Indicators & the management dashboard

Which numbers are tracked, how they are defined, and where they are read.

50documents on this topic
24organizations represented
4issues named
11sourced citations
0sourced statistics

The state of it

One of 7 topics within ROI measurement.

Every other topic here asks whether the value is real. This one asks a narrower and more answerable question: which numbers are on the page, who defined them, and where does anyone read them.

The material divides cleanly. Vendors and integrators publish indicator sets - task completion, cost per completed task, work displaced, safety violations, time to first value - and they are specific, calculable and mostly sound. Consultancies publish evidence about what happens when organizations track indicators at all. Put together they produce a conclusion neither states on its own.

The starting error is measuring adoption instead of work. Seats bought, licences renewed, weekly active users: these are the metrics the software industry has always reported, they are trivially available from the vendor console, and they answer a question nobody is asking. A tool used daily that completes nothing is indistinguishable, on that dashboard, from one that has replaced a process. The frontier labs make this argument themselves, which is worth noting twice - once because it is correct, and once because a model vendor arguing for outcome-based measurement over seat-based measurement is not a disinterested party. It is a position, and it is recorded here as one.

But measuring more is not the fix. The most useful finding in this band is counterintuitive and comes from survey data rather than a framework. Grouping organizations by who leads the digital agenda, the group that tracks the most indicators - over 90% of the 46 measured, technology-led - reports the lowest bottom-line result, at 20% of revenue. The group tracking 76% reports 41%. Breadth of measurement and financial outcome come apart, and the naive reading of "firms that measure broadly do better" inverts at the top end.

What separates them is not how many indicators exist but whether each one has a definition, an owner and a place it gets read. An indicator with no calculation method produces two different numbers from two teams and a meeting spent reconciling them. An indicator with no owner is nobody's to move. An indicator that lives in a team's own tooling and never reaches a management review is telemetry, not a measure.

That is the recommendation this band supports: a short specified set, read on a standing cadence, beats a broad unspecified one - and the specification, not the count, is the work.

The issues, by agreement

How many independent organizations name each issue as a problem. An issue is only as real as the number of separate publishers that identify it, so the count is the ranking. Bars are organizations, not documents. Where the count reads ours, no publisher here states the issue and the analysis is our own.

names it as a problemdisputes itqualifies it
Adoption is measured because the vendor console reports it, not because it answers anything
1
The dashboard exists, and no standing meeting reads it
1
The organizations tracking the most indicators report the weakest financial result
1
An indicator is named without a calculation method, a source, a frequency or an owner
ours

Who takes which position

The chart above counts positions; this shows whose they are. Read down a column for what one organization holds across the whole topic, and across a row for who lines up on one issue. Where a cell carries more than one position, the strongest is shown and the rest are in the tooltip.

Ddisputes it Qqualifies it Nnames it as a problem Pproposes a fix
Indicators & the management dashboard: 4 issues against the 6 organizations cited on them. The number under each name is how many of these issues it is cited on.
Issue Google Cloud · 3 UST · 3 Deloitte · 2 Capgemini · 1 OpenAI · 1 World Economic Forum · 1
Adoption is measured because the vendor console reports it, not because it answers anything · P N · P ·
The dashboard exists, and no standing meeting reads it P P · N · ·
The organizations tracking the most indicators report the weakest financial result P · N · · ·
An indicator is named without a calculation method, a source, a frequency or an owner P P · · · P

A dot means this organization is not cited on that issue. It does not mean they are silent on it: an organization is cited where its document takes a position we could locate, and the absence of a citation is the absence of a finding, not a finding of absence. Who is represented lists everyone working on this topic, including those not cited above.

Where they disagree

No contradictions recorded on this topic yet.

The issues in full

Each issue carries the organizations that name it, the numbers behind it, and the remedies proposed - with the concrete steps under each. Every citation points at a section of a named document, so any count here can be checked.

Issue 011 organization name itnewest evidence Aug 2026

Adoption is measured because the vendor console reports it, not because it answers anything

Seats, licences and weekly active users are counted and reported upward. None of them distinguishes a tool that has absorbed a process from one that is opened daily and completes nothing.

This is the default rather than a decision: usage counts arrive free with the subscription while work-completed figures have to be defined, instrumented and argued over. The alternative that the material converges on is a unit of successfully completed work and what it cost to produce - which reframes model choice too, since cheaper tokens that need three attempts and a human review are not cheaper per finished outcome. Note who is making that argument: a model vendor benefits from a measure that rewards work done over seats held. The reasoning is sound and the interest is real, and both belong on the page.

