In every organisation using AI, someone is already deciding which tools people use and what for, whether or not anyone asked them to. The choice for leaders is who makes that decision. It affects how quickly teams put AI to work, how often the organisation pays twice for one tool, and whether the leadership team can answer a client or regulator asking what AI is in use. The decision also covers AI that suppliers add to software already in use, sometimes switched on by default.
One person decidesLink to this section
One named person approves every tool and every new use of one. Everyone knows who to ask and gets a consistent answer, and the organisation sees what it pays for in one place. The cost is that person's time: new uses keep arriving from every team that takes up a tool, and each waits its turn.
Easy to miss. It is one person's judgement applied to every team's work, so teams they understand get quick answers and the rest tend to wait longer or get a cautious no.
When to choose it. It makes sense while AI use is small, with time set aside for the role and a deputy. Written reasons keep answers consistent through absences, and become the first draft of policies later.
A small group decidesLink to this section
Requests go to a group that meets on a schedule. Several people assess each tool on cost and risk, duplicate purchases get caught, and a record of each decision keeps answers consistent as members change. A group costs its members' time, spent on small decisions as well as large, and teams wait for the next meeting.
Easy to miss. On risk a group is usually more balanced than one person, but these groups are often drawn from IT, legal and finance, who see what could go wrong more readily than what a tool would do for the work.
When to choose it. Choose it once there is more AI use than one person can handle. Include someone from the teams using AI, and answer every request within a set time. Requests that keep returning are ready to become policies.
Teams decide, within policiesLink to this section
A central group sets policies, such as approved tools and limits by type of information, and teams choose within them, asking when a case falls outside. Team leads make more of the decisions, and several teams can start at once. Keeping the policies current takes someone's time.
Easy to miss. When policies are written for a legal reader, team leads refer every unclear case to the central group, and the organisation is back to a small group deciding, with more paperwork.
When to choose it. It is best once central decisions start to repeat. Write for the team leads applying the policies, try them on real requests first, and add the answer to each unclear case to the policies.
Teams decide, with adviceLink to this section
Teams decide for themselves, with central advice they may depart from. People see their judgement trusted, and uses fit each team's work. It gives up consistency and money: the same question gets different answers in different teams, and similar tools often get bought more than once.
Easy to miss. Busy teams ask for advice when they are stuck, usually after choosing a tool and starting to pay for it.
When to choose it. It suits organisations where experienced team leads run work too different for one set of policies. Make advice quick to get, with a set time to reply, and ask teams to record what they choose, so similar purchases can be combined before they renew.
Each team decidesLink to this section
Each team chooses its own tools and uses, within the law and anything the leadership team puts off limits. Decisions are quick, and made by the people using the tool. The costs are duplicated licences, a total bill nobody has added up, different policies for anyone working across teams, and problems that can stay in the team that found them.
Easy to miss. The responsibility stays with the leadership team: a client, insurer or regulator still asks the organisation, and expects one answer. That includes the terms each team accepted: free and personal plans often let the supplier use what people type to improve its models, where plans bought for an organisation usually rule that out.
When to choose it. It is worth choosing where teams do separate work. It relies on a short off-limits list, and a named person in each team who can say what it uses and why, and passes problems on.
How to chooseLink to this section
Answer these together, with the people who run the teams.
- How many teams will use AI a year from now? One person or one group can answer only so many requests.
- How alike is the work across teams? Similar work suits central decisions and shared tools.
- What does a wrong decision cost? If the worst case is a duplicated licence, teams can decide. If it is client or personal information in the wrong tool, keep that central or off limits.
- Whose time is set aside? Without it, the arrangement exists only on paper.
- If a client asked tomorrow what AI the organisation uses, who would answer?
Tensions with other decisionsLink to this section
- One person deciding at a fast pace. At a Quick or faster pace there are soon more requests than one person has time for, so teams wait or go ahead without asking. A small group can have the same problem as requests grow, so give it a set time to reply. Let teams decide within central policies, or slow the pace.
- A careful risk appetite with teams deciding. Safe first and Cautious rely on a list of approved tools. When teams decide with advice, or each team decides, their lists become more different over time, and the leadership team knows every tool in use only if every team keeps its list current. Move to Teams decide, within policies, or bring the appetite up to Balanced.
When to revisit itLink to this section
Set a review date now, and revisit sooner when:
- requests wait longer than promised
- a large renewal is due, or a tool is being paid for twice
- more teams take up AI, or the pace changes
- a supplier adds AI to software already in use
- a client, insurer or regulator asks what AI is in use, and the answer is slow to assemble
If the leadership team would like support with this decision, see Getting outside help.
Something missing or out of date here? Tell us.