Services
JTP Analytics provides independent marketing measurement support for organisations that already have meaningful data and need better answers from it. Engagements can cover a complete modelling project, a focused analysis, an independent audit or ongoing support for an internal or agency team.
Bespoke modelling
Models are built around the decision you need to make. Explore three common starting points below; the data, assumptions and deliverables are agreed during scoping.
What it is
Marketing Mix Modelling (MMM) predicts sales or another key performance indicator (KPI) using historical outcomes, marketing spend and relevant business factors. It models delayed effects, diminishing returns and uncertainty so the predictions can inform planning.
The question it answers
How might sales or another KPI change at different spend levels, and how should the budget be allocated across channels?
When to use it
When several channels are active, you have enough consistent history to model them, and you need evidence for strategic budget decisions.
What you’ll receive
A validated predictive model with documented assumptions, channel contribution estimates, marginal response curves and budget allocation scenarios, all with appropriate uncertainty. Experimental results can help calibrate the model where available.
As total budget changes, the optimiser reallocates spend toward similar marginal returns across channels.
What it is
Customer lifetime value (CLV) modelling estimates future value from the probability of another purchase, expected spend, retention and margin. It can update those expectations as new purchases occur and makes uncertainty visible.
The question it answers
Which customers create lasting value, and how much can you reasonably invest to acquire or retain them?
When to use it
When you have customer or cohort history over time and need to compare acquisition quality, set investment limits or prioritise retention work.
What you’ll receive
Value forecasts by useful cohort or segment, retention and purchase assumptions, uncertainty ranges, and a clear view of which inputs matter most. The outputs can inform acquisition and retention decisions.
A repeat purchase updates the forecast. The shape and timing are estimated from customer history.
What it is
Product adoption modelling forecasts how a launch may spread over time. A Bass diffusion model distinguishes innovators, who adopt independently, from imitators, whose adoption grows with the existing adopter base. With suitable launch data, it can be extended to estimate how pre-launch marketing may influence market share.
The question it answers
How quickly might adoption grow, when could it peak, and how might pre-launch marketing shift eventual market share?
When to use it
When planning a launch or rollout with comparable product history or early adoption data, especially when the timing of adoption affects supply, staffing or marketing plans.
What you’ll receive
Weekly and cumulative adoption forecasts, ranges for peak timing and market potential, and documented assumptions. Where the data support it, scenarios can show how different pre-launch marketing plans may affect market share.
Contact makes the diffusion idea visible; a fitted Bass model does not track individual interactions.
Incrementality testing
Each method answers the same question: what was the incremental impact of the intervention? The right design depends on where you can withhold activity, what can be randomized and whether a credible comparison exists.
What it is
The Synthetic Control Method (SCM) builds a comparison for one exposed geography from a weighted combination of non-exposed geographies. The weights are chosen to closely match the test geography before the intervention; the difference after launch estimates incremental impact.
The question it answers
What was the incremental impact of the new channel in the test geography?
When to use it
Well suited to geo-experiments where the activity can run in one region while comparable donor regions remain unexposed. A good pre-test fit and limited spillover matter.
What you’ll receive
A test design, pre-period fit checks, an estimated incremental effect with uncertainty, and a clear account of donor-region choices and limitations.
The comparison is built from regions that never receive the new channel.
What it is
Interrupted time series analysis (ITSA) uses the pattern before a known intervention to project what would have happened without it. The observed national outcome is compared with that projected path after launch.
The question it answers
What was the incremental impact of the national intervention, relative to the expected path without it?
When to use it
Useful when an intervention must happen nationally and there are no unexposed control geographies. It needs enough history to understand trend and seasonality, plus careful checks for other changes at the same time.
What you’ll receive
An estimated change in level or trend with uncertainty, a visual comparison to the projected path, and sensitivity checks around seasonality and competing explanations.
The level and slope can both change. Concurrent national changes need particular scrutiny.
