Diagnostic reportConfidential — Leadership
Operations analytics · August 2026
Why are orders missing their promised ETA?
A diagnostic investigation of operational behaviour between order placement, rider assignment, rider movement, restaurant arrival and ETA outcome.
Matched population
14,366
orders analysed
Period
5 – 31 Aug
2026
Supporting datasets
17,351
August assignments · 12,147 clean no-reassignment sample
This report quantifies the problem and narrows the likely root-cause space. It does not set out a remediation strategy.
01 — Executive diagnostic summary
n = 14,366 matched orders
ETA failure appears to be driven more by effective rider availability and assignment conditions than by raw rider headcount
01
Concurrent rider workload is strongly associated with ETA failure
| 0 existing active orders | 18.0% |
| 1 | 32.4% |
| 2 | 40.6% |
| 3+ | 54.3% |
ETA failure rises from 18% to 54% as concurrent rider workload increases.
02
Actual Rider → Restaurant time is much longer than Google travel time
Google median
7.0m
Actual median
13.75m
Difference
+5.5m
52.0% took >5 min longer than Google · 32.7% >10 min · 11.7% >20 min.
Analytical label: Rider Execution Gap = actual Rider→Restaurant − Google Maps ETA
03
Reassignment complexity is strongly associated with failure
18.7
27.9
36.5
51.4
49.0
54.4
0
1
2
3
4
5
Reassignments per order, against ETA miss rate (%). Orders experiencing repeated assignment attempts show rapidly deteriorating ETA outcomes.
04
Rider headcount alone does not explain zone performance
Amerat
7.86
orders per active rider-day
3.47×
network rider throughput
10.1%
ETA miss
High rider utilisation can coexist with strong ETA performance.
05
Some low-throughput zones perform badly
Khoud 6
| Orders / rider-day | 1.79 |
| ETA miss | 27.7% |
| Manual assignment | 56.6% |
| Busy rider | 33.7% |
| Rider→Restaurant >20m | 42.2% |
06
Time of day matters
8–10 AM
34.3%
9 PM–1 AM
17.2%
Yet evening carries much higher volume.
The same network behaves very differently depending on time of day.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC01 / 19
02 — What are we actually investigating?
Framing
“Rider available” and “rider capable of reaching the restaurant quickly” are not necessarily the same thing
Observed sequence
Order placed
Rider pool exists
Rider assigned
Rider begins / continues movement
Restaurant arrival
ETA outcome
Dimensions visible in today’s data
Existing rider workload
Assignment source
Reassignment count
Road ETA
Actual movement time
Zone
Time of day
Rider distribution
Central diagnostic question
Why does a rider whose road journey should take approximately 7–8 minutes frequently require 15, 20 or 30+ minutes to reach the restaurant?
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC02 / 19
03 — Concurrent rider workload
n = 14,366 matched orders
Rider workload shows one of the clearest monotonic relationships with ETA failure
| Other active orders | Orders | Share | ETA miss | Median R→R | >30m R→R |
|---|---|---|---|---|---|
| 0 | 11,725 | 81.6% | 18.0% | 12.9m | 8.0% |
| 1 | 2,097 | 14.6% | 32.4% | 18.2m | 20.5% |
| 2 | 463 | 3.2% | 40.6% | 21.2m | 31.1% |
| 3+ | 81 | 0.6% | 54.3% | 25.5m | 43.2% |
ETA miss rate by concurrent workload
0
18.0%
1
32.4%
2
40.6%
3+
54.3%
Median Rider→Restaurant (minutes)
0
12.9m
1
18.2m
2
21.2m
3+
25.5m
Moving from no existing order to one existing order corresponds to a 14.4 percentage-point increase in ETA failure.
Riders carrying 3+ existing orders require approximately twice the median Rider→Restaurant time of riders carrying none.
Association only. These relationships do not establish causation.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC03 / 19
04 — Road travel vs operational travel
n = 14,366 matched orders
Road conditions explain only part of the time riders actually take to reach restaurants
Google median
7.0 min
Actual median
13.75 min
Median difference
+5.5 min
Share of orders slower than the Google estimate
> 5 min slower
52.0%
> 10 min slower
32.7%
> 20 min slower
11.7%
Zero existing orders
| 7m | |
| Actual | 12.93m |
| Gap | +4.77m |
| ETA miss | 18.0% |
1+ existing orders
| 8m | |
| Actual | 18.62m |
| Gap | +9.47m |
| ETA miss | 34.5% |
The Google journey estimate changes by approximately 1 minute. Actual arrival time changes by almost 6 minutes.
