What is Predictive Intelligence?

Predictive Intelligence uses historical and current operational information to estimate where future risk, constraint or opportunity may emerge. Its purpose is not simply to produce a prediction. The objective is to give the relevant person enough useful warning to investigate and act earlier.

How is this different from Power BI?

Power BI helps users analyse and visualise information to understand what has happened and what is happening. Predictive Intelligence adds models designed to estimate what may happen next. Predictive results can still be surfaced through Power BI where that is the most useful place for the user.

How is this different from generative AI?

Generative AI is typically used to create or interpret content such as text, summaries or answers. Predictive models are designed to estimate a defined future outcome or risk using relevant historical and current signals. The technologies may work together, but they solve different problems.

Does Predictive Intelligence make decisions automatically?

Not by default. Our preferred model is decision support. The model identifies and explains a signal, while the responsible person reviews the information and decides what action to take. Controlled automation may be appropriate for specific low-risk activities where governance permits it.

What data does it need?

The required data depends on the decision being predicted. A warehouse model may use orders, inventory, movement, location, workload and historical completion data. The important question is whether the available information contains useful signals related to the outcome being predicted.

Does it only work with Dynamics 365?

No. Predictive Intelligence does not have to be limited to Dynamics 365. Where the required operational data can be accessed securely and reliably from another ERP, warehouse or operational platform, the use case can be assessed on its own merits.

Do we need Microsoft Fabric?

Not necessarily. Fabric may help where a model requires governed data from several systems, or where historical analytical information needs to be prepared for repeated use. A focused use case contained within one platform may not require an additional data platform. Architecture should follow the decision problem.

How accurate are the models?

There is no responsible universal accuracy figure. Performance depends on the use case, data, prediction target, operating conditions and the cost of different types of error. Each model should be evaluated against an agreed baseline and the business decision it is intended to support.

How do you know why a prediction was made?

Every model output is designed to show the contributing factors, the direction of the trend and the strength of the signal alongside the score. A prediction the user cannot investigate is difficult to act on and difficult to trust.

What happens when the model is wrong?

Predictive models are probabilistic and will sometimes be wrong. That is why confidence, explainability, human review and model monitoring are important. The design should consider both false positives and false negatives and the business consequence of each.

Can we build a model for our own use case?

Potentially. A predictive use case is strongest where there is a meaningful recurring decision, sufficient historical information, measurable outcomes and an opportunity to act before the outcome is fixed. We can assess whether a proposed use case has enough value and predictive readiness to justify a proof of value.

How do you prove whether a model creates value?

Start with the current decision and establish a baseline. Then evaluate whether the model identifies useful signals earlier and whether those signals lead to different actions or outcomes. The business case should come from measured evidence rather than an assumed AI benefit.

How are models monitored?

Input drift, prediction distribution, error patterns, usage, feedback and business outcome are all tracked against the deployed model version. A model should not be deployed and forgotten. Operations change, and the model has to keep pace with them.

Where does the prediction appear?

Wherever the decision is actually made. That may be the Predictive Intelligence workspace, Power BI, a Dynamics 365 or Business Central experience, Teams, email or an operational dashboard. Prediction should not force users into another application if that makes the workflow worse.

InteliSense Predictive Intelligence

See what may need attention
before it becomes obvious.

ReactiveProactive

Your ERP and operational systems record thousands of signals about how the business is working. Predictive Intelligence uses those signals to identify emerging risk, constraint and opportunity so the people responsible can investigate and act earlier.

Explore the models

The earlier you can see the problem, the more options you still have.

In short

What this page answers.

Predictive Intelligence is decision support built on your operational data. The models shown here are demonstrations, and we say so plainly.

What is Predictive Intelligence, and what is it not?
What makes an operational decision worth predicting?
Which model types apply to which questions?
What do our three current models demonstrate today?
How is value proved against a baseline?
How is prediction kept under human control?
How do we know when a model is production ready?
What happens after deployment, including retirement?

The reactive gap

Most operational systems tell you what happened. The difficult question is what deserves attention next.

The issue is not lack of information. It is timing.

If the pattern is recognisable, the opportunity is usually earlier warning rather than more reporting.

