Back to Work: The 5 AI Decisions to Make Before Year-End

An AI decision at this point in the year isn’t a tool choice — it’s an executive call on what gets launched, what gets killed, what gets funded, and who owns it. The arithmetic is blunt: roughly 16 working weeks remain before December 31, and an automation project takes 6 to 12 weeks from scoping to production. Anything not settled in September slides into next fiscal year by default. Here are the five decisions to make now, each with the symptom that shows it hasn’t been made, the question to put on the leadership agenda, and the timeline it genuinely requires.

The Real Countdown: 16 Weeks, Not Four Months

Between the September restart and December 31 you have roughly 16 working weeks once you strip out Thanksgiving week, the December holidays, and the year-end stretch when nobody starts anything new. The genuinely usable window is shorter still: past mid-October, an implementation won’t land before February.

That isn’t a calendar constraint, it’s a decision constraint. September is the only month where three cycles overlap: teams are back at full capacity, next year’s budget is being built, and there’s still enough runway to ship.

Decision 1 — Kill or Scale the Pilots You Already Have

The symptom. A proof of concept launched in the spring, technically working, used by two people, and nobody can say whether it’s in production or not. It costs almost nothing, so nobody kills it; it returns nothing either. That’s the worst possible state: it consumes executive attention without producing a decision.

The question for the leadership meeting. “What measurable gain has this produced since launch, and what exactly would it take to put it in the hands of the whole team?” If nobody can answer inside two minutes, the answer is: shut it down.

Realistic timeline. Two weeks to decide, four to eight weeks to scale whatever deserves it. A pilot closed cleanly isn’t a failure, it’s information you paid for and are now collecting.

Decision 2 — Set the 2027 AI Budget During Budget Season

The symptom. AI shows up in the planning deck as a “TBD” line, or not at all. The consequence lands in Q1: every request becomes a one-off exception that has to be defended individually, which slows everything down.

The question for the leadership meeting. “What order of magnitude are we provisioning for AI in 2027, split between recurring licenses, one automation build, and training?” An imperfect number on the page beats an empty cell every time.

Realistic timeline. One two-hour session, scheduled before the budget cycle closes. To put a credible number down without a quote in hand, work from the 2026 AI project cost guide: it gives current ranges by project type, licenses and maintenance included.

Decision 3 — Clean Up Governance After the August 2 Deadline

The symptom. The EU AI Act’s August 2 milestone has passed, nobody internally can say whether it applied to you, and there’s no policy, no usage register, and no record of training. Meanwhile your teams use AI daily.

The question for the leadership meeting. “Who by name owns AI compliance here, and which written document backs them up?” The absence of a name is the real warning sign — and it applies whether your exposure comes from the EU AI Act, from customer security questionnaires, or from your own insurers.

Realistic timeline. Four to six weeks to produce a one-page policy, a usage register, and a half-day AI literacy session. It isn’t a heavy program; it’s a program that never gets started. The concrete obligations are covered in our review of the AI Act after the August 2026 deadline.

Decision 4 — Name an AI Owner With Time Actually Allocated

The symptom. AI is carried by an enthusiastic volunteer who works on it evenings and weekends, on top of their actual job. It moves while they have slack; the moment their day job spikes, everything stops.

The question for the leadership meeting. “What share of this person’s time is formally freed up, and what are we taking off their plate in exchange?” An owner without a workload trade-off isn’t an owner, they’re a volunteer.

Realistic timeline. Immediate on the appointment, two to four weeks to genuinely rebalance their workload. Budget at least 10% of their time — half a day a week. Below that, the role is decorative.

Don’t confuse this with change management on the team side: naming an owner is an org decision, bringing employees along is a different exercise, covered in our 90-day adoption plan.

Decision 5 — Pick ONE Process to Automate by December

The symptom. A list of five to eight automation ideas, all interesting, none committed. Each one waits for the others to be settled. It’s the most common blocker and the most expensive, because it doesn’t look like a blocker — it looks like prudence.

The question for the leadership meeting. “If we could only do one, and it had to be live on December 15, which one would we pick?” The constraint forces the ranking that open discussion never produces.

