|
Описание: |
Role Description
A consultant engagement to derive and validate the values that populate the two deterministic scoring engines behind a video media planning platform.
The formulae are fixed and client-owned. What does not yet exist are the configuration values that make them produce a number a planner can defend to an advertiser — per-channel sub-dimension scores, saturation constants, maximum reach ceilings, universe sizes, and the cross-channel overlap coefficients used for de-duplication. Deriving those values from the source data the client holds, and stating clearly where a value cannot be derived, is the substance of the role.
Work is front-loaded. The first phase establishes what source data exists, what must be derived from it, and what is not obtainable at all. The remainder derives and validates the values, fits the reach curves, builds and checks the overlap matrices, and documents each derivation to a standard the client’s own analysts can review, challenge, and maintain after handover.
This is not a modelling role in the sense of inventing a method. The methodology is the client’s intellectual property and is fixed. The consultant populates and calibrates it, evidences it, and hands it over. Writing production code is not part of the role — a data engineer on the delivery team implements the engines; this role supplies and validates what goes into them.
About the Project
Client: a global media network, engaged through a delivery partner. End users are media planners at one of the network’s agencies.
Product: a video-only, omnichannel media planning web application. A planner defines a campaign, advertiser, budget, flight dates, target audience, and objectives through a six-step wizard. The platform generates budget-optimised plans across video channels, scores them using two deterministic engines, and exports a client-facing Media Plan (PDF) and an operations Activation Plan (XLSX).
Channels in scope: linear TV, BVOD, SVOD, AVOD/FAST, YouTube, online video, social video, and video out-of-home. The market is primarily UK/EMEA.
Scoring: two deterministic engines. A weighted channel quality index built from sub-dimension scores across content, ad experience and data signal quality, weighted by campaign objective; and a reach and frequency model using exponential saturation against effective spend with cross-channel de-duplication. No machine learning is used.
Position today: both engines exist as working spreadsheet prototypes populated with a small illustrative channel set. Extending them to the full production channel inventory, across the audience bands the platform supports, and evidencing every value in the process, is the work.
Delivery: three-month fixed window, sprint cadence, hard go-live date. All values are analyst-maintained through the platform’s admin layer; there are no external data feeds in the first phase.
Key ResponsibilitiesSource data assessment (first)
* Establish what source data exists — what the client holds, what is available from published industry sources, and what is not obtainable at all.
* Assess it for coverage, comparability, recency and reliability, and judge whether it can support the value it is meant to produce.
* State the position on every value — which can be derived from evidence, which must be set as an informed prior, and which cannot be established — with the reasoning behind each.
* Raise data gaps as dependencies with a named owner and a date, early enough to be acted on.
Effectiveness index derivation
* Derive per-channel sub-dimension scores on a 0–10 scale across the content, ad-experience and data-signal dimension groups.
* Derive the attention-based modifier scores applied to ad experience, and the range over which the modifier operates.
* Establish objective-led weighting sets for each dimension group across the four campaign objectives, and the channel-default weightings they blend with.
* Set and justify the scaling bounds that convert a weighted signal score into a multiplier, and the objective-derived weight controlling how strongly signal quality affects the final index.
* Sanity-check the resulting index — that it produces a defensible ordering of channels, responds sensibly to a change in objective mix, and lands within its intended interpretation bands.
Reach and frequency model derivation
* Derive per-channel universe sizes, for all adults and per age band.
* Fit saturation constants per channel against observed reach data, calibrated to the reference campaign duration and frequency.
* Derive maximum reach ceilings per channel and per audience band, reflecting ad-tier penetration, subscriber behaviour and inventory limits.
* Build the pairwise cross-channel overlap matrices per demographic band, and validate them for symmetry, bounds and plausibility.
* Establish per-channel CPM and working-media rates per buying path.
* Set the uncertainty band applied to modelled reach estimates, and justify its width.
Validation
* Validate engine output against expected results across a set of representative test cases, and investigate discrepancies.