Warehouse
The backlog becomes visible after service is already under pressure.
Inventory
Excess stock becomes obvious after working capital has already been consumed.
SLA
Teams discover the order is late when there is little time left to intervene.
Labour
Capacity problems become obvious after workload has accumulated.
Suppliers
Performance deterioration becomes visible after operations are affected.
Cash
Pressure becomes visible after commitments are already difficult to change.
Pipeline
Risk becomes obvious after expected revenue has already moved.

The earlier you can see the problem, the more options you still have.

Predictive Intelligence is designed to move operational insight closer to the point where intervention is still possible.

Each layer answers a different question

Reporting is not replaced. It is extended.

Prediction sits alongside reporting and real-time insight rather than making them obsolete.

01
Reporting
What happened?
02
Real-time insight
What is happening?
03
Prediction
What may happen?
04
Decision support
What deserves attention?
05
Action
What should we do about it?

From signal to decision

Turn operational history into earlier warning.

Signals
Historical data, current conditions and business context.
Model
Patterns relevant to the outcome being predicted.
Risk
Where risk, constraint or opportunity may emerge.
Explanation
Which factors are contributing and how strongly.
Attention
What the responsible person should investigate.
Decision
The person reviews the evidence and decides.
Outcome
What actually happened is recorded.
Learning
Model and operating process improve from evidence.

Signals
Historical data, current conditions and business context.
Model
Patterns relevant to the outcome being predicted.
Risk
Where risk, constraint or opportunity may emerge.
Explanation
Which factors are contributing and how strongly.
Attention
What the responsible person should investigate.
Decision
The person reviews the evidence and decides.
Outcome
What actually happened is recorded.
Learning
Model and operating process improve from evidence.

Not just a score

A useful model answers more than one question.

A prediction is more useful when the user can understand why, how urgent it is and what they can still do about it.

What?
What is likely to require attention?
Why?
Which factors are contributing to the signal?
When?
How urgent is it, and how much time remains?
Impact?
What may happen if nothing changes?
Action?
What should the user investigate first?
How certain?
How much uncertainty sits behind the output, expressed in the way that model genuinely supports.

Model types

Different questions need different models.

Not everything described as predictive is a forecast. Being precise about the model type keeps expectations honest.

Forecasting
Estimates a future quantity, such as where workload may exceed available capacity.
Risk prediction
Estimates which records or work items may be heading for an unwanted outcome, so attention can be ranked.
Optimisation and recommendation
Analyses current conditions and history to identify a better arrangement or a prioritised action to investigate.

What makes a good use case

Prediction is only useful when somebody can still act.

The first question is not whether a model is possible. It is whether knowing earlier would change anything.

Questions we work through

Would knowing earlier actually change anything?
Which decision is repeatedly made too late?
Where do managers rely heavily on experience?
Where does the business discover problems after the outcome?
Which process contains enough history?
What signals exist before the outcome?
What intervention is still possible?
Can the outcome be measured?
Who would act on the warning?

When we would advise against building a model

No intervention exists by the time the signal would appear
The outcome is already fixed when the prediction would arrive
Historical data cannot support the prediction target
The cost of predicting exceeds the value of knowing
The decision occurs too rarely to justify an operating capability
Prediction would create unacceptable operational or ethical risk

Where reporting, a rule or better process discipline would solve the problem, we will say so.

Predictive models

Focused intelligence for specific operational decisions.

Three warehouse and fulfilment models, each built around a decision an operational leader already makes. Outputs shown on this page are illustrative examples, not customer results.

Warehouse intelligence

Demonstration

Warehouse Slotting Advisor

Model type:: Optimisation and recommendation model
Evidence status:: Demonstration

Are products stored in the right places for the way customers order today?

What it enables

Analyses historical pick, put-away and movement data against the current warehouse layout to identify potential re-slotting opportunities.

Predictive value

This model analyses demand and movement that has already happened. It prioritises where layout inefficiency exists today rather than forecasting future demand. If demand forecasting is incorporated later, that would add a predictive dimension it does not currently claim.

Decision impact

Gives warehouse managers a prioritised view of potential re-slotting actions for investigation.

Signals it may use

Pick frequency
Order patterns
Items frequently picked together
Bin location
Travel path
Movement history
Velocity
Current layout
Product characteristics where available

Model limitations

Recommendations depend on the quality of bin, movement and location data. Physical constraints and product handling rules are decided by the warehouse team.

