Prescriptive analytics
Definition
Prescriptive analytics
Prescriptive analytics is the branch of analytics that tells you what to do next. It takes a forecast, adds rules and limits, then ranks the options by value. The output is a decision, not a chart. Most teams stop one rung short of that.
It sits at the top of the four-stage analytics maturity model, above descriptive, diagnostic, and predictive work. Yet Gartner’s 2023 analytics framework put fewer than 15% of large enterprises on operational prescriptive workloads by 2024.
That gap is the opening. Once a predictive model flags a customer as likely to churn, prescriptive analytics names the offer, the discount, and the channel to use — and ranks it against the alternatives.
For outsourcing buyers, the term matters because the work is now sold as a service. Analytics providers in Manila, Bengaluru, and Kraków staff the modelling and optimisation layers that most mid-market firms never build in-house.
Key takeaways
- Prescriptive analytics returns a ranked action, not a dashboard, by pairing prediction with optimisation and business rules.
- Fewer than 15% of large enterprises had operationalised prescriptive workloads by 2024, on Gartner’s count.
- The standard pipeline runs four layers: data, predictive model, optimisation engine, and decision interface.
- Firms embedding machine learning in prescriptive workflows reported 1.5 to 3 times the cost impact of descriptive-only AI.
- Pilots start under $50,000, while production systems typically run $250,000 to $1 million in the first year.
How it works
Prescriptive analytics feeds a predictive model’s output into an optimisation engine bounded by business rules, then ranks each candidate action by expected value. The engine tests thousands of combinations and surfaces the move that best serves your objective.
| Stage | What it does | Common tools |
|---|---|---|
| Data layer | Pulls historical and real-time signals into a feature store | Snowflake, Databricks, BigQuery |
| Predictive model | Forecasts likely outcomes such as churn, demand, or fraud | scikit-learn, XGBoost, TensorFlow |
| Optimisation engine | Searches the action space under stated constraints | Gurobi, CPLEX, Google OR-Tools |
| Decision interface | Surfaces the ranked recommendation to a person or system | Tableau, Power BI, custom apps |
| Feedback loop | Measures what happened after the action, then retrains | MLflow, Airflow, dbt |
Business rules do the unglamorous work. They stop the optimiser recommending something legal but daft, like a discount below cost or a shift pattern that breaks labour law, and they encode the constraints your operation lives with.
Machine learning matters because the optimisation engine needs accurate probabilities to weigh each action. The 2023 McKinsey State of AI report found ML-driven prescriptive workflows delivered 1.5 to 3 times the cost impact of descriptive-only AI.
The link to predictive analytics is direct: prescriptive work consumes predictive output and acts on it. Without solid forecasting upstream, the recommendations downstream are guesses dressed up in maths — expensive ones at production scale.
Examples
Prescriptive analytics turns up wherever a decision repeats at scale and the cost of getting it wrong is measurable. Routing, pricing, scheduling, and underwriting are the classic homes, and the payback shows up in fuel, margin, or service level.
UPS ORION, the parcel giant’s route-optimisation system, has run on prescriptive models since 2013 and saves roughly 10 million gallons of fuel a year, per UPS’s own 2022 sustainability disclosures. Drivers get a turn-by-turn sequence, not a heatmap.
Netflix’s recommendation engine is prescriptive at the action layer. It doesn’t only predict which titles you’ll like; it decides the artwork, the row placement, and the trailer you see.
Netflix has publicly attributed more than $1 billion a year in retained subscriber value to that system, which makes the recommender a revenue engine rather than a convenience.
Maersk, the Danish container line, uses prescriptive models to reroute ships around port congestion and weather. It reported in early 2024 that voyage times fell 4–8% on affected lanes during the Red Sea diversions.
Hospital groups run the same loop on theatre scheduling, matching predicted case length and staff availability against a prescribed roster that keeps operating rooms full without pushing clinicians into unsafe overtime.
Retail pricing teams use the same shape at the shelf edge: a demand forecast per SKU, a margin floor, then a prescribed markdown schedule that clears stock without gutting the category.
Philippine BPO operators now run prescriptive workforce models on agent scheduling: forecast call volume per skill queue, then prescribe exact headcount, breaks, and overtime offers to hit SLA at the lowest cost.
It’s a natural fit for knowledge process outsourcing teams that already own the client’s data, the models, and the reporting layer.
Buyers who’d rather shortlist a provider than build the capability can talk to Outsource Accelerator about analytics teams already running these models.
The pattern holds across all of them: a prediction is interesting on its own, but only a prescription is specific enough to get actioned in production.
Related terms
These adjacent terms round out the analytics stack, and they turn up in the same procurement conversations. Knowing where prescriptive analytics sits against each one keeps scoping honest when you brief a vendor.
- Predictive Analytics: the forecasting layer sitting directly below prescriptive work in the maturity model.
- Descriptive Analytics: the dashboards and KPI reports that summarise what already happened.
- Diagnostic Analytics: the explanatory layer answering why a result moved the way it did.
- Business Intelligence: the broader discipline of turning raw data into commercial decisions.
- Data Analytics: the umbrella term covering all four stages of analytics maturity.
- Incremental Analysis: the marginal cost and benefit technique prescriptive engines automate at scale.
- Knowledge Process Outsourcing: the offshore delivery model for high-judgement analytics work.
FAQ
What’s the difference between predictive and prescriptive analytics?
Predictive analytics estimates the probability of an outcome, say a 70% chance a customer churns next month. Prescriptive analytics takes that number and recommends a specific move — a 15% loyalty discount by email on Tuesday — then ranks it against every alternative.
Do you need machine learning to do prescriptive analytics?
Not strictly. Simple systems run on linear programming and hand-coded business rules alone. Machine learning sharpens the predictive inputs and lets the optimiser handle non-linear, high-dimensional problems that rule-based tools cannot scale to.
How expensive is it to set up?
A pilot on existing cloud infrastructure can launch under $50,000 with off-the-shelf libraries like Google OR-Tools. Production systems with commercial solvers and dedicated engineers run $250,000 to $1 million in year one, per benchmarks Forrester published in 2024.
Which industries benefit most?
Logistics, financial services, retail, healthcare, and telecom lead adoption, because their core decisions repeat millions of times a day: routing, pricing, underwriting, and inventory. Any sector with a repeatable decision and a measurable objective qualifies.
Can you outsource prescriptive analytics?
Yes, and specialist providers across the Philippines, India, and Eastern Europe deliver the modelling, optimisation, and reporting work as a managed service.
If you want prescriptive analytics running without hiring a data science team first, Outsource Accelerator can point you toward vetted providers.
Independent
