Making the Numbers Travel
Identifying the best numbers to protect your programme during budget discussions.
01 / What you'll find in this paper
The numbers that protect an AI programme are not the ones that prove activity. They're the ones that travel.
Senior leaders and budget controllers don't have spare capacity to interpret what every metric means for the business. The numbers that land are the ones built to be understood in a sentence. We've identified five principles, grounded in research on how executives absorb information under cognitive load, for choosing the numbers that survive the conversation, and what to do with the rest.
"For one of our clients, AI usage moved from 23% weekly active to 68% in eight months. The stat had been available every week, but it didn't get on the CMO's radar until somebody said it out loud in a meeting."
From the author · Ian Crocombe · Deft
That gap is what this article is about. The measurement problem is usually already solved. The portability problem is not. Most large marketing organisations are sitting on the right numbers and losing the argument anyway, because the numbers aren't built to travel.
When finance-led pressure arrives, and for the remainder of 2026 and into 2027 it will continue to do so, the programmes that survive are the ones with the right two or three numbers that land with the leadership team. The ones that use the weight of the data to win the argument.
Key take-outs
- Activity metrics lose budget meetings. Outcome and capability metrics demand attention.
- The number that reaches the CEO/CFO is the one that can be said aloud without a slide.
- Three numbers travel. Five don't. Brevity is key.
- Focus on the numbers that reflect shifts 'from X to Y' as opposed to simply 'Y'.
- Build the measurement framework before the pressure arrives. Not during.
02 / The context
Activity numbers don't survive budget season.
Most AI programmes are reported with activity metrics: Who has access? How many completed training? How many skills were added last month? These are easy to collect and easy to display. They are also the first things to lose a finance discussion.
The numbers that hold under pressure are different: Output per pound. Time saved per workflow. Speed to market. Capability score across the team. Revenue growth relative to peers. These are the numbers that turn an AI programme from a nice-to-have into a can't-succeed-without.
Anthropic's own data puts actual AI usage in most enterprises at 33% of potential. Early adopters are growing revenue one and a half times faster than their peers. The case for investment is real. But it only lands when you've already built the metrics that make it credible inside your own organisation. Citing someone else's stat on a slide in a budget meeting is not the same as having your own numbers.
"The teams that protect their AI investment when cost pressure arrives aren't the ones with the best argument. They're the ones with the numbers already in place."
Ian Crocombe · Deft
Agreed before the budget cycle opens. Tracking outcomes and capability, not pure activity. The measurement framework has to be built before the pressure arrives. Not assembled during it.
03 / The principles
Five things that make numbers travel.
Each is derived from research on how senior leaders take in information under cognitive load, and on what teams successfully protecting their programmes already are doing in practice.
Principle 01 · The corridor statistic: If it needs a chart, it won't travel.
A number that needs a chart to explain won't travel further than the person holding the chart. The format that survives is one sentence, one shift, one timeframe. "We went from 23% to 68% in eight months." Direction, distance, time. Said in passing in a corridor, and repeated verbatim by the recipient in the next meeting.
Heath & Heath, Made to Stick (2007): the stickiness test is whether the listener can re-tell it without the slide.
Principle 02 · Ratios over absolute numbers: Ratios scale in the listener's head. Raw numbers don't.
"For each hour we spend in AI, we save two and a half hours" works for a 30,000-person marketing organisation and for a team of ten. BCG and McKinsey use ratios in executive briefings for the same reason.
Daniel Kahneman: ratios invoke automatic proportional comparison; raw numbers require effortful calculation that does not happen in a senior leader's day.
Principle 03 · The rule of three: Three numbers travel. Five don't.
Cowan's revision of the working memory literature puts the chunkable limit at three to four pieces of information under real cognitive load, not Miller's classic seven. Executives making dozens of decisions a day are operating at the bottom of that range.
The dashboard with twelve KPIs loses because none of it survives the next conversation.
Principle 04 · Movement before magnitude: Lead with the change, not the level.
"68% weekly usage" tells a senior leader nothing without the target, the baseline and the comparison. "From 23% to 68% in eight months" carries direction, distance and momentum in one sentence.
Prospect theory (Kahneman & Tversky, 1979) is unambiguous: people evaluate outcomes relative to a reference point, not in absolute terms.
Principle 05 · Repetition, not comprehensiveness: Lock the three numbers. Repeat them unchanged.
Ebbinghaus's forgetting curve illustrates how 77% of new information is lost within a week of learning it, without reinforcement through repetition. The number that reaches the CEO/CFO isn't the most thorough one. It's the one repeated, unchanged, across every steering committee, corridor conversation and email summary.
Consistency of the stat is the distribution mechanism, not the comprehensiveness of the report.
04 / The format
One page that travels upward.
When there's one shot to make a number land, in a board pack, a steering committee, a corridor conversation, this is the page we use.
One-page template
Headline
Movement + timeframe: "AI usage has gone from 23% to 68% weekly active in eight months."
Three numbers
56 min/day average daily AI usage (up from 15 min, Sept 2025)
92 min saved perceived daily time returned (1 : 1.5 ratio)
82% team confidence score
So what
"We are on track to hit one hour daily by end of FY26."
Nothing else travels. Everything else is an appendix for those who ask. The structure underneath the page is McKinsey's standard Situation-Complication-Resolution: it does the cognitive work for the audience so they don't have to figure out why the number matters. Headline first, evidence second. The insight at the top, not the bottom.
The teams that protect their AI investment in a restructure or around budgeting time are the ones who agreed these three metrics before the budget cycle opened, and have been repeating the same numbers ever since.
The weight of the data is what wins the argument.
Download a PDF version of this article.
Download the full report
Done. Your download is ready.
Download the report (PDF)05 / About · Working with Deft
We build the operating capability to move at a different speed than the organisations around you.
Deft helps CMOs and marketing leaders build the functional capability, across strategy, adoption, and operations, to move at a different speed than the organisations around them. We work with your tech, across the full AI programme and stay until the change sticks.
If your AI programme is approaching its first real budget conversation, or you can see one coming, that's the ideal time, as a leader, to start a conversation with us. We believe that a human, open conversation is the easiest and best place to start.
"The measurement framework has to be built before the pressure arrives. Not assembled during it."
Ian Crocombe · Deft
Ian Crocombe · CEO and founder, Deft · [email protected] · deft.be