Capacity intelligence

Demonstration

Warehouse Labour & Backlog Forecast

Model type:: Forecasting model
Evidence status:: Demonstration

Where is workload likely to exceed available capacity?

What it enables

Uses current workload, historical throughput and relevant operating conditions to estimate where backlog or labour pressure may emerge.

Predictive value

Estimates where future workload may exceed available operational capacity, with uncertainty expressed in the way the method supports.

Decision impact

Helps warehouse leadership investigate where capacity, priorities or workload may need attention. It does not schedule people.

Signals it may use

Open orders
Order lines
Pick workload
Historical throughput
Shift capacity
Backlog
Cut-off times
Work mix
Warehouse zone
Day and time patterns
Known operational constraints

Model limitations

Unrecorded capacity changes, absence and unplanned work reduce forecast usefulness. Resourcing decisions stay with leadership.

Fulfilment intelligence

Demonstration

Pick Wave SLA Risk Predictor

Model type:: Risk prediction model
Evidence status:: Demonstration

Which warehouse work is most likely to miss its required service window?

What it enables

Uses order, wave, workload and execution signals to estimate which work deserves earlier attention because it may be at greater risk of missing its required service window.

Predictive value

Ranks higher-risk work before the service failure is confirmed. A ranking score is not a probability unless a deployed model has been calibrated and presented that way.

Decision impact

Helps operational teams focus attention where intervention may still change the outcome.

Signals it may use

Wave size
Order priority
Lines
Items
Current progress
Available time
Warehouse zone
Historical completion patterns
Workload
Resource pressure
Cut-off
Exceptions

Model limitations

Risk is estimated, not certain. Both false positives and false negatives carry an operational cost, so thresholds are agreed with the operation.

Evidence status

Be clear about what has been proved and what has not.

Every model on this page sits at Level 1. They demonstrate the approach and the operating experience using synthetic data. We have no verified customer outcomes to publish for Predictive Intelligence yet, and we will not imply otherwise.

Level 1
Demonstration
Shows the concept and the intended operating experience using synthetic data.
Level 2
Proof of value
Tested against representative or customer data, compared with an agreed baseline.
Level 3
Production use
Operating inside a real business workflow with an owner and monitoring.
Level 4
Verified outcome
A measured business outcome exists against the original baseline.

Warehouse Slotting Advisor

Are products stored in the right places for the way customers order today?

Movement intensity by zone, shown as the current layout and a recommended pattern for the team to investigate.

Optimisation model · warehouse layout

Zone movement map

Movement intensity is an unitless index derived from historical pick and put-away activity. It describes today's demand pattern rather than forecasting future demand.

Synthetic demonstration data

Zone A1 · front pick face
High movement
Highest movement, closest to dispatch
Zone A2 · front reserve
Moderate movement
Mixed velocity, some slow movers held forward
Zone B1 · mid aisle
Low movement
Capacity available for faster items
Zone B2 · mid reserve
Low movement
Stable, mostly replenishment stock
Zone C1 · rear aisle
High movement
Frequently picked items sitting furthest from dispatch
Zone C2 · rear reserve
Low movement
Low movement, appropriate for slow lines

The recommended view redistributes movement towards the zones with the shortest travel to dispatch. It is a prompt to investigate, not an instruction to relayout the warehouse.

Recommendation

Illustrative example

Item and current location: Item 40218 · Zone C1, rear aisle
Signal: High pick frequency
Observation: The current location creates avoidable travel relative to the demand pattern.
Recommendation: Review a move to a faster-access pick zone.
Why: Frequently picked, commonly co-picked with items in Zone A1, high contribution to total travel.
Signal strength: High confidence · twelve months of movement history
Potential effect: Directional only. Any operational effect should be measured against a baseline before it is claimed.

Warehouse Labour & Backlog Forecast

Where is workload likely to exceed available capacity?

Expected workload against available capacity by shift window, with the periods worth investigating marked in words as well as colour.

Forecasting model · capacity

Expected workload against available capacity

Forecast horizon: next six shift windows. Values are an illustrative workload index, not hours, lines or labour units.