How to fix it — 1 approach, 3 steps

Define one unit of completed work per use case, and cost it

For each deployed use case, name the thing that counts as one finished piece of work, and track how many are completed and what each costs end to end - including retries and human review.

Done when Each live use case defines one completed unit of work, completion counts and full cost per unit including retries and review are instrumented, and seat and active-user counts have been removed from management reporting.

  1. Write down, for each live use case, what one completed unit of work is.0-30 daysHead of transformation
  2. Instrument completion counts and full cost per completed unit, retries and review included.30-90 daysHead of data
  3. Retire seat and active-user counts from management reporting once the completion measure is running.90-180 daysCFO
The evidence — 3 documents
OrganizationDocumentPosition
DeloitteConsultancy · October 2025AI is capturing the digital dollar: what is left for the rest of the tech estateOur reading Names the measurement gap outright: indicator usage across 46 measures has flattened rather than shifted toward outcomes, and use of process-effectiveness measurement fell from 81% to 75% in a year in which AI attention rose, which it calls a missed opportunity rather than a caveat.Return on key performance indicatorsnames it
OpenAIFrontier lab · August 2026A scorecard for the AI ageOur reading Argues the software industry habit of counting seats, active users and renewals answers the wrong question, and proposes cost per successfully completed unit of work as the replacement. Written by a model vendor, whose commercial interest points the same way as the argument.Measuring work accomplished rather than seats purchasedproposes a fix
USTConsultancy · November 2025CIOs guide to agentic AIOur reading Sets out a specific task-level set - completion rate, work reduction, cost and cycle time per task, safety violations - explicitly in place of technical model scores, which are equally unrelated to whether work got done.Task-level performance indicatorsproposes a fix

Issue 021 organization name itnewest evidence Nov 2025

The dashboard exists, and no standing meeting reads it

Indicators are collected into a dashboard owned by the delivery team. The management cadence that decides funding never opens it, so measurement and decision stay in separate rooms.

A dashboard nobody reads on a schedule is telemetry - useful for the team operating the system, invisible to the people who continue or stop the investment. What closes the gap is unremarkable and rarely done: one place where the measures live, a standing slot in an existing review rather than a new meeting, trend and anomaly visible rather than a point-in-time snapshot, and an explicit route from what the numbers show to a decision that gets taken. The absence of that last step is why dashboards accumulate without changing anything.

How to fix it — 1 approach, 3 steps

Give the indicator set a standing slot in a meeting that already exists

Put the measures into the existing management review as a fixed agenda item, with a named person presenting and a decision expected - not a new forum.

Done when The measures are a fixed item in the existing review that owns funding, presented by a named person as a trend rather than a snapshot, and each review records what decision the numbers led to or that none was needed.

  1. Choose the existing review that owns funding, and add the measures as a standing item.0-30 daysCFO
  2. Name who presents, and require trend rather than a point-in-time snapshot.30-90 daysHead of transformation
  3. Record, against each review, what decision the numbers led to - or that none was needed.ongoingChief of staff
The evidence — 3 documents
OrganizationDocumentPosition
CapgeminiConsultancy · June 2025The blueprint to scaling AI for business transformationOur reading Names neglected ongoing monitoring and evaluation as one of the pitfalls that stops AI scaling, and asks that indicator progress be carried upward to leadership rather than held inside the delivery team.Cadence in practicenames it
Google CloudHyperscalerGoogle’s agentic AI transformation frameworkOur reading Proposes an established review cadence in the governance meetings themselves, and a defined feedback loop from what the indicators show to a change actually being made.Reporting cadence and the route to actionproposes a fix
USTConsultancy · November 2025CIOs guide to agentic AIOur reading Proposes a single place for the indicator set, tied into the business review rather than standing beside it, and tracking trend and anomaly so that drift is visible early enough to act on.Dashboards and business reviewsproposes a fix

Issue 031 organization name itnewest evidence Oct 2025

The organizations tracking the most indicators report the weakest financial result

Breadth of measurement is treated as maturity. In the survey evidence the group tracking the widest set of indicators reports the lowest bottom-line outcome, and a narrower set sits alongside roughly double the result.

