The pilot has gone well. Reports are reaching customers sooner, employees have adapted to the new process and the early results are encouraging. The managing director has a reasonable question: what happens next? More licences, another pilot, or a decision to make the new approach part of everyday operations?

That is a different problem from getting an experiment to work. A pilot can demonstrate that a tool is useful under controlled conditions. It does not, by itself, establish who supports it, how exceptions are handled or whether the benefit will continue when the original enthusiasts are busy with other work.

Article 3 used a hypothetical 20-person equipment servicing company to illustrate a first AI pilot. This final article picks up where that exercise leaves off. The question is no longer how to get started, but how to turn a promising result into a lasting business capability.

A working pilot is not yet a working service

Research from the British Chambers of Commerce in March 2026 found that most participating businesses adopting AI were using generic tools rather than deeply integrated systems. More specialised adoption remained less common. The research covered 668 UK businesses, mainly SMEs, and should not be treated as a description of every company. [1]

The OECD's April 2026 survey likewise found that targeted, secure integration of AI into business operations remained uneven. Its sample covered more than 2,000 SMEs in 12 countries, but was not representative of the SME population as a whole. Taken together, the studies suggest that progressing beyond experimentation is still a challenge. [2]

An everyday business process needs rather more than a successful demonstration. It needs agreed inputs, predictable outputs, someone to resolve exceptions and a sensible contingency when the service or supplier is unavailable. The arrangement also needs to survive changes to the tool, the people using it and the business around it.

Make the improvement somebody’s responsibility

For a smaller business, I would keep ownership fairly straightforward. One manager should be accountable for the process and its results. The people who understand the work should set the quality standard and approve consequential decisions. A technical or external supplier may be responsible for the system, but should not become the default owner of the business outcome.

In the hypothetical servicing company, the operations manager could own the new reporting process, with an experienced engineer retaining responsibility for technical approval. Somebody must also look after access, supplier settings, incidents and updates. These responsibilities can sit with a small number of people; they simply need to be clear.

A good test is what happens when the employee who ran the pilot is on holiday. Can another colleague follow the procedure, check the output, resolve an error and explain the result to a customer? If not, the pilot is still dependent on an individual rather than operating as a dependable part of the business.

The Handover Question

Can the new process be operated, checked and supported when the people who delivered the pilot are unavailable? If not, the business has further work to do before scaling.

Choose the next opportunity on its own merits

Once an initial use case has succeeded, there can be pressure to apply the same technology everywhere. I would resist that. A system suitable for drafting engineer reports may be entirely unsuitable for changing customer prices, allocating technicians or agreeing contractual commitments without further controls.

A September 2026 working paper from The Productivity Institute found that stronger management practices, particularly those concerned with performance measurement, were associated with a greater likelihood of AI adoption among UK firms. The research does not establish that management practices caused better commercial outcomes. Nevertheless, it reinforces the value of making deliberate, measurable decisions rather than accumulating unrelated experiments. [3]

The Social Market Foundation's October 2026 study of 1,015 UK SME leaders found that businesses reporting more advanced AI use tended to report better outcomes than basic users. That is an association based on self-reported survey data, not proof that deeper adoption causes higher revenues. The report was sponsored by Intuit, which is relevant context when interpreting the findings. [4]

The next decision should therefore consider expected value, the condition of the underlying information, integration effort, risk and the capacity to support the change. Several modest improvements within one well-understood process may be more valuable than deploying a new assistant into every department. Equally, an existing software feature or a simpler process change may still be the right answer.

Keep the benefits visible after the enthusiasm fades

Article 4 distinguished time released from money actually saved. That distinction becomes even more important when a pilot enters routine use. The business should track measures linked to the original objective: perhaps report turnaround, accuracy, rework, customer response or the amount of additional work completed without extra expenditure.

There is no need for an elaborate dashboard. A small number of measures, reviewed consistently by the process owner, should show whether the improvement is continuing. Seasonal demand, changes in staff and differences in the work itself all complicate comparisons. It may not be possible to attribute every movement precisely to AI, but the assumptions should be visible rather than hidden behind a headline about hours saved.

Ongoing costs also matter. Licences, monitoring, changes to supplier terms, security controls, checking time and support can alter the business case after the pilot. A sensible management review should allow the company to improve the arrangement, restrict its scope or stop using it if the original case no longer holds.

Consider what the business could do differently

There is a more interesting possibility beyond making an existing process faster. If reliable AI assistance changes the cost or effort involved in delivering work, it may allow a company to offer a service that was previously uneconomic, respond to customers in a different way or grow without expanding its administrative workload at the same rate.

For the hypothetical servicing business, that might eventually mean more timely maintenance information for customers or a more consistent reporting service across the installed equipment base. Those are possibilities, not proven outcomes, and would require their own commercial case, appropriate data rights and technical assurance. The point is that a successful process change can lead to a different question about what the company is capable of offering.

This is where AI adoption becomes a matter for business leadership rather than simply an IT project. Decisions about pricing, services, skills, information ownership, suppliers and investment need to be considered together. Junior employees must still have opportunities to develop the understanding needed to question AI output, while governance has to keep pace with any increase in automation or access to business systems.

The proposed RJA AI Value & Adoption Framework brings these considerations together through discovery, prioritisation, design, piloting, validation and ongoing improvement. It is intended as a proportionate management approach, not a certified standard or a claim that every business needs a major transformation programme. The first five stages help decide whether a change is worthwhile. The final stage is where responsibility for sustaining it becomes real.

The conclusion to this series is therefore fairly simple. The technology matters, but lasting advantage will depend on the decisions made around it: which work to change, how to make the result dependable, what value is realised and whether the business is prepared to operate differently. I suspect those disciplines will distinguish productive AI adoption from a collection of impressive demonstrations.

Richard James Bowles is Founder & Principal of Richard James Advisory Ltd, providing independent senior technology, transformation and board-level advice to founder-led and growing businesses.

Sources and methodological notes

[1] British Chambers of Commerce, Future of Work: AI in the Workplace (18 March 2026). Source — Survey of 668 UK businesses, mostly SMEs; results are not population-wide estimates.

[2] OECD, Empowering SMEs in the age of AI: The 2026 OECD D4SME Survey (13 April 2026). Source — Non-representative sample of over 2,000 SMEs in 12 countries; targeted and secure integration was uneven.

[3] The Productivity Institute, More than just Plug and Play: Early Evidence on Organisational Capital and AI Adoption (28 September 2026). Source — Working paper using UK firm-panel data; associations do not establish causality.

[4] Social Market Foundation, Byte-sized growth: Supporting SMEs to get more from AI (19 October 2026). Source — Survey of 1,015 UK SME leaders, with self-reported outcomes; Intuit-sponsored study.

Method: The framework and operational recommendations are proposed advisory judgements, not conclusions established by the cited studies.

Illustrative scenario: The 20-person business is hypothetical. No actual client outcomes, permissions or engagements are implied.