AI for Operations
Scheduling and dispatch that adjusts when the day does not go to plan.
What is AI operations?
AI for operations builds work schedules around your real constraints, assigns jobs to the people qualified to do them, and reworks the plan automatically when something changes, notifying whoever is affected.
Schedules are built the night before and wrong by ten the next morning. Someone calls in sick, a job overruns, an urgent request arrives, and the plan stops describing reality.
Rebuilding it by phone consumes a supervisor entire morning, and the version in their head is not visible to anybody else.
We build scheduling that accounts for the constraints that actually bind: who is qualified for what, travel time between jobs, customer windows, and priority accounts. When something moves, the plan reworks and the affected people are told.
Why people can override everything
Duara builds operational scheduling from Nairobi for businesses running field teams, delivery fleets and shift rotas where the plan changes daily.
We keep people able to override every decision the system makes. A scheduler who cannot overrule the plan stops trusting it within a fortnight, and a system nobody trusts gets worked around rather than used.
Who this is for
- You run field teams, deliveries or shifts
- One person spends the morning rescheduling
- Urgent jobs disrupt the whole day
- Nobody outside dispatch knows the current plan
Industries we serve
- Logistics and courier
- Field service and maintenance
- Construction
- Healthcare home visits
- Security services
- Utilities
Problems this solves
Scheduling automation pays where the plan changes daily. Where it holds for a week, a spreadsheet is fine.
- The plan is wrong by mid-morning. Absence, overruns and urgent jobs arrive daily, so the schedule built last night stops matching reality almost immediately.
- Rescheduling consumes a supervisor. One person spends the morning on the phone rebuilding the day, which is expensive and produces a plan only they can see.
- Travel time ignored in planning. Jobs are sequenced by geography on a map rather than by actual travel time, which in Nairobi are very different things.
- Nobody knows the current plan. Field staff work from what they were told at seven, and customers are given times nobody is tracking.
What we deliver
Schedules built from real constraints, automatic rework when the day changes, notification to everybody affected, and a plan visible to the whole business rather than one person.
Schedules built around your actual constraints
Automatic rework when something changes mid-day
Notifications to affected staff and customers
A shared view of the current plan
Skills and qualification matching on assignment
A record of what changed during the day and why
Features
- Constraint-aware planning. Qualifications, travel time, customer windows and priority accounts are all inputs rather than afterthoughts.
- Live travel times. Sequencing uses actual journey times rather than distance, which matters more in Nairobi than almost anywhere.
- Automatic rework. When a job overruns or somebody is absent, the plan rebuilds without a supervisor doing it by phone.
- Human override everywhere. Schedulers can overrule any assignment. A system that cannot be overridden stops being trusted.
- Shared visibility. The current plan is visible to office, field and management rather than living in one person head.
Benefits
- More jobs completed per day. Better sequencing against real travel time typically adds capacity without adding vehicles or staff.
- The morning is returned to the supervisor. Rescheduling stops being a person job, which frees the most experienced operator for actual supervision.
- Customers get accurate times. Windows are based on the live plan rather than on optimism, which reduces the calls asking where somebody is.
- Disruption stops cascading. One overrun is absorbed by a rework rather than derailing everything scheduled after it.
Technologies we use
- Constraint modelling. Skills, certifications, shift rules, vehicle capacity and customer windows encoded as real limits.
- Live travel time data. Journey estimates that reflect current conditions rather than straight line distance.
- Rescheduling engine. Rework triggered by absence, overrun or urgent insertion, preserving as much of the existing plan as possible.
- Mobile dispatch. Field staff receive their current assignments on an ordinary smartphone, updated as things change.
- Notification routing. Affected staff and customers told automatically when their part of the plan moves.
Scheduling quality depends on constraint accuracy more than on algorithm sophistication. A perfect optimiser given wrong travel times produces confident nonsense.
How the process works, step by step
Operational scheduling usually takes five to seven weeks, since constraint capture is slow and cannot be rushed.
-
Step 1
Free scheduling review
We watch how the day is currently planned and how often it changes, which decides whether this is worth automating.
