A firm can discuss AI at every strategy meeting, have partners using general-purpose tools and run a pilot, yet lack one governed application producing an evidenced outcome. Activity feels like maturity from inside the organisation.

An adoption curve can create a shared language for the difference. It should not become a league table or imply that every professional-services firm must pursue the same destination. Readiness varies by use case, and a firm can be advanced in one area while unprepared in another.

The purpose of the model is to identify the next responsible decision.

Five useful stages

Stage 1: unengaged

AI has no material leadership attention, owned investigation or known use. Informal use may still exist, which creates a reason to check rather than assume absence.

The appropriate action is awareness tied to the firm’s work, duties and clients. Avoid beginning with general market statistics or fear of competitors. Identify which tasks may be affected, which risks already exist through unapproved use and who will bring evidence to leadership.

Stage 2: curious

Leaders and practitioners are discussing AI. Individuals may use tools, though there is no coordinated use-case portfolio, accountable owner or agreed evidence.

This is legitimate exploration. It becomes dangerous when the organisation describes informal activity as controlled deployment. Establish an interim policy, understand current use and choose one problem worth testing.

Stage 3: experimenting

The firm has at least one structured experiment with a defined task, participants, data, baseline, success criteria, guardrails and stop conditions. Somebody owns the test and negative evidence will affect the decision.

An unsuccessful pilot can still demonstrate mature experimentation if it prevents a poor investment or exposes a missing foundation. A demonstration with enthusiastic users and no comparison does not meet the definition.

Stage 4: operational

At least one application is used in normal work under an operating model. The firm can show value, quality, cost, adoption, incidents and limitations with suitable evidence. Roles, policies, supplier controls, evaluation and human accountability are active.

The threshold is not a universal percentage of weekly users. Meaningful adoption depends on how often eligible work occurs and who should use the service.

Stage 5: strategic

AI capability materially shapes service, investment, workforce or market choices and is reviewed with the wider strategy. The organisation keeps learning as models, regulation, suppliers and client expectations change.

Client-facing claims are specific and evidenced. Strategic use does not mean placing AI in every service or making it the firm’s public identity.

Assess evidence across dimensions

Do not assign a stage from the most visible activity. Examine several dimensions separately.

Use and value

Which problem is being addressed? What baseline and outcome evidence exist? Has saved time become usable capacity, improved service or realised cost?

Information

Is the required information lawful to use, accessible, current, permissioned and accompanied by provenance? Can errors be corrected and retention controlled?

Technology and supplier

How does the system authenticate, integrate, log and change? What do the contract, data terms, hosting, sub-processors and exit arrangements say?

People and workflow

Can users recognise limitations, review output and handle exceptions? Is verification practical inside workload? Who is affected even if they do not operate the tool?

Governance and risk

Are permitted use, approval, evaluation, human decision boundaries, incident response, monitoring and accountability explicit? Have current legal, regulatory, professional, security and privacy questions been reviewed?

Operating capability

Is there ownership, support, budget, service monitoring and a route to improve or retire the application? A successful pilot without these conditions remains experimental.

A firm may sit at Stage 4 for an internal research task and Stage 1 for automated client advice. Record the profile rather than averaging it into false precision.

Check claims against observable behaviour

Strategy documents and tool licences are weak evidence by themselves. Look for:

  • a named use and accountable owner;
  • actual eligible users and work;
  • approved data and access;
  • pre-agreed evaluation and guardrails;
  • an operating policy people understand;
  • issue, correction and escalation records;
  • current supplier and risk review;
  • decisions changed by evidence;
  • budget and capacity for operation;
  • a clear route to stop.

Interview practitioners and inspect artefacts. Leaders may believe a policy controls use while colleagues rely on public tools outside it. Conversely, cautious informal practice may be more sophisticated than the programme label suggests.

The assessment should protect people who disclose unsafe use so the firm can understand and correct it. It is not a disciplinary hunt.

Move from awareness to structured learning

At Stage 1, give a senior owner time to understand specific implications for the firm. Include practitioners, risk and clients where appropriate. The output is a decision about whether any use deserves further work, not an obligation to pilot within three months.

At Stage 2, inventory current use and set practical boundaries. Select a bounded problem using value, information fitness, reversibility and consequence. Choose internal work where every output can be reviewed when that creates a safer learning environment, while recognising that internal use can still affect confidentiality and professional quality.

At Stage 3, run a genuine pilot. Establish baseline, representative cases, comparison, guardrails and communication before results exist. Test ordinary work and failure. A four-to-six-week duration may suit some tasks and should never be copied without regard to volume and learning needs.

The pilot-design companion article provides a fuller method.

Cross the operational gap deliberately

Moving from pilot to service requires more than scaling access. Decide:

  • eligible tasks and exclusions;
  • owners and users;
  • data and system architecture;
  • supplier and commercial model;
  • evaluation before and after changes;
  • human review and accountability;
  • security, privacy and records controls;
  • training, support and incident response;
  • value, quality and cost measures;
  • retirement and exit.

Re-test at operational volume and with a more representative group. Results achieved with intensive pilot support, self-selected users and curated cases may disappear under normal operating conditions.

Avoid “AI scar tissue” as a rhetorical inevitability. Poorly governed rollout can damage confidence. Transparent handling of a negative result may strengthen it. Explain what happened, contain harm, change the service and preserve the learning.

Decide whether strategic integration is useful

Stage 5 is not the correct ambition for every firm. AI should become a strategic capability only when it materially affects how the firm creates and protects value.

Integrate it with data, technology, people, service and investment decisions. Use a recurring review such as WHNN® to connect present evidence with next options, while maintaining specialist governance at the cadence risk requires.

Do not use client-facing AI claims as maturity proof. State what the firm does, where human judgement remains and which benefit is evidenced. Marketing language must survive professional scrutiny.

Avoid treating the curve as a ladder

Firms can move backwards when a supplier changes, a use case fails or new regulation alters the control environment. That may be responsible governance rather than failure.

Stages can also be skipped conceptually if existing capabilities are strong, but the underlying evidence cannot be skipped. A firm with mature data and technology still needs use-specific evaluation and accountability.

Assess each material use, then form a portfolio view. Stop duplicate experiments, share safe foundations and prevent a successful low-risk application from granting blanket permission for higher-consequence work.

The downloadable AI adoption curve framework can give partners shared language, provided it is used as a diagnostic rather than a score. A structured AI maturity assessment should return a profile, evidence gaps and a specific next decision. It may conclude that the firm should operate, experiment, prepare foundations or avoid the proposed use.

The most useful answer is not the highest stage. It is the position the firm can prove, the risk it can govern and the next step its evidence has earned.