What it is
An A/B test randomly assigns eligible people to two experiences. Comparing a pre-selected outcome between the groups estimates the effect of the change while both run at the same time.
The question it answers
What was the incremental impact of the new experience on sign-ups?
When to use it
When you can randomize at the right level, prevent people switching between versions and reach a useful sample size. It works well for digital journeys, messages and customer-level holdouts.
What you’ll receive
A test plan with the primary metric and sample-size target, randomization and quality checks, and an estimate of the difference with uncertainty and a practical recommendation.
The curves illustrate possible counts across repeated tests with equal-sized groups. Actual lift and uncertainty come from the test data.
Model audits and validation
Can you trust this model for the decision you are making? An independent audit looks beyond a headline accuracy score to find where the model works, where it breaks and which conclusions its inputs can support.
What we check
We compare out-of-sample (OOS) predictions and uncertainty with outcomes not used to fit the model, alongside a simple benchmark. The holdout design depends on the model and its intended use, with a separate final check where the data allow.
What it reveals
A model can fit its training data closely yet miss new periods or regions. For Marketing Mix Modelling, predictive fit is a useful check, but it cannot by itself validate causal channel effects.
What you’ll receive
A reproducible holdout design, observed-versus-predicted results, benchmark comparison and a clear view of where performance is strong or weak.
A holdout should reflect the model’s purpose; one headline score is never the whole audit.
What we check
We inspect residuals, the gaps between observed and predicted outcomes, across time, geography, products and recurring calendar periods. Repeated patterns can be more revealing than the overall average error.
What it reveals
Errors that cluster by season, region or product can signal missed structure, a data issue or assumptions that need revisiting.
What you’ll receive
Diagnostic plots, a ranked account of the patterns found and specific checks or model changes that would test the likely explanations.
The repeated pattern shows where to investigate; it does not prove the cause.
What we check
We compare media spend with impressions and implied cost per thousand impressions over time, then test alternative activity inputs. We also inspect input timing and lineage for information leakage.
What it reveals
Spend can rise because of media price inflation even when delivered impressions barely change. A model using spend as the activity input may confuse higher costs with more exposure. Impression-based inputs can better reflect delivered activity when the counts are reliable and comparable.
What you’ll receive
A driver definition and timing review, spend-versus-impression trends, sensitivity checks with alternative inputs and recommendations for a credible media activity measure.
Impressions measure delivery rather than cost. Their quality and comparability still need checking.
Where relevant, the audit also reviews assumptions, priors, response curves, uncertainty and reproducibility. It concludes with prioritised findings and a practical recommendation: use the model, use it with caveats, or repair it before relying on its outputs.
Advisory Roles
JTP can help an in-house team build measurement skills or provide specialist support to an agency on a project or ongoing basis.
What this role covers
Workshops and mentoring help marketing and technical teams understand, challenge and use measurement. Good support should leave the team more capable of making its own decisions.
Typical support
- Marketing measurement foundations for mixed audiences.
- Practical MMM, Bayesian modelling and causal inference training.
- Workshops on interpreting uncertainty and explaining results to senior stakeholders.
- Technical mentoring, model review and office hours for analysts.
How it works
Sessions are tailored to the team’s current knowledge, data and live commercial questions.
Read the Impression case study to see how applied MMM training and model reviews helped build an in-house team’s capability.
What this role covers
Specialist measurement input for agencies that want to strengthen their offering without adding a permanent role immediately.
Typical support
- Behind-the-scenes modelling and independent technical review.
- Method development, project scoping and pitch advice.
- Help answering difficult client measurement questions.
How it works
Support can be scoped to one client project or arranged as ongoing advisory time.
Read the Impression case study to see how specialist collaboration and independent model reviews supported an agency team.
Is JTP a good fit?
JTP is best suited to UK and US organisations with mature data, established marketing activity and stakeholders ready to act on the findings. Direct-to-consumer businesses are a particularly strong fit, although the methods apply across many sectors.
JTP works with no more than four clients concurrently so each engagement receives meaningful senior attention.