Synthetic demonstration data

Mon AM
State: Normal
Expected workload62
Capacity80
Illustrative range
55 to 70
Contributing signals
Order intake close to the usual pattern
Time still available to act
No action expected
Expected workload within available capacity
Mon PM
State: Normal
Expected workload74
Capacity80
Illustrative range
64 to 85
Contributing signals
Carry-over from the morning wave
Time still available to act
Around 6 hours before the window opens
Expected workload within available capacity
Tue AM
State: Watch
Expected workload81
Capacity80
Illustrative range
70 to 94
Contributing signals
Higher order intake. Two replenishment runs due
Time still available to act
Around 18 hours before the window opens
Expected workload close to available capacity
Tue PM
State: Risk
Expected workload96
Capacity80
Illustrative range
82 to 111
Contributing signals
Promotional order lines. Backlog carried from the morning
Time still available to act
Around 24 hours before the window opens
Expected workload above the illustrative risk boundary
Wed AM
State: Risk
Expected workload104
Capacity84
Illustrative range
88 to 122
Contributing signals
Peak intake day. Backlog accumulating across zones
Time still available to act
Around 42 hours before the window opens
Expected workload above the illustrative risk boundary
Wed PM
State: Watch
Expected workload88
Capacity84
Illustrative range
73 to 104
Contributing signals
Residual backlog. Cut-off pressure on next-day orders
Time still available to act
Around 48 hours before the window opens
Expected workload close to available capacity

The range shown is illustrative. A deployed forecast should express uncertainty in a way the underlying method genuinely supports, rather than presenting a single confident number.

Watch and Risk boundaries here are illustrative. In production they are agreed with the operation against the cost of a false alarm, the cost of a missed problem, the service requirement and the capacity available to intervene.

The forecast highlights where pressure may build. Decisions about resourcing, priority and workload remain with warehouse leadership.

Pick Wave SLA Risk Predictor

Which warehouse work is most likely to miss its service window?

A ranked worklist with the contributing factors, the time still available and the attention the model suggests.

Risk prediction model · fulfilment

Work most likely to need attention

Synthetic demonstration data

Wave 4182 · next-day cut-off
High risk · illustrative score 72/100
Time remaining
1h 40m to cut-off
Progress
38% complete
Evidence behind the score
Data basis: wave, workload and completion history for this zone
Contributing signals
- Backlog increasing in the zone
- Wave larger than the usual profile
- Current throughput below the expected level
Recommended attention. Review resource or priority allocation for this wave.
Wave 4176 · standard service
Medium risk · illustrative score 48/100
Time remaining
3h 05m to cut-off
Progress
55% complete
Evidence behind the score
Data basis: open exceptions and pick path context
Contributing signals
- Two open exceptions on lines
- Pick path crosses a congested aisle
Recommended attention. Check the open exceptions before they hold the wave.
Wave 4169 · standard service
Low risk · illustrative score 17/100
Time remaining
5h 20m to cut-off
Progress
74% complete
Evidence behind the score
Data basis: progress against the historical completion pattern
Contributing signals
- Progress ahead of the historical completion pattern
Recommended attention. No action expected. Monitor only.

A risk score ranks attention. It should not be read as a probability unless a deployed model has specifically been calibrated and presented that way.

Contributing signals show what the model associated with the output. They are evidence for investigation rather than proof of cause.

Prediction should communicate uncertainty rather than create false certainty.

Opportunity areas

Where else operational history may contain a useful signal.

These are areas we can assess. They are not currently packaged products.

Slotting
Backlog
SLA
Labour
Capacity
Replenishment

Decision support

The model highlights the signal. The person owns the decision.

Prediction supports the decision. It does not remove ownership of it.

Identify
The model highlights the signal.
Explain
The model shows the contributing factors.
Review
The responsible person examines the evidence.
Decide
The person owns the decision.
Act
The action is taken in the operating system.
Capture
The outcome is recorded for learning.

Identify
The model highlights the signal.
Explain
The model shows the contributing factors.
Review
The responsible person examines the evidence.
Decide
The person owns the decision.
Act
The action is taken in the operating system.
Capture
The outcome is recorded for learning.

Recommendation versus automation

Not every recommendation should execute automatically.

Level 1
Insight
Show the risk.
Level 2
Recommendation
Suggest what to investigate.
Level 3
Assisted action
Prepare the action for approval.
Level 4
Controlled automation
Execute approved low-risk actions under defined rules. Not appropriate for every use case.