* Investigate the known open defect where de-duplicated reach can exceed the audience denominator, and establish whether the cause sits in the derived coefficients, the accumulation method, or the implementation.
* Define the acceptance tests the delivery team validates the engines against, including the expected values.
Documentation and handover
* Document every derivation — source, method, assumptions, confidence and refresh cadence — to a standard the client’s analysts can review and challenge.
* Distinguish a derived value from an estimated one in writing, in the deliverable itself.
* Participate in methodology working sessions with the client’s analysts, and defend a derivation under challenge.
* Hand over cleanly, so the client’s analysts can maintain and refresh the values themselves through the platform’s admin layer once the engagement ends.
Required QualificationsExperience
* 5+ years in a quantitative analyst role where the output was a derived value other people depended on — scores, indices, model parameters, benchmarks — rather than reporting or dashboard production.
* Deriving parameters for a model specified by someone else, and working inside that specification rather than around it.
* Curve fitting to observed data, and judging whether a fit is good enough to rely on.
* Working from incomplete, inconsistent or only partially comparable source data, and stating plainly where the result is weak.
* Producing written analytical documentation that another analyst reviewed and challenged.
* English: strong written and spoken, C1+ effectively. The role runs sessions with client-side specialists and its principal deliverable is written.
Technical Acumen
* Advanced spreadsheet modelling — the methodology is spreadsheet-native, and the deliverables are workbook-based. This is the primary working tool, not a fallback.
* SQL to an analytical standard — aggregation, window functions, and joining across inconsistent sources.
* Python or R for curve fitting, validation and sensitivity checking. Production engineering is not required.
* Weighted scoring and composite index construction — normalisation, weighting, blending, and the failure modes of each.
* Matrix validation — symmetry, bounds and plausibility checks on pairwise coefficient matrices.
* Versioned, effective-dated reference data as a concept — knowing which values a result was produced against, and being able to reproduce a historical result.
Domain Knowledge — required
* Media measurement fundamentals — reach and frequency, audience universes and denominators, cross-channel duplication and de-duplication, impressions and CPM arithmetic. The requirement is the underlying reasoning rather than fluency in any single market’s currency.
* The UK video landscape and its measurement sources — broadcaster and platform reporting, industry panels and cross-platform studies — and a working view of which source is reliable for what.
* Enough media planning literacy to know what a planner does with these numbers, and what makes a value indefensible in front of an advertiser.
Judgement & Soft Capabilities
* Says a value cannot be derived from what exists, rather than producing a plausible number to fill a cell.
* Separates what has been derived from what has been assumed, and labels the difference in the deliverable rather than holding it in their head.
* Documents while working, not afterwards.
* Works within a method they did not design, and raises a concern with it through the people who own it rather than correcting it silently.
* Defends a derivation under challenge from a client-side specialist, and revises it when the challenge is right.
Nice to Have
* Background in a media agency, broadcaster, publisher or measurement vendor.
* Direct experience with UK cross-platform measurement datasets and panel data.
* Experience building or validating an audience duplication or overlap model.
* Prior work on a composite score or index used commercially.
* Experience with attention or ad-effectiveness research.
* Experience handing a model over to a client team to own and maintain.
* Familiarity with video and CTV channel economics — buying paths, working media rates, rate cards.
Ideal Candidate Profile
You are an analyst who has been handed someone else’s model and asked to make it real. You know the hard part is rarely the arithmetic. It is establishing what the available data can honestly support, and being straight about it when the answer is less than everyone hoped for.
You are comfortable saying that a value cannot be derived. You would rather deliver forty defensible numbers and a list of the twelve that need a decision than fifty-two numbers of unstated quality. You label an estimate as an estimate, and you write down what would have to change for it to become a measurement.
You work inside the method you are given. When you think something in it is wrong, you raise it, evidence it, and let the people who own it decide. You do not quietly adjust it and mention it afterwards.
You document as you go, because you know the deliverable is not the number. It is the number plus the reason anyone should believe it, in a form the client’s own analysts can pick up, challenge, and maintain once you have gone.
Відгукнутись на вакансію |
