Contents · 7 min read
01 / 05 · Early gains
Why can the fastest return fool you?
In short
Early gains are real, but speed alone does not show which bet will keep growing. The fastest first return may come from work that soon reaches its limit.
In your next review, your leadership team may see that generic report automation cut hours at once and put it first in line for more funding. But once the report takes little time, that gain stops growing, while using AI to spread what your best people know can improve how the next person prepares and follows up.
Now compare two uses in your own company. Generic report automation may cut hours at once, but it stops adding value once the report takes little time. Using AI to spread what your best people know may start more slowly, yet each use can improve how the next person prepares and follows up.
That slower path can look weak in a short review because its gain is harder to count. The ROI Paradox explains why work that builds a rare skill can seem less sound when early returns lead the choice.
To rank the two, you must first see where the quick gain ends.
02 / 05 · The ceiling
When does an AI bet reach its limit?
In short
A bet plateaus when it only speeds up work your company already does. The gain ends once that work is fast enough, and a rival can buy much the same tool.
The tool may still be useful, so keeping it can be the right call. Faster search or report writing can cut cost and make daily work easier.
But the gain has a clear end. Once the task takes little time, another round of speed adds little value. If your rivals use the same base tool, they can reach much the same point.
The enterprise software briefing shows how search and workflow features can become expected when base models make them easier to match. Your company may need those features, but buying them does not make your position stronger for long.
The same limit applies to how people use AI. As Build Builders, Not Users argues, people reach a ceiling when they only speed up tasks they already know.
The next test is whether repeated use changes what your team can do.
03 / 05 · Repeated use
What must improve each time the bet is used?
In short
A compounding bet leaves your team better prepared for the next use. It keeps what people learn from the work, then changes the next response based on what customers do.
A faster task does not become a compounding bet merely because people repeat it. Something from one use must improve the next one.
Have the work owner record the useful change
Keep the point where human judgment changed the result, so the next person can use it.
Have the team update the next run
Change the steps or guidance while the lesson is still clear.
Have the customer owner record the response
Use that response to improve the next preparation or follow-up.
This takes practice because the team must learn to shape the work, not just ask for a draft. AI for leadership teams makes the same split between using AI on a task and rebuilding work around what AI makes possible.
The people must also keep making sound calls. The Capability Paradox explains why AI should make your people more able, rather than leave them busy while their skill stands still.
But repeated learning becomes an edge only if it stays specific to your company.
04 / 05 · The edge
When does learning become hard to copy?
In short
Learning becomes an edge when it depends on company context a rival cannot soon reproduce. It must also improve a business result as the record of work grows.
| Plateaus | Compounds |
|---|---|
| Each use repeats the same fast task. | Each use changes the next result. |
| The same tool is sold to every rival. | The work uses context only your company has. |
| Customer response is stored but not used. | Customer response changes the next action. |
| A rival can copy the gain soon. | Reproducing the gain takes sustained work. |
| The team tracks activity or time saved. | The team tracks a business result that should deepen. |
Use five checks before you call any bet a lasting edge.
What gets better after each use?
Ask the owner to show one change that came from prior work.
What stays with your company?
Check whether the gain depends on your records and your people's calls. If an agent needs those records, decide how it can use your systems before the work grows.
Does customer response change the next action?
A stored note has little value until it changes how your team prepares or follows up.
How fast could a rival copy the gain?
If buying the same tool is enough, the bet may help you keep pace but not pull ahead.
Which result should improve?
Name one result before you fund more work. The Industrial B2B CEO briefing shows the contrast between quick task gains and gains that appear over time in win rates and retention.
Once a bet clears these checks, the remaining choice is how much weight to give it.
05 / 05 · Funding choice
Which AI bets should get the most money?
In short
Fund the bets whose repeated use improves a named result and makes the gain harder to copy. Keep useful tools that plateau, but rank them below work that builds a stronger company skill.
Early return data still matters because it shows whether the work has a sound start. It should inform the choice, but it should not make the choice on its own.
Use this order with your leadership team:
Have each owner name the result
State what should improve if the bet works, such as win rate or retention.
Have finance record the first gain
Track the time or cost saved, so the useful short-term return remains visible.
Have the owner show what deepened
Point to the change in the work that came from prior use or customer response.
Have the leadership team test how easy it is to copy
Lower the rank if a rival can buy the same gain without your history.
Have the CEO set the funding order
Put the most weight behind bets that improve the result and grow harder to copy.
That funding choice now belongs in your AI strategy.
Set the funding order
Your leadership team now needs to rank its AI bets by whether repeated use improves a named result and makes the gain harder to copy. GRAIL's See The Opportunity gives the team three workshops to use AI in its own work and identify where it matters most. It turns that shared view into a ranked roadmap, so the funding order reflects the opportunities that can strengthen with use rather than the tools that only save time once.
Three to four weeks, three workshops with your management team: hands-on use of AI, custom AI agents for each leader, and a ranked roadmap of opportunities. Fixed fee, and no IT project to get started.
Questions CEOs ask about this
How can we tell which AI investments will compound over time and which will plateau?
An AI bet compounds when each use improves the next result and builds know-how rivals cannot quickly copy; it plateaus when it only makes the same work faster. In your next budget meeting, the tool with the fastest early return may look strongest even though its gain ends once the task is fast enough. You must decide whether to fund the quick gain or the bet that keeps getting stronger.
What makes an AI capability improve with repeated use?
Each use must leave a useful change in the next run. The effect grows when customer response also changes how your team prepares or follows up.
Do data advantages still compound when foundation models are widely available?
Raw data alone is not enough. It becomes harder to copy when it is tied to your work and improved by the calls your people make.
Which productivity tools quickly become table stakes?
Generic search and report generation can reach parity fast. Workflow automation can do the same when rivals can buy the same base tool.
How should compounding potential affect investment priority?
Keep useful bets with quick returns, but rank them below bets that keep improving a business result. Give more weight to gains that also take longer to copy.
What metrics show whether an advantage is deepening?
Track the business result named at the start, such as win rate or retention. Then check whether later uses improve that result without relying on the same one-time saving.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- Build Builders, Not Users: why task use reaches a limit while building new systems makes later work easier. (accessed October 29, 2026)
- AI for leadership teams: why experience helps leaders rebuild work around what AI makes possible. (accessed October 29, 2026)
- Industrial B2B CEO briefing: the contrast between quick task returns and gains seen in win rates and retention. (accessed October 29, 2026)
- Enterprise SaaS commercial briefing: why common AI features can become expected and easy to match. (accessed October 29, 2026)
- The ROI Paradox: why rare skills can look weak when leaders rank only early returns. (accessed October 29, 2026)
- The Capability Paradox: why AI systems should make people more able over time. (accessed October 29, 2026)
- AI Agent Access to Your Systems: the choices required when an agent uses company records. (accessed October 29, 2026)
Next scheduled review: February 4, 2027