Model workspace

Every model has a place to be reviewed, acted on and evidenced.

What needs attention, what changed, how serious it is, what is driving it and what to investigate first.

What needs attention, and what changed since last time
How serious the signal is and what is driving it
Risk distribution, trend and top flagged records
Model version, evidence basis and uncertainty where the method supports it

Models change because operations change

Every prediction needs an owner.

A prediction model is an operational capability, not a one-off development exercise.

A production model should be monitored across input drift and prediction distribution
Model performance, false positives, false negatives and calibration where probabilities are presented
Threshold behaviour, adoption, outputs reviewed, actions taken and outcomes captured
Model version and cost of operation
A prediction model is an operating capability, not a one-off development exercise.

Use the data already created by the operation

Predictive Intelligence starts with operational signals.

The intelligence should follow the decision, not the ERP brand. Where the required data can be accessed responsibly from another platform, a model may still be viable without a replacement programme.

Systems

Dynamics 365 Business Central
Dynamics 365 Finance
Dynamics 365 Supply Chain Management
Dynamics 365 CRM
Dataverse
Microsoft Fabric
Warehouse systems
Other ERP
Other operational systems

Signals

Transactions
Orders
Inventory
Movement
Capacity
Demand
Supplier performance
Production
Customers
Cases
Finance
Historical outcomes

Business Central

Sales and purchase orders, items, inventory, warehouse entries, locations, bins where applicable, planning data, customers, vendors and financial information.

Explore Business Central

Finance & Supply Chain

Warehouse work, waves, inventory, products, orders, production, planning, procurement, suppliers, finance and operational history.

Explore Finance & Supply Chain

Microsoft Fabric

Where the signal crosses systems, Fabric may provide governed data from multiple sources. It is not mandatory for every model.

Explore Data & AI

Where prediction fits

From transaction to measured outcome.

Prediction is one layer of a wider data and AI maturity picture.

01
Transaction
What happened?
02
Reporting
What does it mean?
03
Automation
What can happen automatically?
04
Prediction
What may happen?
05
Recommendation
What deserves attention?
06
Action
What are we doing about it?
07
Outcome
Did it help?

See the wider Data & AI picture

What is different about prediction

Five capabilities that are easy to blur and should not be.

Power BI
Helps me understand the business.
Generative AI
Helps me understand and create information.
Automation
Executes predictable processes.
Predictive Intelligence
Helps me see a future operational risk or opportunity earlier.
AI agents
Coordinate information and approved actions towards a defined task.

Your operation may contain a better use case

The strongest predictive model may be one specific to your business.

A use case should not proceed simply because data exists.

Would knowing earlier actually change anything?
Which decision is repeatedly made too late?
Where do managers rely heavily on experience?
Where does the business discover problems after the outcome?
Which process contains enough history?
What signals exist before the outcome?
What intervention is still possible?
Can the outcome be measured?
Who would act on the warning?

Model viability

Decision value
Data availability
Signal quality
Actionability
Measurability
Risk

Prioritisation

Business value against predictive readiness.

A simple way to decide what to prove now, what needs data preparation first and what should be left alone.

High value · high readiness
Prove now
High value · low readiness
Prepare data
Low value · high readiness
Low priority
Low value · low readiness
Do not build

Prove the signal

Before scaling the model, prove that earlier warning changes a real decision.

Decision
The decision currently being made too late.
History
The historical data available to test the signal.
Baseline
How the decision performs today.
Model
The predictive approach against that decision.
Review
Business review with the people who would act.
Action
What somebody does differently.
Outcome
Measured against the baseline.
Verdict
Is the measured improvement worth the cost and effort of operating the capability? Go, change or stop.

Decision
The decision currently being made too late.
History
The historical data available to test the signal.
Baseline
How the decision performs today.
Model
The predictive approach against that decision.
Review
Business review with the people who would act.
Action
What somebody does differently.
Outcome
Measured against the baseline.
Verdict
Is the measured improvement worth the cost and effort of operating the capability? Go, change or stop.

The value hypothesis we test.

Current decision: The decision the business already makes.
Current problem: Why the decision is currently worse than it could be.
Current timing: When the problem is discovered today.
Earlier signal: The signal that could realistically arrive sooner.
Intervention: What somebody could still do at that point.
Behaviour change: What would be done differently.
Measurable outcome: The result that would move, and how it is measured.
Cost to operate: What running the capability would cost each year.

