Six months into the year is enough distance to see past the optimism of the January plan and the excuses that accumulate by December. It is the honest middle, and most companies never formally use it that way for their AI spending.
Without a structured check-in, the second half of the year tends to repeat whatever pattern the first half set, for better or worse, simply because nobody stopped to look.
Short answer: a mid year AI review answers four questions. What shipped against what was planned, what the fully loaded spend has actually been against budget, what the measured return looks like against the baseline, and what changes for the second half. Skipping any of the four turns the review into a status update rather than a decision point.
Why this needs to be a distinct exercise
A general budget review touches AI spending as one line among many. That format is not built to catch the specific failure modes AI projects tend toward: stalled pilots, unmeasured returns, and ownership that quietly lapsed. A dedicated review, even a short one, catches what a general review misses.
Question one. What shipped versus what was planned
Compare the roadmap from January or the last quarterly reset against what actually went live. Gaps here are not automatically a problem, but they need an honest reason attached, not a vague sense that things got busy.
Question two. What has the fully loaded spend actually been
Include build hours, review hours, and ownership hours, not just subscription costs. Most companies discover here that actual spend has run higher than the invoice suggested, which is useful information for scoping the second half more realistically.
Question three. What does the measured return look like
Against the baseline captured before each project started. Where no baseline exists, that gap is itself a finding, and it should shape how future projects get scoped.
Question four. What changes for the second half
| Finding | Second half action |
|---|---|
| Project shipped, return proven | Expand or replicate the approach elsewhere |
| Project shipped, return unclear | Fix measurement before adding scope |
| Project stalled | Diagnose the blocker, name a new owner, or cut it |
| Never started | Rescope for Q3 or move to the parked list honestly |
What to bring to the review meeting
Actual numbers, not impressions. The fully loaded spend, the measured return where it exists, and a status against the January roadmap. A review built on impressions produces a conversation about how everyone feels, which is a weaker foundation for a second-half decision than a page of numbers.
What good outcomes from this review look like
Not every project needs to have succeeded. A useful mid year review sometimes concludes that a project should be cut, and that is a better outcome than carrying an unexamined initiative to December purely because nobody wanted to have the conversation earlier.
FAQ
Who should be in this review?
Whoever owns the AI roadmap, finance, and the operational owners of each live project.
How long should it take?
Half a day for a company with two to four active projects, longer only if the underlying numbers are not readily available.
What if the news is mostly bad?
Better to know now, with six months left to correct course, than to discover it during year-end reporting.
Should this connect to the annual budget process?
Yes, directly. The findings here should shape the AI section of next year’s budget request.
What is the most common finding in these reviews?
Missing baselines. Companies can usually say what they spent. Far fewer can say what it actually returned.
Where to go next: Block half a day this month to answer the four questions above honestly. It is the cheapest correction you will make all year.