Additional delay appears to be introduced outside pure road travel. The current data does not identify what that additional time consists of.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC04 / 19
05 — Assignment source
n = 14,366 matched orders
Manual assignments are associated with roughly twice the ETA failure rate of app-accepted assignments
Driver App Accept
9,185 orders
15.7%
ETA miss
12.4m
median R→R
6.8%
>30m R→R
Manual Dispatcher
5,181 orders
30.4%
ETA miss
16.6m
median R→R
17.7%
>30m R→R
The confounder — orders where the selected rider already carried another order
App
1.3%
Manual
48.7%
Manual assignment is likely also identifying a more difficult class of orders. The two populations are not comparable as observed.
Controlled comparison — riders with zero other active orders
App + free rider
15.7%
Manual + free rider
25.8%
A substantial performance difference remains even after controlling for this basic workload dimension. This warrants deeper investigation but does not establish causality.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC05 / 19
06 — Reassignment complexity
n = 14,366 matched orders
Orders with repeated assignment attempts are substantially more likely to miss ETA
18.7%
27.9%
36.5%
51.4%
49.0%
54.4%
0
1
2
3
4
5
Reassignments per order
Orders with three assignment changes miss ETA in approximately half of cases.
Reassignment may either contribute to delay or simply indicate an order that was already difficult to fulfil. Today’s data does not distinguish between these interpretations.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC06 / 19
07 — Zone rider throughput
Zone-level, August 2026
More orders per rider does not automatically mean worse delivery performance
Orders per active rider-day — network average
2.27
network avg 2.27
1
2
3
4
5
6
7
8
9
10
40%
0%
ETA miss %
8.5
0
Orders per active rider-day — bubble size ≈ order volume
1 · Amerat 7.86 / 10.1%
2 · Al Khuwayr S 4.08 / 15.1%
3 · Mabelah1 3.09 / 22.7%
4 · Koudh2 2.69 / 24.9%
5 · Bawshar 2.54 / 22.6%
6 · Khoud6 1.79 / 27.7%
7 · Al Hail N 1.72 / 22.6%
8 · Khoud1 1.70 / 25.0%
9 · Al Hail S 1.42 / 26.3%
10 · Al Mawaleh N 1.35 / 33.9%
Amerat
1,367 orders · 6.4 average active riders/day · 7.86 orders per rider-day · 3.47× network average · ETA miss 10.1% · busy rider 20.7% · Rider Execution Gap +4.0m · Rider→Restaurant >20m 22.6%
Amerat operates at extremely high rider throughput while maintaining one of the strongest ETA outcomes. This challenges the hypothesis that high rider load necessarily means inadequate rider supply.
Al Khuwayr South
1,525 orders · 13.9 active riders/day · 4.08 orders per rider-day · 1.80× network average · ETA miss 15.1%
Again: high throughput does not necessarily create poor ETA performance.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC07 / 19
08 — Two routes to the same failure rate
Zone-level, August 2026
Different zones reach similar ETA failure through very different operating patterns
High rider-throughput, weaker outcomes
Mabelah1
| Orders/rider-day | 3.09 |
| ETA miss | 22.7% |
| Manual | 45.1% |
| Busy rider | 23.3% |
| Execution gap | +6.6m |
| R→R >20m | 35.7% |
Koudh2
| Orders/rider-day | 2.69 |
| ETA miss | 24.9% |
| Busy rider | 22.4% |
| Execution gap | +7.0m |
| R→R >20m | 34.6% |
Bawshar
| Orders/rider-day | 2.54 |
| ETA miss | 22.6% |
| Manual | 52.6% |
| Assignment >5m | 19.9% |
Low rider throughput, poor outcomes
| Zone | Orders / rider-day | ETA miss |
|---|---|---|
| Al Mawaleh North | 1.35 | 33.9% |
| Al Hail South | 1.42 | 26.3% |
| Khoud1 | 1.70 | 25.0% |
| Al Hail North | 1.72 | 22.6% |
| Khoud6 | 1.79 | 27.7% |
Some of the poorest ETA outcomes occur where rider throughput is below network average.
Rider quantity alone cannot explain zone performance.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC08 / 19
09 — Khoud 6 / Al Mawaleh North case
Zone-level, August 2026
Some zones appear to have rider capacity but continue to experience high operational delay
| Indicator | Khoud 6 | Al Mawaleh North |
|---|---|---|
| Orders / rider-day | 1.79 | 1.35 |
| ETA miss | 27.7% | 33.9% |
| Manual assignment | 56.6% | 38.6% |
| Busy rider | 33.7% | 24.4% |
| Rider Execution Gap | +6.4m | +9.2m |
| Rider→Restaurant >20m | 42.2% | 55.1% |
| Assignment >5m | 23.2% | 22.0% |
| Reassignment | 24.1% | — |
These areas do not resemble a straightforward “too many orders / too few riders” scenario.
