The single most common reason a data programme runs out of oxygen in year two is that nobody can answer, clearly, what it was supposed to deliver. A value model is the instrument that prevents that. It is also the single most under-invested artefact in enterprise data.
What a value model is — and is not
A value model is a written, quantified statement of the financial outcome a data investment is expected to produce, with the assumptions behind each figure. It is not a business case (which justifies spend), not a forecast (which predicts revenue), and not a benefits register (which lists what might happen). It is narrower and more useful than any of them.
A good value model fits on one page. It contains four sections.
1. The figure the board is asked to believe
One number. The three-year NPV, or the annual operating saving, or the cost of inaction. The board will only ever quote one figure. Pick the one that is most defensible and name it first.
2. The assumption chain
Every input that multiplies into the headline number, with a source. Not "efficiency uplift 12%" — "time spent by the 60-strong underwriting team on submission triage reduces from 31% to 22%, based on timesheet analysis of 140 submissions in Q3, source: the underwriting director." Where a figure is an estimate, label it as such and put a range on it.
3. The dependencies
The operational or organisational changes required for the figure to land. A licence saving of £1.8m depends on the retirement of 388 dashboards. A partner-hours saving depends on partners using the new agent. If the dependency is outside the data team's control, the value model should say so.
4. The measurement plan
How you will prove, month by month, that the value has landed. What baseline will you hold yourself to, what will you report, and to whom. Without this, a value model is a promise. With it, it is an operating instrument.
A worked example
This is a lightly anonymised value-model summary from a recent engagement — a G15 housing association consolidating three post-merger estates.
Each figure had a named owner, a source, and a measurement plan. The platform decision — Databricks on Azure — came after the value model, and was the cheapest option that could realistically deliver the £1.9m of licence consolidation. Other choices would have been defensible against a different value model, but not this one.
Why value modelling comes first
If you have a value model before a technology shortlist, two things become possible:
- Architecture becomes consequence, not choice. You are not picking a platform because a competitor has it; you are picking the platform that delivers the model. This kills the tooling wars and reduces vendor lobbying.
- The programme has a kill switch. If month six shows the model is wrong, you stop. If you have no model, you do not stop — you iterate a technology you cannot justify.
In the last three years, every engagement we have declined at the discovery stage has been one where the client refused to put numbers on the outcome. Every one has since been delivered by another firm and stalled within eighteen months.
When the model is honest about its limits
Not everything is quantifiable. Regulatory risk reduction, brand protection, recruiting attractiveness — these appear in value models as ranges, with explicit acknowledgement of their uncertainty. The goal is not spurious precision. The goal is that every figure on the page has been contested by someone who would have to deliver it.
The one-page template
We give every client a stripped-back template: one page, the four sections above, and a set of questions that must be answered or left explicitly blank. Blanks are useful — they mark the parts of the strategy where you do not yet know what you are promising.
Questions in the template
- What is the single headline number and over what period?
- What are the three largest components and what are their sources?
- Which components are estimated — and in what range?
- Which dependencies are outside the data team's direct control?
- Who owns the measurement of each component?
- On what date will we know if the model was wrong?
If the last question makes you uncomfortable, you are writing the model correctly.
Value modelling is not an exercise
The process of writing a value model is where most of the value is. The final document is almost a by-product. In every case we have worked, the act of forcing a named operational leader to own a number has uncovered a hidden assumption, an unpriced dependency, or a political fault line that was going to stop the programme anyway. It is better to find it in week two than in month twenty-two.