Grouped by who leads the agenda, the technology-led group tracks more than 90% of the 46 indicators measured and reports earnings at 20% of revenue. The group tracking 76% reports 41%. This does not show that tracking fewer measures causes a better result - the grouping is by leadership, not by measurement policy, and the likelier reading is that a technology-led programme measures what its systems emit while a finance-led one measures what its accounts recognise. That reading is the useful one either way: it says the constraint is which measures connect to the accounts, not how many exist. It also sits against the same publisher reporting that broad measurement correlates with higher value, which is why both halves are stated here rather than the convenient one.

How to fix it — 1 approach, 2 steps

Cut the reported set to measures the accounts recognise

Prefer a small set where each measure maps to a line the finance function already recognises, over a broad set that only the delivery systems emit.

Done when Every reported measure is marked according to whether it maps to a line the finance function recognises, and the management set has been cut to those that do, with the rest held as delivery telemetry.

  1. List every measure currently reported and mark which map to a recognised financial line.0-30 daysCFO
  2. Cut the management set to those that map, and hold the rest as delivery telemetry.30-90 daysHead of transformation
The evidence — 2 documents
OrganizationDocumentPosition
DeloitteConsultancy · October 2025AI is capturing the digital dollar: what is left for the rest of the tech estateOur reading Reports the technology-led group tracking over 90% of indicators and reporting earnings at 20% of revenue, against a group tracking 76% and reporting 41% - and separately reports that broad measurement correlates with higher enterprise value, which pulls the other way.Archetypes by who leads the digital agendanames it
Google CloudHyperscalerGoogle’s agentic AI transformation frameworkOur reading Proposes selecting a balanced set of indicators across business value, risk and compliance, and operational efficiency, chosen with stakeholders in each domain rather than by one function.Balanced selection across domainsproposes a fix

Issue 04Our analysisnewest evidence Nov 2025

An indicator is named without a calculation method, a source, a frequency or an owner

A measure is agreed in a steering meeting and appears on a slide the following month. How it is calculated, from which system, how often, and who is accountable for moving it were never written down.

The failure surfaces as two teams reporting different values for the same named indicator and a review spent reconciling them rather than acting. The specification is four fields - calculation, data source, collection frequency, owner - and writing them is most of the work of making a measure real. The same discipline is what allows a target to be set against the indicator, which is the point at which it starts changing behaviour rather than describing it.

How to fix it — 1 approach, 3 steps

Give every indicator four fields before it appears in a report

No measure enters management reporting until its calculation, source system, frequency and accountable owner are written down and agreed.

Done when Every measure in management reporting has a written calculation, source system, frequency and accountable owner, anything that could not be specified has been removed or fixed, and each survivor carries a target and a named owner.

  1. Write the calculation, source system, frequency and owner for every measure already being reported.0-30 daysHead of data
  2. Remove or fix any measure that cannot be specified, rather than carrying it unspecified.30-90 daysCFO
  3. Set an explicit target against each surviving measure and name who is accountable for it.30-90 daysHead of transformation
The evidence — 3 documents
OrganizationDocumentPosition
Google CloudHyperscalerGoogle’s agentic AI transformation frameworkOur reading Proposes specifying each measure with its calculation method, data source, collection frequency and owner, selected jointly across business value, risk and operational efficiency rather than by one function.Establishing an indicator frameworkproposes a fix
USTConsultancy · November 2025CIOs guide to agentic AIOur reading Goes further and attaches an explicit business target to each measure, on the argument that an untargeted indicator describes rather than holds anyone to account.Targets attached to each indicatorproposes a fix
World Economic ForumInstitution · October 2025Green CPO AI playbookOur reading Applies the same specification discipline where the measures cross an organizational boundary, which is where an unstated definition does the most damage.Indicator definitions in a procurement contextproposes a fix

Who is represented

This dossier is drawn from 24 organizations working on the subject, 6 of which are cited directly in the issues above.

Consultancy — 10

Capgemini 4 Deloitte 4 UST 1 Boston Consulting Group 8 McKinsey & Company 4 EY 2 Infosys 2 Accenture 1 Arthur D. Little 1 Oliver Wyman 1

Institution — 3

World Economic Forum 2 Cloud Security Alliance 2 IAB 1

Academic — 2

Carnegie Mellon SEI 1 UC Berkeley 1

Hyperscaler — 4

Google Cloud 4 IBM 2 Lenovo 1 Microsoft 1

Frontier lab — 2

OpenAI 2 Anthropic 1

Vendor — 3

IntuitionLabs 2 Predibase 1 TechWolf 1