A month in, and your AI lead has nothing to show you. No tool live. No dashboard. Just conversations about data readiness and a few vendor calls. You start wondering whether to be worried.
The answer depends on what they have produced, not on the absence of visible output. The same observable quiet that signals a healthy programme at day 30 signals a stalling one at day 90. The two look identical from the outside unless you know what to look for at each stage.
What does healthy AI progress look like at 30, 60 and 90 days?
The healthy arc runs in three distinct phases. Assessment before deployment in the first month, opportunity identification and prioritisation in the second, and a board-ready roadmap with a credible measurement frame by month three. Founders who expect a live tool at day 30 are reading a healthy programme as a failing one. Tool deployment before assessment is almost always a warning sign, not a win.
A useful shorthand is discover, prioritise, plan. Month one is about understanding what is actually there, which is rarely what anyone assumes. Month two narrows the field from everything AI could theoretically do to the three to five things worth doing first. Month three builds the case in language the business can act on.
The arc matters because it runs counter to what boards and founders often expect. Early pressure for visible output pushes delegates toward the wrong moves: selecting a tool before the problem is clear, running a proof of concept that impresses in a demo but goes nowhere, or chasing a headline ROI number that evaporates under scrutiny.
Why does reading the signals wrong matter for your business?
The most expensive mistake founders make when they delegate AI is pulling the mandate or pushing for output before the foundation is there. Both errors set the programme back by months. Getting this wrong on a delegate who was doing the right things costs the work already done and the credibility needed to restart. Knowing what to expect is what keeps the programme on the right arc.
Misreading early quiet as failure leads founders to intervene at exactly the wrong moment. A delegate who has spent the first month doing a thorough assessment gets replaced or restructured. A new delegate starts the assessment phase again. Months of good work disappear, and the business learns that AI programmes in this firm do not survive their first quarter.
The reverse error is just as costly. AI theatre takes hold when a delegate feels pressure to show something visible. The common pattern is a cycle of flashy demos, headline tool names, and a roadmap that looks busy but has not been through any real prioritisation. The Addepar test is worth borrowing. Would this initiative still matter if it did not use AI? If the answer is no, someone has optimised for looking busy rather than building something real.
What should your delegate hand you at each stage?
By day 30, the healthy deliverable is a single-page current-state assessment. By day 60, three to five prioritised opportunities each with a named business outcome. By day 90, a phased plan with a credible measurement framework and realistic expectations about when financial returns arrive. These are the specific deliverables that distinguish a delegate making the right progress from one who has drifted.
Day 30. The assessment maps existing AI activity across the business, data readiness, technology infrastructure, organisational capability, and risk exposure. The AI activity piece often surprises, because shadow use in many businesses runs significantly ahead of anything the organisation has formally sanctioned. The key signal at day 30 is discipline rather than output. A delegate who has spent the month interviewing, mapping workflows, and turning down vendor pitches is doing it right. One who has signed up for tool trials and is running demos has skipped the step that makes everything else work.
Day 60. The opportunity list should name a concrete business outcome for each item, rank by a combination of impact and feasibility, and flag a rough investment estimate with a time-to-value range. The outcome needs to be specific, not “improve efficiency” but something along the lines of “reduce the time the compliance team spends on quarterly reporting from four days to one.” Scope discipline matters as much as the list itself. A delegate who hands you twenty possibilities has not done the prioritisation work. Three to five, with real outcomes attached, is the signal.
Day 90. The phased plan typically runs across a foundation phase in the first six months, an expansion phase running to month eighteen, and a longer return horizon beyond that. The measurement framework should distinguish between early indicators (adoption rates, time saved, error rates) and realised financial outcomes. The expectation to set clearly is that meaningful returns take 12 to 24 months, sometimes longer. A delegate who says this at day 90 is telling you the truth, and the board needs to hear it.
When should you ask questions, and when should you hold back?
Ask when a checkpoint passes without a deliverable. Hold back when the work is being done but nothing is visible yet. The distinction sounds simple, but founders consistently get it backwards, asking for demo-worthy output at day 30 and staying quiet at day 60 when a stalling delegate knows no intervention is coming. The checkpoint dates give you the right moments, and the specific deliverables give you the right questions.
Three questions are worth having at each review point. At day 30, ask what the current-state assessment says about the biggest readiness gap in the business. “We’re still scoping the assessment” is a materially different answer from “we’ve identified three departments running significant shadow AI with no governance framework.” At day 60, ask which three opportunities made the cut and what the specific outcome is for each. Vague answers are a signal that the prioritisation step has not happened. Named outcomes with measurable baselines are the signal it has.
At day 90, ask what the measurement framework says about when you should start seeing early indicators. A delegate who cannot answer this has a timeline with tools attached, not a programme with goals.
The board expectation question is often the hardest. Many boards arrive at the 90-day review expecting a live product. Setting that expectation before the review, not at it, is one of the most useful things a founder can do for their delegate.
What else shapes how quickly this lands?
Three factors outside the delegate’s control often determine pace more than the programme design itself. Data quality, the buy versus build decision, and how much change management capacity the business actually has. A founder who understands these constraints reads delays correctly, as factors of the environment rather than failures of the person carrying the brief.
Data quality is consistently cited as the biggest practical barrier to AI adoption. Research cited in Schellman’s 2025 AI implementation analysis, drawing on Gartner data, puts poor data at the root of adoption failures for 77% of firms. A delegate who spends the first month uncovering data quality problems is doing exactly the right thing. Understanding the soil condition before planting anything is diligence, not delay.
On the buy versus build question, vendor-led implementations succeed at roughly twice the rate of internal builds for owner-managed businesses at this scale. A delegate who recommends starting with a proven external solution rather than a custom build is reading the risk profile correctly.
Change management is the factor founders most commonly underestimate. AI adoption stalls far more often at the people layer than at the technical layer. Employees who fear job displacement or distrust AI outputs create passive resistance that no configuration change fixes. A delegate who builds structured change management into the programme from day one is solving the harder problem first.
The 30-60-90 frame is not a project milestone tracker. It is the founder’s equivalent of understanding what a sound audit looks like before the auditor arrives. You do not need to understand the technical work to read the signals well. You just need to know what healthy progress looks like at each stage, and when the right question is “what did you produce?” rather than “what did you switch on?”
If you want to think through how this applies to your own programme, Book a conversation.
