| no-selection rate | no-selection jobs (status: 8) divided by jobs with applicants. |
| completion rate | completed jobs (status: 3 or status: 16) divided by jobs with applicants. |
| worked slot | a slot with a valid paid payment record. The company charge on it averaged SGD 90.89 in the twelve weeks of shifts starting 22 June to 13 September 2026. |
| gross value | every credit earned on a shift that auto-select filled. It is an upper bound, because a hiring manager would have filled many of those shifts by hand. |
| net value | the part of the gross value that would not have happened without auto-select. No comparison group exists, so every net number here is a replay estimate. |
| a miss | a no-selection job at an outlet with locations.is_auto_select_applicant = 1 that still had a live eligible applicant at expiry. 703 jobs in the twelve weeks. |
Part A — What auto-select earns today against before it existed
1. Auto-select began on 3 December 2025 and reached full volume in its first week
The first auto-selected job was job 591235 at outlet 1921, for a shift starting 3 December 2025 at 22:00 (fact from data). The week of Monday 1 December 2025 already held 31 auto-selected jobs (fact from data). Between 1 December 2025 and 13 September 2026 the system made 2,876 real auto-picks (fact from data).
The measurement splits into three periods. Every number in this table is a fact from data. Averages are per week.
| measure | before: 2 Jun to 30 Nov 2025 (26 weeks) | opt-in: 1 Dec 2025 to 21 Jun 2026 (29 weeks) | all outlets on: 22 Jun to 13 Sep 2026 (12 weeks) |
|---|---|---|---|
| jobs with applicants a week | 1,722 | 1,754 | 1,786 |
| auto-selected jobs a week | 0 | 47 | 126 |
| manager-selected jobs a week | 1,490 | 1,456 | 1,480 |
| no-selection jobs a week | 147.9 | 161.7 | 84.1 |
| no-selection rate | 8.59% | 9.22% | 4.71% |
| completion rate | 68.31% | 66.57% | 69.50% |
| credits a week | 116,008 | 121,028 | 124,022 |
| credits per job with applicants | 67.4 | 69.0 | 69.5 |
The opt-in period looks worse than before because only outlets that opted in ran the cron, and because that period contains December to February, the worst window of every year. At the outlets that did opt in, the no-selection rate fell 3.77 points against the outlets that had not yet opted in (replay estimate).
2. The season does not explain the drop in the no-selection rate
The same calendar weeks one year earlier are the check. The platform rate went from 7.89% in the twelve weeks before 22 June 2026 to 4.71% in the twelve weeks after it, a drop of 3.18 points. In 2025, with no auto-select, the same two windows were 7.93% and 7.70%, a drop of 0.23 points. The difference is 2.95 points (all fact from data). The two before-windows, 2025 and 2026, are almost the same rate, which makes the after-window difference easy to read. The same after-window weeks were 8.24% in 2024 and 7.70% in 2025 (fact from data).
3. Gross value: SGD 7,176 a week on the shifts auto-select filled, which is an upper bound
This counts every credit earned on a shift that auto-select filled. Many of those shifts would have been filled by a hiring manager anyway, so it is the ceiling, not the gain.
| window | auto-selected jobs | of which completed | worked slots | credits a week | share of credits in the period | credits a year | tag |
|---|---|---|---|---|---|---|---|
| opt-in, 29 weeks | 1,363 | 802 (58.8%) | 819 | SGD 2,454 | 2.0% | n/a | fact from data |
| all outlets on, 12 weeks | 1,513 | 1,010 (66.8%) | 1,023 | SGD 7,176 | 5.8% | SGD 373,000 | fact from data |
The year figure is the weekly figure times 52 weeks. Reading the gross number as the gain is an assumption, and it holds only if no hiring manager would have picked any of those jobs by hand.
4. Net value: about SGD 3,500 a week, inside a range of SGD 2,000 to SGD 4,500
The central reading starts from the 2.95-point drop in the no-selection rate that remains after the same weeks one year earlier are subtracted. It applies that drop to the 1,786 jobs with applicants a week, keeps the 66.8% of auto-selected jobs that complete, and multiplies by 99.9 credits per completed job (1.10 worked slots at 90.7 credits each). That gives 52.7 jobs rescued a week, 35.2 of them completed, and about 3,520 credits.
| reading | where the change comes from | credits a week | credits a year | share of the 124,022 credits a week | tag |
|---|---|---|---|---|---|
| lower | the June 2026 switch-on, 12 weeks against the 12 before it | about SGD 2,000 | about SGD 104,000 | 1.6% | replay estimate, biased low |
| central | the same weeks one year earlier, on the no-selection rate | about SGD 3,500 | about SGD 182,000 | 2.8% | replay estimate |
| upper | the December 2025 opt-in switch, applied to today's volume | about SGD 4,500 | about SGD 234,000 | 3.6% | replay estimate |
| ceiling | every credit on every auto-selected shift | SGD 7,176 | SGD 373,000 | 5.8% | fact for the credits, assumption for the attribution |
Credits per job with applicants rose from 67.4 to 69.5, about 3% (fact from data). That rise is consistent with the central reading.
