One intact team of 15 can be trained two to three weeks after you sign. A multi-role AI enablement program at a company of 50 to 500 people runs 6 to 12 weeks from signature. The question your CFO cares about, did this change how people work, gets a defensible answer at about 90 days. Three numbers, because three clocks are running, and most published timelines report them as one.
That is why buyers feel lied to. A vendor says “four to six weeks,” delivers on time, and the buyer still cannot answer the board question in month two. Nothing was missed. The delivery clock stopped; the other two were still running.
The three clocks inside an AI enablement timeline
Write these down separately before you put a date in front of anyone. Corporate AI enablement gets sold as one tidy phase diagram, and every argument about a late program starts by collapsing these three into it.
| Clock | What it measures | How long | When you can read it |
|---|---|---|---|
| Delivery | Signature to the last scheduled session | 2–3 weeks for one team, 6–12 weeks for three or more roles | The day the last session ends |
| Behavior | Last session to the point where people reach for the tool unprompted | 8–12 weeks of near-daily use | Week 10 to week 14 |
| Proof | Start to a result you can put in a board pack | About 90 days | Day 85 to day 100 |
Nobody can sell you out of the behavior clock. The best evidence on habit formation comes from Lally and colleagues, writing in the European Journal of Social Psychology in 2010. Volunteers performed a chosen daily behavior for 12 weeks; the median modeled time to reach 95% of an automaticity plateau was 66 days, and individual estimates ran from 18 to 254. Those are simple daily acts, not drafting a client memo with an assistant, so treat 66 days as the floor rather than the target.
Delivery is the only clock a vendor fully controls, which is why it is the only one most of them publish.
The calendar nobody publishes before session one
Every timeline on the first page of search results starts at day one of training. Yours starts the morning someone says we should do something about this, and at a company with no learning team that stretch often runs longer than the sessions.
Scoping, 3 to 5 business days. Which roles need their own drill material, what tools you already pay for, which task each group practices on. Role count drives the calendar and the price more than headcount does, as our breakdown of what an AI enablement program costs sets out.
Proposal and approval, 5 to 15 business days. The widest variance in the project, and none of it is the vendor’s. One signer with a card: three days. Finance plus legal plus a security questionnaire: three weeks. Read our piece on what belongs in the statement of work before this stage rather than during it, because the clauses that slow legal review are predictable.
Document access, 2 to 10 business days. Hands-on training runs on your material: real intake forms, real client emails, real board decks. Someone has to pull them, scrub anything sensitive, and send them. In our experience this step, not the trainer’s calendar, sets session one.
Build from your documents, 5 to 8 business days. Turning that material into drills that hold up in a room takes design hours. A vendor who skips it is selling a generic deck with your logo on the cover.
Scheduling the room, 3 to 20 business days. Fifteen people from one department is easy. Four cohorts across sales, finance, operations, and legal during a quarter close means negotiating with four calendars that already lost.
Take the midpoints and the honest answer to “when could we start” is three to five weeks after the first call, of which the vendor’s own work is about eight working days. Say it early. Heard in week four it reads as incompetence; heard in week one it is something a buyer can plan around.
What you owe the vendor, and the date it sets
A delivery date is a conditional promise. Every program that has slipped on us slipped on a dependency the client owed, so an honest schedule names them with dates attached:
- A named owner with two hours a week. Not a sponsor who approves things. Someone who answers an email in a day and can get 30 people into a room. Without one, add two weeks.
- A document pack. Eight to fifteen real artifacts per role. Redacted is fine; synthetic is not, because people can tell and they check out inside ten minutes.
- Working licenses on the accounts attendees will use. Not a shared demo seat, and checked the week before rather than the morning of. A locked tenant burns the first 40 minutes and that time never comes back.
- One written answer on what staff may put into a tool. A paragraph, not a policy.
That last one is where most voluntary delay lives: more programs stall waiting for a perfect acceptable-use policy than for any other reason. Write the paragraph, run the training, write the policy properly next quarter when you have seen what people do.
Twelve weeks, laid out
The shape of a program for roughly 120 people across three roles, counted from signature.
Weeks 1 and 2. Document pack arrives, drills get built, cohorts get scheduled, licenses get verified. Capture the baseline here too: ask each attendee which recurring task eats the most time and how long it takes now. Two questions, five minutes, and the only before-reading you will ever get.
Weeks 3 to 6. Sessions run, one cohort at a time, half a day each, on that role’s own material. Space them a week apart rather than stacking them, because people need live work in between to try things on. Expect wins in the room by the second hour, and a few quiet people per cohort who need a follow-up.
Weeks 6 to 10. The part most vendors leave out and the part the behavior clock needs: short reinforcement touchpoints, a channel where people post what worked, an office-hours slot where a stuck draft gets unstuck. Cheap to run, and the difference between a program that shows up in output and one that fades by the second Friday. How that stretch gets billed varies widely, as the comparison of per-seat, per-cohort, and retainer pricing lays out.
Weeks 10 to 12. Re-ask the baseline questions, collect what people produced, write the one-page read for whoever signed the invoice.
Four of those weeks are delivery; five are waiting for humans to change. One team with documents in hand and a single approver reaches the same place in about five weeks total.
What moves the date, in both directions
Three levers pull the calendar in. Fewer roles: four cohorts down to two takes 2 to 3 weeks off delivery, so group roles that share a workflow rather than roles that share a floor. An intact team instead of a cross-company cohort: volunteers from six departments make a pleasant session and no behavior change, because nobody returns to a group that practiced the same thing. A pre-signed master agreement: if AI enablement services are something you will buy more than once, put the paper in place while nothing is pending and save 1 to 2 weeks every time after.
Four things push it out:
- Quarter close and year-end. Finance and sales cohorts cannot be scheduled in the last two weeks of a quarter. Plan around it or lose three.