Realistic timeline. One week to choose, 6 to 12 weeks to deliver. To make the call on evidence rather than instinct, run the candidates through the 20-point readiness checklist: the highest score wins, even if it isn’t the most exciting idea on the list.

At PIWA, we see the same pattern every year: the companies that ship something in December are rarely the best resourced — they’re the ones that agreed in September to drop four ideas out of five.

The Decision Table

Decision	Who decides	Timeline	Cost of not deciding
Kill or scale pilots	CEO + business sponsor	2 weeks	Executive attention burned with no return
2027 AI budget	CEO + finance	One 2-hour session	Every spend becomes a one-off exception
Governance and compliance	CEO + named owner	4 to 6 weeks	Regulatory exposure, untracked usage
AI owner with allocated time	CEO alone	Immediate + 2 to 4 weeks	Work stops at the first workload spike
The process to automate	Leadership team	1 week	Five stalled ideas, zero results in 2026

The September-to-December Calendar

September — Decide

All five decisions land this month, ideally across two meetings: one for decisions 1, 4 and 5 (operational arbitration), one for decisions 2 and 3 (budget and governance). By September 30 you should be able to name the chosen process and its owner.

October — Scope and Launch

Detailed scoping, technical access, baseline measurement. This is also the month the AI policy and usage register get published. Target: development committed before October 20.

November — Build and Test

Development, testing against real data, adjustment with actual users. The densest month, and the one where the owner truly needs that half-day a week protected.

December — Ship and Measure

Production rollout in the first half of the month, user training, measurement of the real gain. The second half is for consolidating, not starting. You enter 2027 with a number rather than an intention.

Three Numbers to Remember

• 16 working weeks between the September restart and December 31, of which roughly 10 are genuinely usable to deliver a build.
• 10% of an owner’s time, or half a day a week: the threshold below which the role produces nothing.
• $9,000 to $20,000 as an order of magnitude for a complete first automation build in year one — the number to put in the 2027 budget in September, and refine later.

FAQ

Which AI decisions need to be made before year-end 2026?

Five decisions belong in September: kill or scale the pilots already running, provision the 2027 AI budget during budget season, clean up governance after the August 2 AI Act milestone, name an AI owner with time genuinely allocated, and pick a single process to automate by December. Each one is settled at the leadership table, not in a technical meeting — these are executive calls about funding and priority, not tool selections.

Why is September decisive for AI projects?

Because three cycles overlap: teams are back at full capacity, next year’s budget is being built, and there’s still enough runway to ship inside the current fiscal year. An automation build takes 6 to 12 weeks from scoping to production with training included. Past mid-October, a newly launched implementation won’t produce a measurable result before February.

What should I do with an AI pilot that has been running for six months?

Settle it within two weeks, one way or the other. Ask two questions: what gain has it produced since launch, and what exactly would it take to extend it to the whole team. If nobody can answer quickly, shutting it down is the correct decision. A pilot closed cleanly is still useful information; a pilot left drifting consumes executive attention and never produces a result.

How much time does an internal AI owner actually need?

At minimum 10% of their time — half a day a week — with a real reduction in their other responsibilities to match. An appointment without a workload trade-off doesn’t create an owner, it creates a volunteer who will last until their first busy period. The right question at the leadership table isn’t “who wants to take this on” but “what are we removing from their plate.” This is the most frequently skipped step in AI org charts.

Should I wait until January to launch an automation project?

Rarely. Waiting until January pushes production to spring, a six-month gap against a decision made in September. If the target process is stable, documented, and backed by an available owner, an implementation launched in October can be live by mid-December. If the scope isn’t defined, however, use the fall for scoping and start in January on solid ground rather than rushing a vague build.

Next Step: Settle All Five in One Session

None of these calls is technical, but they’re hard to make alone: they require a credible order of magnitude on cost, a clear read on what’s deliverable in 16 weeks, and someone to ask the uncomfortable questions in the room. That’s the format of an AI workshop: a short, decision-oriented session that ends with a chosen process, a named owner, and a budget on the page.

Let’s settle your year-end decisions — a working session to make the five calls and build a realistic calendar through December 31.