5. The money did not follow the halving, because auto picks complete at 66.8% and manager picks at 78.2%
The no-selection rate roughly halved. The completion rate rose 1.19 points, from 68.31% to 69.50%. The reason is what happens after an automatic selection (all fact from data, twelve weeks of shifts starting 22 June to 13 September 2026). The rates in this table are shares of the selected jobs of each kind, not of all jobs with applicants.
| outcome of the job | auto-selected jobs | manager-selected jobs |
|---|---|---|
the hiring manager rejects the jodder (status: 10) | 14.1% | 8.7% |
the jodder cancels (status: 11) | 11.4% | 9.8% |
| the job completes | 66.8% | 78.2% |
So auto-select turns many "nobody selected" jobs into "selected, then rejected or cancelled" jobs. About two thirds of a rescued job becomes a worked shift.
6. One outlet switched the flag off in August, and its no-selection rate rose from 8.6% to 11.0%
FairPrice Xtra Changi Business Park is outlet locations.id = 693, company NTUC FairPrice (Hyper). The flag came on for every outlet on 22 June 2026. This outlet's last auto-selected shift started 11 August 2026. It is the closest the data comes to a test, because the same outlet ran without auto-select, then with it, then without it again. All numbers are facts from data.
| measure | before: 1 Apr to 21 Jun 2026 (82 days, flag off) | during: 22 Jun to 11 Aug 2026 (51 days, flag on) | after: 12 Aug to 13 Sep 2026 (33 days, flag off) |
|---|---|---|---|
| jobs with applicants | 1,352 (16.5 a day) | 1,023 (20.1 a day) | 608 (18.4 a day) |
| no-selection rate | 122 of 1,352 (9.0%) | 88 of 1,023 (8.6%) | 67 of 608 (11.0%) |
| completion rate | 914 of 1,352 (67.6%) | 729 of 1,023 (71.3%) | 397 of 608 (65.3%) |
| credits a day | SGD 1,017 | SGD 1,273 | SGD 1,040 |
Two readings follow. While the flag was on, the no-selection rate barely moved, so the day-before run was not reaching this outlet's misses. After the switch-off the no-selection rate rose 2.4 points and the completion rate fell 6.0 points, while the other four FairPrice Hyper outlets rose in the same weeks. The credits the switch-off costs are between SGD 650 and SGD 2,900 a month (replay estimate). The lower figure takes the 2.4-point rise in the no-selection rate on 18.4 jobs a day, at this outlet's own 57% auto completion and SGD 86.47 a slot. The upper figure takes the whole 6-point completion drop. One outlet, 33 days, so this is directional only.
7. Every net number above is an estimate, for four reasons
| reason | what the data shows | effect on the reading |
|---|---|---|
| no clean comparison group | 24 of 2,381 outlets have the flag off today, and all of them ran auto-select from 22 June until early August 2026. One outlet posts 59% of their jobs. | The flag-on against flag-off split cannot measure the effect. It gives 0.40 points, which is not usable. |
| the flag has no history | Only today's value of locations.is_auto_select_applicant is stored. An outlet that was on in July and off today counts as off for every week. | The "opted in early" group is built from outcomes, not from settings. |
| seasonality | December to February is the worst window every year: 9.92% no-selection in 2024/25 and 10.77% in 2025/26. | The opt-in period looks worse than before partly because of the calendar, not because of auto-select. |
| the biggest outlets switched the flag off | Changi Business Park alone had 216 auto-selected jobs between 23 June and 11 August 2026, then switched off. | Today's platform rate already excludes the outlets with the most misses. |
Platform growth adds a fifth caution. Jobs with applicants a week rose from 1,722 to 1,786, and credits per worked slot rose from 89.1 to 90.7 (fact from data). Use credits per job with applicants, not credits a week.
Part B — What the proposed rule would add on top of today
8. The options, one table, with the constants stated once
These constants apply to every row below:
- SGD 90.89 is the company charge on one worked slot.
- 66.8% of automatic picks complete.
- 86% of re-selected parent applicants complete.
- One year is 52 weeks.