-
Step 2
Constraint capture
We record the rules that actually bind, including the informal ones nobody has written down.
-
Step 3
Shadow scheduling
The system plans alongside your scheduler without being used, so its output can be compared against theirs.
-
Step 4
Assisted scheduling
Your scheduler works from the generated plan and overrides freely, which is where trust is built.
-
Step 5
Live running
The plan drives dispatch, with rework and notifications automatic and override always available.
Quality assurance and how we work
The commitments attached to systems that direct people work.
- Override is always available. Schedulers can overrule any assignment without justification. A plan that cannot be overridden gets worked around instead of used.
- We shadow before we replace. The system plans alongside your scheduler first, so its quality is demonstrated rather than asserted.
- Informal constraints get captured. The unwritten rules about who works with whom and which customer needs which person are recorded, because ignoring them produces unusable plans.
- Changes are logged with reasons. What moved during the day and why is recorded, so disputes about what was promised have an answer.
- Staff see their own plan. Field workers are told what changed rather than discovering it when they arrive somewhere.
Frequently asked questions about AI operations
Can staff override the schedule?
Always, and without justification. A scheduler who cannot overrule the plan stops trusting it within a fortnight, and then it gets worked around.
Does it handle Nairobi traffic?
It uses live travel times rather than distance, which matters far more here than in most cities for what a team can complete in a day.
What happens when a job overruns?
The plan reworks automatically, preserving as much of the existing schedule as possible, and everybody affected is notified.
Do field staff need special devices?
No. An ordinary smartphone receives assignments and updates as the plan changes.
How long until operations automation is live?
Five to seven weeks. Constraint capture is the slow part and rushing it produces plans nobody can use.
Why shadow before going live?
So the system output can be compared against your scheduler for a period. Quality is demonstrated rather than claimed.
What about unwritten rules?
Those get captured deliberately. Which staff work well together, and which customers need a particular person, are real constraints even if undocumented.
Can it handle qualifications and certifications?
Yes, and assignment respects them, so somebody without the required certification is never sent to that job.
Does it work for shift rotas as well as field work?
Yes, though the constraints differ. Shift rules, rest periods and skills mix replace travel time as the binding factors.
Will it reduce our headcount?
Most clients gain capacity rather than shed staff. The usual result is more jobs completed by the same team.
Can customers be notified automatically?
Yes, when their window changes, which noticeably reduces the calls asking where somebody has got to.
What if the plan it produces looks wrong?
Override it and tell us. A plan that looks wrong usually means a constraint was captured incorrectly, which is fixable.
Does it integrate with our job system?
Yes, to the common field service and dispatch systems, so jobs flow in and completions flow back.
How does it handle urgent insertions?
It reworks around them, showing what was displaced so the scheduler can decide whether that trade is acceptable.
Can it plan several days ahead?
Yes, though the further out the plan, the more it changes. Most value sits in same-day and next-day planning.
What if our constraints change?
They are configurable, and your team can adjust them without us for anything other than structural changes.
Does it track what actually happened?
Yes, including changes made during the day and why, which is what makes later disputes answerable.
Is this useful for a small team?
Below about six field staff, a good spreadsheet usually competes. We will say so during the review.
Can it optimise for cost or for speed?
Either, and the trade-off is yours to set. Optimising purely for one usually produces plans the other side of the business objects to.
What does operations automation cost to run?
It scales with staff and job volume. We project the monthly cost from your own operational figures before anything is agreed.
Glossary of terms
Scheduling vocabulary as used here.
- Dispatch
- Assigning jobs to people and telling them where to go and when.
- Constraint
- A real limit the plan must respect, such as a qualification, a shift rule or a customer window.
- Rework
- Rebuilding the schedule mid-day after something changes, ideally preserving most of the existing plan.
- Utilisation
- The proportion of available working time actually spent on productive jobs rather than travel or waiting.
- Service window
- The time range promised to a customer for arrival.
- Override
- A scheduler decision to overrule the automated assignment, which should always be permitted.
Related services in AI Implementation
Talk to us about AI operations
Tell us how often your daily plan changes after it is published. If the answer is rarely, this is not worth buying and we will say so.
Start your project