The first 90 days in an AI adoption role should leave the company with a clear account of how selected work happens, a usable starting measure, and evidence from a supported trial. The lead and sponsor should then be able to decide what to continue, change or stop, with someone responsible for the next step.
For an internal adoption lead in a company of 100 or more employees, that is a more useful first-quarter expectation than a quota of demonstrations or tool sign-ups. Judge the work the lead produces alongside the access, decisions and staff time the employer supplies. A trial waiting for approval is different from a trial that ran and failed its checks.
Use 90 days as a planning horizon. The scope should fit the role, the work and the company's starting conditions; the calendar does not establish that an intervention works.
Agree what the first quarter can accomplish
Start with the sponsor and the people who own the proposed work. Confirm the adoption leader's authority and responsibilities, then translate them into outputs that someone can inspect at the quarter-end review.
A useful agreement names:
- The work to understand. Choose a department and recurring task, with access to the people who perform and check it.
- The trial boundary. State which inputs and tools may be used, what remains outside the trial and who must approve it.
- The employer's commitments. Name the sponsor, workflow owner, required reviewers and protected staff time.
- The review decision. Agree what evidence would support continuing, revising, expanding or ending the trial.
Do not make the new lead responsible for permissions they cannot grant. If the role includes coaching and coordination but not purchasing or data approval, keep those decisions with the authorized owners. The AI adoption manager role explains the broader remit; this first-quarter agreement makes a small part of it deliverable.
Listen to how people finish the work
The first month should produce an account of real work, including the checks, interruptions and handovers that a demonstration can omit. Ask colleagues to show a recent task from its starting input to its accepted output. Where approved access permits, watch the work rather than relying only on descriptions of an ideal process.
In her January 2026 article on the first 90 days of SME adoption, adoption adviser Kim Mason describes clients asking where to begin and proposes first understanding existing AI use and unclear boundaries. This is a practitioner's proposed approach for smaller businesses, not a measured result. Its useful challenge to a new lead is straightforward: find out what colleagues already do before announcing what they ought to adopt.
Suppose an industrial spare-parts distributor hires Sam Taylor to support AI adoption. A sales coordinator prepares quotes from customer requests and the company's approved catalog. A line in a request names a part and a quantity, but the catalog supplies it in boxes. Sam watches the coordinator pause to establish whether the customer wants individual units or that many boxes.
A polished AI draft would not resolve the customer's meaning. Sam records who can clarify the quantity and how the coordinator distinguishes the requested item from its supply unit. Prices, available stock and delivery commitments still need confirmation in the current authorized systems.
The first useful output is a short task account: inputs, finished output, responsible people, important checks and unresolved questions. Include colleagues who are not enthusiastic about AI. Their account may reveal a constraint that volunteers have already learned to work around.
Establish a baseline before changing the task
A baseline records what happens under the existing process so the trial has something meaningful to compare with. Collect it before introducing AI where practical, and retain the conditions that affect the comparison.
HM Treasury and the Evaluation Task Force's guidance on evaluating AI interventions, updated in May 2026, recommends defining business-as-usual and planning baseline collection early. It is government evaluation guidance, rather than a validated onboarding schedule for private companies. The relevant principle is to establish the starting conditions before claiming a change.
For Sam's quote-preparation task, record:
- The complete preparation effort. Include reading the request, assembling draft line items, checking them and correcting mistakes.
- The accepted output. Note whether part identity, quantity and supply unit are correct before the quote proceeds.
- The complications. Retain ambiguous requests, missing records, unsuccessful attempts and the reasons work was paused.
- The comparison conditions. Record who performed the task and whether the requests differed materially in difficulty.
Decide which observations the coordinator can collect without making the measurement itself an unreasonable burden. A small, consistently described sample can inform the next trial decision, but it cannot automatically establish a company-wide return on investment. Changes in workload or staff experience may also explain a before-and-after difference.
Count checking and rework as part of the task. A faster first draft is not a benefit if finding and correcting its errors consumes the apparent saving. Keep the whole-work measurement guide beside the trial record when choosing measures.
Run one supported trial that can be reviewed
Once the inputs, permissions and support are ready, test a bounded change. Choose work whose output can be checked by the person responsible for it. Agree a way to pause the trial and return to the existing process if the output or support becomes unacceptable.