The rules are on the solution page, the solution in five rules. Rule 1 selects at each job's last safe moment. Rule 3 re-selects the parent job's applicants after a dropout. The replay window is the twelve weeks of shifts starting 22 June to 13 September 2026, except where a row says otherwise. The first row is counted from payment records, not replayed. The credits columns come from the unrounded weekly counts, so they do not always match the rounded slot column.
| option | jobs selected a week | worked slots a week | credits a week | credits a year | tag |
|---|---|---|---|---|---|
| today's two runs, 09:00 and 17:00 the day before the shift (12 weeks) | 126 | 85 | SGD 7,176 | SGD 373,000 | fact from data, gross upper bound |
| the short-notice run that shipped on 16 September 2026, replayed on 8 weeks of shifts starting 20 July to 13 September 2026: 68 of 631 no-selection slots (11%) | 9 | 6 | SGD 550 | SGD 28,000 | replay estimate |
| rule 1 at a 6-hour floor: 452 of 703 misses (64%) (12 weeks) | 38 | 25 | SGD 2,300 | SGD 119,000 | replay estimate |
| rule 1 with night selections pulled to 21:59 the evening before: 359 of 703 misses (51%) (12 weeks) | 30 | 20 | SGD 1,800 | SGD 94,000 | replay estimate |
| rule 1 at a 4-hour floor: 492 of 703 misses (70%) (12 weeks) | 41 | 27 | SGD 2,500 | SGD 129,000 | replay estimate |
| rule 1 at a 3-hour floor: 517 of 703 misses (74%) (12 weeks) | 43 | 29 | SGD 2,600 | SGD 136,000 | replay estimate |
| rule 3, re-select the parent job's applicants after a dropout, 6 hours or more of lead: 193 of 396 lost reposts with a free parent applicant (49%) (12 weeks) | 16 | 14 | SGD 1,300 | SGD 65,000 | replay estimate |
| of those 193, the 34 clones that never got an applicant of their own, so new value on top of rule 1 (12 weeks) | 3 | 2 | SGD 220 | SGD 11,000 | replay estimate |
| the four outlets with the flag off switched back on, with rule 1. Changi Business Park alone: 123 of its 150 misses (82%) (12 weeks) | 10 | 7 | SGD 620 | SGD 32,000 | replay estimate |
| pick quality: auto completion rises from 66.8% towards the 78.2% of manager picks, on about 100 automatic picks a week (12 weeks) | no new jobs | 11 | SGD 1,000 | SGD 54,000 | assumption |
Two notes on the rows. The three other flag-off outlets have no replay number in the source, so only Changi Business Park is counted; at that outlet automatic picks completed at 57%, not 66.8%, so its own figure would be lower. The pick-quality row is an assumption, not a plan: rule 5 on the solution page narrows the gap, and closing it needs a better ranking, which is separate work.
9. The options overlap, so do not add the rows together
Rule 1 at a 4-hour floor contains rule 1 at a 6-hour floor. Rule 3 at a 6-hour lead overlaps rule 1, because many lost reposts are inside the 703 misses. Only three pieces are additive:
- rule 1 at a 6-hour floor, about SGD 119,000 a year.
- the 34 clones rule 3 reaches that never got an applicant of their own, about SGD 11,000 a year.
- Changi Business Park, whose misses sit at a flag-off outlet and are outside the 703, about SGD 32,000 a year.
Rule 3 adds one more piece where it overlaps rule 1. It selects at the moment of the dropout instead of hours later, and it selects someone who already chose this shift and completes at 86% instead of 66.8%. On the 159 lost clones with 6 hours of lead that is about 30 more worked slots in twelve weeks, about SGD 12,000 a year (replay estimate).
| combined reading | what it includes | credits a year |
|---|---|---|
| lower | rule 1 at a 6-hour floor, plus the 34 clones | about SGD 130,000 |
| upper | the same, plus the completion gain on overlapping reposts, plus Changi Business Park | about SGD 175,000 |
10. The change is worth about SGD 130,000 to SGD 175,000 a year, and the first step costs nothing
Auto-select earns about SGD 373,000 a year gross today, of which about SGD 182,000 is the central net reading. The proposed rule would add roughly SGD 130,000 to SGD 175,000 a year on top of that (replay estimate). The range widens if the floor drops to 4 hours and narrows if night selections move to 21:59. None of these numbers can be checked after the change ships, because the current code throws away the result of every automatic attempt. The first step, rule 4, writes one log line per attempt with its outcome. It moves no money by itself and it makes every number on this page measurable instead of replayed. The solution page, build order puts it first for that reason.