Instead, several assignment and rider-execution indicators deteriorate simultaneously: manual dispatch share, busy-rider share, execution gap and long Rider→Restaurant times all sit above network levels while throughput sits below it.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC09 / 19
10 — Which variables move with zone ETA failure?
Adequately sized zones only
Rider→Restaurant delay is the strongest observed zone-level signal
Rider→Restaurant >20m %
+0.74
Assignment >5m %
+0.46
Rider Execution Gap
+0.41
Reassignment rate
+0.41
Manual assignment %
+0.22
Busy rider %
+0.11
Orders / rider-day
−0.58
−0.6
0
+0.8
Correlation with zone ETA miss rate
Rider travel and arrival execution metrics have substantially stronger relationships with ETA failure than raw rider throughput. Throughput moves in the opposite direction to failure.
Correlation ≠ causation
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC10 / 19
11 — Time-of-day effect
n = 14,366 matched orders
Morning is weakest by failure rate; evening contributes the largest number of misses
| Window | Orders | ETA misses | Miss rate | Share of ETA misses |
|---|---|---|---|---|
| 8 AM–12 PM | 1,107 | 308 | 27.8% | 10.2% |
| 12 PM–6 PM | 4,426 | 1,000 | 22.6% | 33.1% |
| 6 PM–1 AM | 7,937 | 1,566 | 19.7% | 51.8% |
| 1 AM–8 AM | 896 | 147 | 16.4% | 4.9% |
Highest risk per order
8 AM–12 PM
27.8% miss rate on 1,107 orders
Largest absolute miss population
6 PM–1 AM
1,566 misses — 51.8% of all ETA misses
These two ideas are distinct. The morning window carries the highest probability of failure per order; the evening window carries the largest number of affected customers.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC11 / 19
12 — Drilldown by time
n = 14,366 matched orders
The network deteriorates sharply around 8–10 AM and improves toward late evening
| Time | Orders | ETA miss | Median R→R | >20m R→R | Execution gap | Global available riders* |
|---|---|---|---|---|---|---|
| 8–10 | 396 | 34.3% | 17.5m | 39.4% | +7.7m | 15 |
| 10–12 | 711 | 24.2% | 16.3m | 34.2% | +7.5m | 30 |
| 12–3 | 2,545 | 22.8% | 15.0m | 31.9% | +6.0m | 39 |
| 3–6 | 1,881 | 22.4% | 14.1m | 28.8% | +5.4m | 44 |
| 6–9 | 3,641 | 22.7% | 14.3m | 30.4% | +6.5m | 61 |
| 9–1 | 4,296 | 17.2% | 12.2m | 24.1% | +3.9m | 65 |
*Global Khedmah availability, not local eligible rider supply.
ETA miss rate across the day
34.38–10
24.210–12
22.812–3
22.43–6
22.76–9
17.29–1
8–10 AM records approximately double the ETA miss rate of 9 PM–1 AM.
Rider→Restaurant performance deteriorates materially in the morning. The mechanism is not established by this data.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC12 / 19
13 — Morning effect on clean assignments
App accept · zero existing rider orders
Morning underperformance persists even when rider workload and manual assignment are removed
Sample restricted to Driver App Accept assignments where the selected rider carried zero existing orders — removing the two strongest confounders identified so far.
| Window | Orders | ETA miss | Median R→R | >20m | Execution gap |
|---|---|---|---|---|---|
| 8–12 | 749 | 25.0% | 16.4m | 33.4% | +7.5m |
| 12–6 | 2,722 | 15.6% | 13.1m | 24.8% | +4.9m |
| 6–1 | 5,109 | 14.9% | 11.8m | 20.8% | +4.3m |
Morning miss rate
25.0%
Evening miss rate
14.9%
Gap retained
+10.1pp
Morning performance cannot be explained solely by busy-rider assignments or manual dispatch. Possible reasons remain unproven in today’s data.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC13 / 19
14 — Potential rider load imbalance
n = 12,147 clean no-reassignment assignments
Some heavily loaded riders receive new orders while apparently zero-load riders are operating in the same zone
Rider already had ≥1 active order
18.4%2,241 orders
Rider already had ≥2 active orders
3.9%472 orders
For these cases the incoming order creates at least 3 simultaneous orders for that rider.
Of those 472 heavily loaded assignments
94.7%
447 / 472
had at least one other rider who served the same zone that day and had zero active orders at the exact assignment timestamp.