In the distributor example, Sam and the workflow owner could test preparing internal draft line items from approved inputs. The sales coordinator checks each line against the source request and catalog. Ambiguous quantities remain questions for clarification. The trial does not authorize sending a quote, substituting parts or making price and delivery commitments.
The sequence is simple, but each step needs evidence:
From starting conditions to a trial decision
Describe the current task
Retain accepted outputs, complete effort and the conditions behind them.
Try the approved change
Record draft checks, corrections, failed attempts and colleague feedback.
Compare and decide
Review quality and effort with the workflow owner before changing the commitment.
Give the next step an owner
Continue, revise, expand or stop within the authority and support available.
The employer must make room for this work. In a February 2026 discussion, designer Tommy Geoco proposed protected experimentation time for a hypothetical large-company design team because regular work already occupied its attention. His proposal is not evidence of a successful rollout or an optimal schedule. It does, however, raise a condition the sponsor should answer: what existing work will move so colleagues can participate?
Avoid turning that question into an expectation of unpaid weekend experimentation. Agree time within the team's working commitments, including the coordinator's checking and the lead's support.
The Cabinet Office's human-centred guidance for scaling AI tools, based on government communications work, recommends retaining sign-off processes and gathering feedback from users and non-users. For this trial, preserve the coordinator's approval role and ask both participants and colleagues who declined what made the method useful or impractical.
Review outputs and dependencies at 90 days
At the quarter-end review, separate what the lead learned and delivered from what the employer still needs to provide. Inspect the records rather than accepting “adoption is progressing” as the result.
| First-quarter output | Evidence to inspect | Employer dependency |
|---|---|---|
Account of selected work | Inputs, accepted output, checks and unresolved questions | Access to colleagues and approved task records |
Starting measure | Quality and complete effort under the existing process | Time and permission to collect observations |
Supported trial or a documented reason it could not start | Checks, corrections, feedback and approval status | Authorized inputs, reviewers and participant time |
Next-step decision | Reason to continue, revise, expand or stop, with an owner | Sponsor commitment and ongoing support |
If approved tool access arrived late, record when it became available and what trial work remains. If access is still missing, identify the decision owner, the requested resolution and the effect on the plan. The lead can still document work and prepare checks using materials they are authorized to use; they should not route restricted information through an unapproved tool to meet a deadline.
A dependency record is useful when it is specific and actionable. It should not become a general excuse for unfinished work the lead could have completed. Ask which output was possible under the conditions supplied, which was not, and what evidence supports that distinction.
If the trial ran, inspect its limitations as closely as its promising results. A method that requires Sam to rescue every draft is not yet ready for colleagues to use independently. A method that preserves quality but adds checking effort may need revision or may not merit continuing.
Finish the review with a decision and a responsible person. Wider rollout is one possible outcome, not a requirement for a successful first quarter. The immediate next step is to agree the output and dependency table with the sponsor before selecting the first trial.
Questions about an adoption leader's first 90 days
What should an AI adoption lead produce in the first 30 days?
An AI adoption lead should produce a clear account of selected work, its owners and checks, plus an agreed trial boundary and a list of required access and decisions. The exact scope depends on the role and starting conditions. Ask a colleague to show a recent task from input to accepted output, then use that observation to prepare the starting measure. A broad tool inventory alone will not explain how the task succeeds.
What if tool or data access is delayed?
An adoption lead should record the missing approval, its owner and the work it prevents, while continuing activities permitted under current access. The sponsor should decide whether to resolve the dependency or revise the trial and timeline. Keep a trial that never started distinct from one that produced poor results. The quarter-end output and dependency review provides a practical way to make that distinction without bypassing approval rules.
Which measures belong in a first-quarter adoption trial?
A first-quarter adoption trial needs measures of accepted output quality and the effort required to finish the whole task, including checking and rework. Record failed attempts and the conditions behind the comparison. For quote preparation, a quick draft matters only alongside correct part identity, quantity and supply unit. Agree a manageable collection method with the workflow owner and use the baseline checklist before interpreting any change as a benefit.
Must an adoption lead achieve company-wide rollout within 90 days?
Company-wide rollout should not be an automatic requirement for a new adoption lead's first 90 days. Expansion needs evidence from the selected work, the required approvals and enough support to sustain it. A well-supported decision to revise or stop a trial can also be a useful outcome. Agree the first-quarter commitments with the sponsor and assess what the lead could deliver under those conditions.
