Strategy

Build, buy or adopt AI after the demonstration

Choose an AI route by testing its fit with real work, the cost of operating it and what the company keeps when a supplier leaves.

A convincing AI demonstration leaves an operations leader with a purchasing decision. The assistant has read documents and prepared a useful answer. Now someone must decide whether to buy a product, configure a platform the company already uses, or commission a company-specific system. Choosing the most impressive demonstration can leave staff with an application that cannot reach the right records or respect their approval process.

Compare those routes against the same work and constraints. The question is which can deliver a worthwhile improvement at an acceptable total cost, with responsibilities the company can sustain. Ownership matters where it gives the business a necessary freedom. It should not become a reason to build something an adequate product already does.

Establish how much of the work is company-specific

Buying means taking an existing product whose standard functions do enough. Adopting, in this comparison, means configuring an existing platform with the company's records, permissions and working steps. Building means developing an application around requirements those routes cannot adequately meet. These boundaries can overlap: a bought product still needs configuration, and a company-specific application can use an existing model.

Ask the operating owner to show where a standard process stops being sufficient. Company terminology alone may be handled through configuration. A decision that depends on relationships between records, local exceptions and evidence held across departments may need more tailored work. Test that distinction with representative records, including incomplete and conflicting ones. A missing approval rule does not become acceptable because the assistant writes fluently.

Keep conventional automation and a change to working practices in the comparison. Where the steps are fixed and the records already structured, adding an AI model may introduce checking work without improving the decision. Cornerstone Red recommends an off-the-shelf product when it is sufficient. An honest comparison must allow that recommendation to win.

Separate ownership from the ability to operate

Owning software does not appoint anyone to keep it running. Compare who will maintain connections to company systems, manage access, investigate incorrect answers and check changes before release. Include staff adoption and ongoing review alongside purchase, configuration and hosting costs. Compare an existing platform on these requirements; familiar tools may still need substantial work to support a different process.

For Cornerstone Red's company-specific delivery, the customer hosts the deployment and owns the delivered system and processes. Operation can sit with the customer's team, Cornerstone Red or a supplied or recruited team. Agree those responsibilities before approval. Employees still decide commercial commitments and give engineering, quality and safety approvals; an agent preparing evidence does not acquire that authority.

Ask what changes when the supplier changes

Compare dependence as well as the initial price. A product supplier may change subscription terms or withdraw a function. A platform may change the way applications connect to it. A model update may change answers that staff previously accepted. Ask who can pause use, what notice the company receives and how the responsible team will check the changed system against its operating requirements.

Customers can change the underlying model in Cornerstone Red's company-specific delivery, whether to another commercial model or to one whose trained settings are published for the customer to run under its licence. That freedom still requires work: the replacement must be checked with company records and approval rules before staff rely on it. Model portability means having a viable choice, not a promise that every replacement will behave identically or cost nothing to introduce.

Make the exit part of the approval decision

Ask what the company keeps if the supplier or adviser relationship ends. For a product, establish which records and review history can be exported and what stops working when the subscription ends. For a build, confirm the delivered code, configurations, operating instructions and rights needed for another team to take over. Customer ownership should be accompanied by evidence that the nominated team can operate independently.

Decision checks for each delivery route
Decision factorBuy a productConfigure a platformBuild company-specific
Fit to company-specific workStandard functions must meet requirements.Configure records, permissions and working steps.Address requirements the other routes cannot meet.
Operating and checking costInclude purchase, connections and staff review.Include configuration, support and staff review.Assign operation; include hosting, maintenance and review.
Ownership and model changeCheck supplier terms, function changes and notice.Check connection changes and who checks updates.Cornerstone Red customers own the system; check replacement models.
What you keep at exitConfirm exportable records and review history.Confirm what remains usable after the relationship ends.Confirm code, configurations, instructions and takeover rights.

Before approval, ask the operating owner to record which route meets the requirements, its full running commitment and the evidence supporting the choice. Approve the adequate option, including off-the-shelf when it wins. Defer where an unanswered question about access or operation could reverse the decision. Reject a proposal whose additional control has no worthwhile use, or whose dependence the company cannot accept.

Where this applies in our work: company-specific AI delivery.