Value is judged against the cost of operating it.

Model and data cost
Integration cost
Monitoring and support cost
Operational effort to review outputs
Measured change in the business outcome

A model that works technically but cannot pay for the effort of running it is not a success.

Measure the current decision before claiming improvement.

How the decision is currently made
Current warning time
Current outcome
Current effort
Current intervention
Current error and failure pattern

Without a baseline, we do not claim value.

Put prediction where the decision happens.

Predictive Intelligence workspace
Power BI
Dynamics 365 workspace
Business Central
Teams
Email
Power Apps
Operational dashboards
API and workflow

More alerts do not create better decisions. If everything is high risk, nothing is high risk.

Implementation

From a decision worth improving to a monitored capability.

Discover
Find the decision.
Qualify
Assess value and viability.
Baseline
Understand current performance.
Prepare
Prepare the relevant data.
Model
Develop the predictive approach.
Evaluate
Test technical and operational usefulness.
Prove
Use with representative users against the baseline.
Integrate
Put the output where the decision happens.
Deploy
Release with an owner, thresholds and controls agreed.
Operate
Run it as a supported capability.
Monitor
Track model behaviour, usage and business outcome.
Improve
Refine as operating conditions change.

Discover
Find the decision.
Qualify
Assess value and viability.
Baseline
Understand current performance.
Prepare
Prepare the relevant data.
Model
Develop the predictive approach.
Evaluate
Test technical and operational usefulness.
Prove
Use with representative users against the baseline.
Integrate
Put the output where the decision happens.
Deploy
Release with an owner, thresholds and controls agreed.
Operate
Run it as a supported capability.
Monitor
Track model behaviour, usage and business outcome.
Improve
Refine as operating conditions change.

After deployment

A model is a capability to operate, not a project to finish.

Operating conditions change, so a deployed model is reviewed on a defined cycle.

Retrain
Performance has drifted but the decision still matters.
Recalibrate
Thresholds or probabilities no longer match the operation.
Replace
A better approach or a simpler solution is available.
Suspend
Conditions have changed enough that outputs cannot be trusted.
Retire
The decision has gone, or the cost now exceeds the benefit.

Retirement is a valid and successful outcome. Keeping a model running after it has stopped earning its place is a cost, not an achievement.

Technical assurance

How we judge whether a model can be trusted.

Detail for the people who will be asked to sign this off.

What we need from you

Predictive work depends on operational knowledge we do not have.

These responsibilities sit with the customer. Where they are not in place, timelines and confidence change.

Decision owner
Owns the business decision the model is intended to improve, and what happens when it flags something.
Operational expertise
Explains operating conditions, exceptions and why the historical data behaves the way it does.
Data owner
Owns the meaning, quality and source knowledge behind the information the model uses.
Outcome owner
Ensures the real outcome is captured so the model's usefulness can be evaluated honestly.
IT and security
Provides appropriate access and governs the connected systems and environments.
Model or service owner
Owns the deployed capability, the review cycle and operational escalation.
Users
Review outputs, decide actions and give feedback on what was useful and what was noise.

Confidence gates

Four points where continuing has to be justified.

Each gate has a defined decision, including the option to stop.

Gate 1
Before qualification closes
Decision and value
Establish that a real decision exists, and that prediction is not being applied to something reporting or automation would solve more simply.
Confirmed at this gate
- The decision and its owner
- The current problem and current baseline
- The intervention still available
- The potential outcome if the warning arrived earlier
- Whether reporting or automation would be sufficient
Decision
Proceed · Use a simpler capability · Reframe the decision · Stop
Gate 2
Before modelling
Data and signal
Establish that the history exists, that the prediction target is definable and that the outcome can be captured.
Confirmed at this gate
- Historical information and coverage
- Prediction target and outcome data
- Timing of the signal against the decision
- Quality, access and security
- Whether a useful signal is plausible at all
Decision
Proceed · Prepare the data first · Change the target · Stop
Gate 3
After the proof of value
Proof
Establish whether earlier warning changed a real decision, technically, operationally and commercially.
Confirmed at this gate
- Technical: performance, error profile, calibration where relevant and robustness
- Operational: users understand the output, alerts are actionable, thresholds are sensible and intervention remains possible
- Business: behaviour changed, the outcome can be measured against the baseline and the value appears sufficient
Decision
Production candidate · Improve and repeat · Stop
Gate 4
Before deployment
Production
A model entering the operation needs an owner and an operating model, not simply a working model.
Confirmed at this gate
- Owner, users and permitted actions
- Security, integration and environment
- Model version and agreed thresholds
- Monitoring, support and cost
- Incident response, fallback, rollback and change control
Decision
Deploy · Remediate first · Do not deploy

Predictive assessment

Find out whether prediction is the right answer.