- Approval chains and security review. Each extra signer adds 3 to 5 business days, and a vendor questionnaire at a company with an IT function adds 1 to 3 weeks. Start it the day the proposal lands.
- Role sprawl. Scope that grows from three roles to six mid-project doubles the build and resets the schedule.
- Treating license choice as the project. Six weeks comparing ChatGPT, Claude, and Gemini before training anyone burns the window in which attention was available. Train people on what you already pay for, which is the case we make in buying licenses is not the same as building capability.
Checkpoints that let you stop paying
A timeline without exit points is a reassurance. Write these three into the schedule with a fail action attached to each.
Week 2: the build check. Ask to see one drill built from your own documents. If it comes back as a slide on prompt frameworks with no trace of your material, the setup hours were not spent. Fail action: hold session one until they have been.
Week 6: the artifact check. By the end of delivery every attendee should have kept something real: a draft, a summary, an analysis that went into actual work. If fewer than half the room has one, the sessions were demos. Fail action: do not pay for reinforcement on a cohort with nothing to reinforce.
Week 12: the behavior check. Re-ask the baseline question and see whether time on that task moved. Fail action: no renewal, known at 90 days instead of a year.
A vendor who resists putting the week-2 artifact into the contract is telling you where their margin comes from. For sequencing the whole program, and what a finished engagement should hand back, see our guide to turning a license rollout into measurable team capability.
When AI enablement produces a number finance accepts
Day 85 to day 100, and only if you captured a baseline in week one.
This clock is slow for reasons unrelated to learning. Usage dashboards update in days and tell you almost nothing: seats active, messages sent, a green line going up. Output changes need a full cycle of real work, and most work runs monthly or quarterly. A finance team needs two closes before anyone can say the second was faster.
The comparison that survives scrutiny is narrow. One named task, one before number, one after number, same people. “Proposal first drafts went from 90 minutes to 35 across eight people” is a sentence a CFO can audit. A 40% productivity claim gets discounted to zero, correctly.
Put the cost in the same frame. Microsoft 365 Copilot lists at $30 per user per month on an annual commitment, an add-on that sits on top of a qualifying Microsoft 365 license rather than replacing it. At 120 seats that is about $43,000 a year.
Set that against the training side. ATD’s 2026 State of the Industry report, built on 2025 data from 340 organizations, put average direct learning spend at $846 per employee, down from $1,254 a year earlier. Formal learning hours per employee went the other way, rising from 13.7 to 16.7. A program at market rates, $2,000 to $15,000 depending on role count, is a routine line against either number.
One timing note. If this leads to a custom automation build, which at a market range of $25,000 to $150,000 it often does, do not scope it before week 12. A team eight weeks into these tools knows which task is worth automating; the same team in week one will guess.
Frequently asked questions
How long does AI enablement take from the day we sign?
Two to three weeks for one intact team with documents in hand, 6 to 12 weeks for three or more roles. Only about four of those weeks are sessions; the rest is document prep, scheduling, and the reinforcement that makes it stick.
Can we just do a one-day workshop?
For one team solving one shared problem, a day works well. For a whole company it does not: a single day cannot carry role-specific drills for four jobs, and it leaves no interval in which anyone practices. If a day is all you have, narrow the audience rather than the content.
Is 90 days realistic for a 200-person company?
For a readable result on two or three roles, yes. For every person in the building trained and changed, no. Teams that get this right pick the roles where AI fluency moves real work, finish inside a quarter, and use the result to fund the next wave.
How long would it take to build in-house?
Budget 20 to 40 hours of design work per role before session one, plus whatever the calendar of the person doing it allows. The knowledge is not the constraint; the hours are. When this has landed on a chief of staff as a second job, the real comparison is external versus nothing.
What delays these programs most?
Document access, then approval chains, then waiting for an acceptable-use policy nobody has been assigned to write. None of the three is the vendor, and all three are fixable in week one.
How long before I can show my CFO something?
About 90 days for a defensible before-and-after on a named task. Artifacts and stories arrive by week six and are worth sharing, but a story is not a number. The one hard requirement is that week-one baseline, which cannot be reconstructed later.
How long does the reinforcement part need to run?
Four to six weeks past the last session, which puts you at the edge of that 66-day median. It is also the cheapest hour in the program, so cutting it to save money is the worst trade on the table.
Do new hires restart the clock?
Not the program clock, but they create a standing need. A company hiring steadily wants a short onboarding module, recorded or run quarterly, rather than a repeat of the full program. Decide that at week 12, when you know which drills transferred.
Why do some vendors quote six months?
Usually because the quote bundles a platform subscription, a course library, or a retainer, so six months is a contract term rather than a time to capability. Ask what is true at day 90 and what months four through six buy. If the answer is access rather than delivery, that is a license priced as a program.
Key takeaways
- Three clocks run at once: delivery (2 to 12 weeks), behavior (8 to 12 weeks past the last session), proof (about 90 days). Quoting one number for all three is where trust goes.
- The calendar starts at your decision, not at session one. Scoping, approval, document access, build, and scheduling add three to five weeks, most of it on your side of the table.
- Four dependencies set the date: a named owner with two hours a week, a pack of real documents, working licenses, and a written paragraph on what staff may put into a tool.
- Capture the baseline in week one, two questions per attendee on one recurring task. Without it, week 12 produces an opinion instead of a result.
- Put checkpoints at weeks 2, 6, and 12 with a fail action on each. A vendor who will not show you a drill built from your own documents has told you where their hours go.
For a date rather than a range, bring three things to a conversation: how many roles need their own material, who can get those people into a room, and whether your documents are ready to share. A free AI-readiness call turns those into a real calendar in about half an hour.
Arthur Wandzel
