Predictive and Advanced Analytics Consulting

Advanced analytics consulting is the work of turning data into decisions that actually change: naming the commercial decision being made worse than it could be, establishing what a better version of it is worth, and building only the prediction that moves it. This page covers the commercial and customer half of that, where prediction touches revenue, price and inventory rather than infrastructure.

• Since 1999demand, commercial and customer strategy
• 200+published case studies
• 10 officeson four continents
• No vendor tiesour model and platform advice carries no commercial stake
• The Interaction Fieldwritten by our founder

A forecast is worth what the decision was worth

The model performs, the dashboard refreshes, and the decision it was built to improve is still made the way it was always made: by the same person, on the same instinct, at the same point in the week.

That is a scoping failure, not a technical one. Work that starts from available data produces models looking for a use. Work that starts from a named recurring decision produces models that get used, because the person who acts on the output was in the room when the threshold was set.

Three questions decide whether a prediction program is worth funding:

• Which recurring commercial decision is being made worse than it could be, and who owns it
• What would have to be true for it to change: accuracy, latency, confidence, and how the answer arrives in the workflow
• What a better decision is worth

If the third question has no answer, do not build the model.

When companies bring us in

• The models perform and nothing changes. A capable data science function, good accuracy scores, and a P&L that looks exactly as it did three years ago.
• Before a platform or tooling commitment. A decision is about to be made on infrastructure before anyone has named the decisions it is meant to improve.
• When pricing or promotion is run on instinct. The data exists to predict response by segment and nobody has connected it to the weekly call.
• When analytics has to justify itself. The function is being asked for a return and has only accuracy metrics to offer.

What our predictive analytics consulting services include

Four bodies of work. Most engagements start with the first and stop where the value runs out.

1. A decision inventory and a value case

Before any modeling, we build the inventory of recurring commercial decisions, who owns each one, how it is made today and what a better version would be worth.

• The decisions ranked by value at stake rather than by data availability
• The accuracy, latency and confidence each decision actually requires, taken from the person who makes it
• The decisions that should not be automated, named explicitly, so the program is not judged against them

2. Prediction as the sensing half of a loop

We design intelligence as a sense, reason, act and learn loop. Prediction is the sensing half, and it absorbs almost all of most budgets, which is why so many analytics functions look busy and change so little.

• Sense tells you what is likely
• Reason turns that into a recommended action inside a business context the model does not hold
• Act needs someone, or something, with authority to move
• Learn needs the outcome to come back, so the next prediction starts better informed

Organizations are usually strong at sense, thin at reason, blocked at act and absent at learn. The blockage at act is a permissions problem wearing a technical costume. Nobody has decided which calls a system may make alone, so everything queues behind a person who is busy. So we set the escalation threshold explicitly, on materiality, reversibility, precedent, exposure and the system's own confidence. Routine decisions then clear without review, consequential ones arrive at the right level with the reasoning and alternatives assembled, and the exception path is known before it is needed.

3. Prediction that looks outside the category

Commercial prediction usually runs inside the categories a company already organizes around: churn on this product, uplift on that promotion, next best action from the existing range. All useful, all bounded, because it can only optimize demand you already capture.

• Demand spaces describe what customers are trying to get done, and the profitable ones cut across the category lines incumbents are structured around
• A model that sees a need forming before it reaches your category is a growth instrument. One that only sees the customer leaving is a retention instrument, and boards do not pay the same multiple for those two stories
• Price belongs in the same frame: what a preference premium is worth, by segment and demand space, is measurable, and it connects brand strength to economic profit

4. An architecture that assumes the models will change

Capability that was frontier not long ago is now mid-tier and available from several suppliers, so we design for models chosen per workload, where moving is a deployment decision rather than a rebuild.

• The durable assets transfer anyway: the decision inventory, the labeled outcome data, the workflow integration and the feedback path
• Classical machine learning stays the right tool for structured prediction at volume, being cheaper, faster and more explainable, while language models handle unstructured input
• We hold no partnership with any model provider, so our advice on model choice carries no commercial stake

How an advanced analytics engagement works

1. Inventory. Interviews with the people who make the recurring commercial decisions, a map of how each is made today, and a ranking by value at stake.
2. Specify. For the top decisions, the accuracy, latency and confidence required, the form the answer has to arrive in, and the escalation threshold. The output is a build specification written from the decision backwards.
3. Build. The prediction, the reasoning layer around it and the workflow integration, with model choice made per workload rather than per supplier relationship.
4. Close the loop. Outcome capture, so the system learns whether its last recommendation was any good, and a measurement line from the decision to the P&L.

Why companies choose Vivaldi for analytics work

• We start at the decision, not the data. The value case comes before the model, which is the step that decides whether anything gets used.
• We are not reselling a platform. No model or infrastructure partnership sits behind our recommendation, so the architecture advice is about your workload rather than our margin.
• Prediction is connected to demand, not only to the funnel. Our demand space work means the models look for needs forming outside your category, which is where growth prediction differs from retention prediction.

How to evaluate a data analytics consulting firm

Worth asking of any firm on your shortlist, including us.

• Ask them to name the decision before the model. A proposal that opens with a data audit and closes with a platform recommendation has skipped the question that decides whether any of it gets used.
• Ask what accuracy the decision requires. A firm that cannot answer has not spoken to the person who owns the decision, and is optimizing a metric nobody will act on.
• Ask what happens when the model needs replacing. An architecture that assumes one provider indefinitely carries a supplier risk with no return attached.
• Ask what they will do about outcome data. The item most often missing is whether the last recommendation was any good, and without it nothing improves.
• Ask who they are paid by. A recommendation from a firm with a platform partnership is not the same artifact as one without.

Analytics strategy consultancy, data engineering firm or platform vendor?

Where this page stops and machine learning starts

This page covers commercial and customer prediction inside connected commerce: demand and sales forecasting, lifetime value and churn, propensity and next best action, price and promotion response, inventory and assortment. General machine learning, model development and evaluation sit on machine learning, predictive analytics and AI strategy. Most organizations need both. If the question is commercial, start here.

Related work

• Machine learning, predictive analytics and AI strategy for general model development, evaluation and AI model strategy.
• Data strategy if the constraint is the state of the data rather than the state of the decisions.
• Dynamic customer journey strategy if the prediction is sound and nothing in the journey acts on it.
• Connected commerce for the wider group this page sits in, whose job is making customer context travel between channels.

This work sits inside our Business Technology Strategy practice at Vivaldi, the strategy firm behind the Demand Landscape and the Interaction Field.

Advanced analytics consulting in the US and worldwide

Our analytics work runs out of ten offices on four continents, with the US team in New York. Commercial prediction is market-specific in a way that infrastructure is not: the same churn model trained on one market usually misprices risk in another, because the alternatives available to the customer are different. We scope by market and reuse the decision inventory rather than the model.