Seven short questions. The result may be a proof of value, a data foundation step, or advice that a simpler capability would serve you better.

1. What would you like to know earlier?

2. Which business area does that decision sit in?

3. When do you currently discover the problem?

4. Where does the history live?

5. Could somebody still intervene if the issue were known earlier?

6. Can the eventual result be measured?

7. Where are you today?

Indicative signal

Answer the questions and we will show a likely next step. This is a recommended route rather than a prediction-readiness verdict, and sometimes the honest answer is that reporting, automation or a data step comes first. Please keep every answer at organisational level: no customer, personal or case information.

Your answers are carried through, so you will not be asked to repeat them. Final qualification is always a conversation.

Where this connects

Prediction rarely arrives on its own.

Data & AI
Where trusted data, reporting, automation and AI need to be built in the right order.
Explore
Optimise
Where an existing Microsoft platform already contains untapped operational value.
Explore
Transform
Where the operating model and priorities behind the data still need defining.
Explore
Recovery
Where a data or analytics programme has lost clarity, confidence or control.
Explore

Evidence

Model demonstrations and customer evidence, kept separate.

Predictive outputs shown on this page are demonstrations and illustrative examples. The customer videos below speak to our Microsoft delivery work, not to Predictive Intelligence deployments.

Customer voice

What customers say about working with us.

Customer voices

InteliSense customers

Hear our customers talk about InteliSense

Why this is different

We understand the systems that create the operational signal.

ERP knowledge
We understand the transactions beneath the model.
Operational context
We understand why the data was created.
Data and AI
We turn operational history into usable intelligence.
Decision first
We start with what somebody needs to decide.
Explainability
We design outputs people can investigate.
Human control
The business remains accountable for the decision.
Integration
We put the insight into the operating workflow.
Continuous evaluation
We monitor whether the model remains useful.

Common questions

Questions leaders ask about Predictive Intelligence.

What would you want to know earlier?

Find the decision you are currently making too late.

Show us an operational decision that depends on ERP, warehouse, supply chain, finance, sales or service data. We can assess whether the history already contained in your systems provides enough signal to help you see the issue earlier.

See what may need attentionbefore it becomes obvious.

What this page answers.

Most operational systems tell you what happened. The difficult question is what deserves attention next.

Warehouse

Inventory

SLA

Labour

Suppliers

Cash

Pipeline

Reporting is not replaced. It is extended.

Reporting

Real-time insight

Prediction

Decision support

Action

Turn operational history into earlier warning.

Signals

Model

Risk

Explanation

Attention

Decision

Outcome

Learning

Signals

Model

Risk

Explanation

Attention

Decision

Outcome

Learning

A useful model answers more than one question.

What?

Why?

When?

Impact?

Action?

How certain?

Different questions need different models.

Forecasting

Risk prediction

Optimisation and recommendation

Prediction is only useful when somebody can still act.

Questions we work through

When we would advise against building a model

Focused intelligence for specific operational decisions.

Warehouse Slotting Advisor

Warehouse Labour & Backlog Forecast

Pick Wave SLA Risk Predictor

Be clear about what has been proved and what has not.

Demonstration

Proof of value

Production use

Verified outcome

Are products stored in the right places for the way customers order today?

Zone movement map

Where is workload likely to exceed available capacity?

Expected workload against available capacity

Which warehouse work is most likely to miss its service window?

Work most likely to need attention

Where else operational history may contain a useful signal.

Warehouse

Supply chain

Manufacturing

Finance

Sales

Service

The model highlights the signal. The person owns the decision.

Identify

Explain

Review

Decide

Act

Capture

Identify

Explain

Review

Decide

See what may need attention
before it becomes obvious.