Methodological limit
This does not prove that rider was online, eligible or within acceptable assignment distance at that instant. Candidate-level eligibility is not present in the current dataset.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC14 / 19
15 — Strengthening the load-imbalance signal
472 heavy-load assignments
In a meaningful subset, zero-load riders received same-zone orders shortly after a heavily loaded rider was selected
| Other zero-load rider receives same-zone assignment | Cases | Share |
|---|---|---|
| Within ±15 minutes | 52 | 11.0% |
| Within ±30 minutes | 96 | 20.3% |
| Within ±60 minutes | 182 | 38.6% |
±15 min
±30 min
±60 min
13
assignments
had a selected rider already carrying 2+ active orders, at least two zero-load riders, and both of those riders receiving same-zone assignments within ±30 minutes.
This creates a credible workload-distribution question, but candidate eligibility data is required before describing any specific assignment as incorrect.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC15 / 19
16 — Forensic case study: Ruwi
12 August, ~1:30 PM Oman time
A rider carrying two active orders received a third while two zero-load riders operated in Ruwi minutes later
T + 0.0 min
Dilawar Hussain assigned
2 active orders before · 3 simultaneous orders after
T + 3.4 min
Makadar Al Balushi
0 active orders at the moment of the original assignment · takes a Ruwi order
T + 3.4 min
Shahbaz Safdar Cheema
0 active orders · takes a Ruwi order
T + 68.1 min
Restaurant arrival
Selected rider reaches the restaurant
Google ETA to restaurant
8 min
Actual Rider→Restaurant
68.1 min
ETA miss
32.7 min
This does not prove either alternative rider was eligible for this particular order, but it demonstrates why candidate-level assignment behaviour warrants closer examination.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC16 / 19
17 — Forensic case study: Wadi Kabir Industrial Estate
6 August, ~12:29 PM
A rider already carrying three active orders received a fourth while zero-load riders took same-zone assignments minutes later
Selected rider
Tassawar Hussain
3
existing orders
4
new concurrent total
Other rider activity in the same zone
Kashif Ali
0 active orders · same-zone assignment 7.5 min later
Badar Abdul Aziz
0 active orders · same-zone assignment 9.2 min later
12:29 PM
4th order assigned
+7.5 min
Kashif Ali — same zone
+9.2 min
Badar Abdul Aziz — same zone
How to read this
Assignment behaviour requiring explanation
Today’s data does not establish that these riders definitely could or should have received the original order.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC17 / 19
18 — What the data strongly shows
Findings summary
The ETA issue is more complex than rider quantity or rider distance
01
Rider workload matters
18.0% → 32.4% → 40.6% → 54.3%
ETA miss increases progressively as concurrent workload increases.
02
Actual rider arrival frequently exceeds road travel expectations
7 min Google → 13.75 min actual
Median across 14,366 matched orders.
03
Assignment complexity matters
Repeated reassignments coincide with materially worse ETA outcomes, rising from 18.7% at zero reassignments to 54.4% at five.
04
Rider throughput alone does not explain performance
3.47× network throughput → 10.1% miss
Amerat.
05
Low-throughput zones can still perform badly
Al Mawaleh North · Khoud 6 · Al Hail South · Khoud1
06
Time of day materially changes performance
34.3% at 8–10 AM · 17.2% at 9 PM–1 AM
07
There are credible rider workload-distribution anomalies
Heavily loaded riders sometimes receive additional orders while zero-load riders appear operational in the same zone around the same time.
KHEDMAH DELIVERY — RIDER ASSIGNMENT & ETA MISS DIAGNOSTIC18 / 19
19 — What the data does not yet prove
Boundaries of the evidence
“There are not enough riders.”
Current zone data measures riders who served an area, not exact eligible supply at assignment time.
“Manual dispatch is causing the problem.”
Manual assignment may identify inherently difficult orders.
“The assignment algorithm chose the wrong rider.”
Alternative rider eligibility is unknown.
“Morning rider shortage causes the 8–10 AM issue.”
The available-rider metric is global, not geographically filtered.
“Concurrent workload is the sole cause of ETA failure.”
It is strongly associated, but multiple dimensions interact.
Today’s investigation substantially narrows the problem. ETA failure is not adequately explained by order volume, raw rider headcount or rider distance alone. The strongest new signals sit between assignment and restaurant arrival: rider workload, assignment complexity, rider movement beyond road expectations, zone-specific operational behaviour and time-of-day effects.
The key unresolved question is not simply how many riders exist. It is why materially different ETA outcomes occur under apparently similar rider-supply conditions.
19 / 19