How to keep your team ahead on AI: a working guide

A team stays ahead on AI by owning a few workflows built on its own documents and by holding a weekly working session that adds the next one. Knowing the most about AI is a weaker position than running real work through it, because a team with working workflows has something to test each new model and feature against.

The rest of this guide lays out the order to do that in, who should own it, and what to skip. It is written for a firm of 10 to 200 people that has paid for AI seats, sees uneven use, and has no learning and development department to hand the problem to.

What does staying ahead on AI mean for a team that sells expertise?

A brokerage team, an accounting practice or an advisory firm sells judgment, and the week is full of work that surrounds that judgment: abstracts, first drafts, data entry, research summaries, client materials. Staying ahead means that work gets done faster and to the same standard, so senior people spend more of the week with clients and on deals.

That gives a practical test. A team is ahead when it can name the workflows it runs through AI, the person who owns each one, the person who reviews the output, and the next one on the list. Familiarity with model names and product announcements is useful background, and it does not show up in the work on its own.

Where most teams are today

Use has moved faster than employer support for it. Gallup reported in June 2025 that the share of U.S. employees who had used AI in their role at least a few times a year rose from 21% to 40% in two years, and that 34% of professional services employees used it a few times a week or more. In the same report, 22% of employees said their organization had communicated a clear plan or strategy, 30% said it had guidelines or formal policies, and 16% of employees who use AI strongly agreed that the tools their organization provides are useful for their work.

The training numbers point the same way. The Conference Board's July 2026 survey of nearly 1,300 workers found that 55.1% use generative AI or AI agents daily or weekly, while 33.3% had used organization-provided AI training in the past six months and 28.3% said their organization provides none. The New York Fed found that 15.9% of employed respondents said their employer offers any AI training, and workers without access would give up an average of 11.4% of salary to get it.

Leaders are planning around this. In Microsoft's 2025 Work Trend Index, 47% of leaders listed upskilling existing employees as a top workforce strategy for the next 12 to 18 months. None of these surveys is specific to small professional firms, so read them as the general picture: people are already using the tools, and most have not been told what the firm wants them to do with them.

Start with a list of your own recurring work

Much of the published guidance on this topic describes a training program, with literacy tiers, course catalogs and newsletters, and it assumes a department to run it. A smaller firm can skip most of that and begin with a list of its own recurring work.

Write down the tasks that repeat every week or every deal and that take senior time without needing senior judgment for most of the steps. On a brokerage team that list usually includes lease abstraction, the first pass of a broker opinion of value, survey workbooks, and call preparation. At an accounting or advisory firm it tends to include proposal drafting from past engagements, document intake, and recurring client reports.

Pick one task from that list. Good first candidates share three traits: the task happens often, the inputs are documents the firm already has, and a senior person can tell quickly whether the output is right.

Train people on their own documents

Our view is that people keep using what they helped build on their own files, so the training and the first workflow should be the same activity. Sit the person who owns the task down with real source documents, build the workflow in front of them, let them correct it, and have them run it themselves before the session ends.

Some foundation has to come first, and it can be short: which tools are approved, what client data may go into them, and what the reviewer is responsible for. We cover what that foundation needs in why AI training comes before AI tools, and the session format in how to run AI training on your team's own work. The Conference Board report reaches a similar conclusion at enterprise scale, recommending that organizations develop "applied capabilities that improve business outcomes, not just AI literacy".

Who should own it: champions, a lead, or an embedded engineer?

Vendor guidance favors champions. Google's advice is to ask for volunteers from every business unit and sustain them with a monthly newsletter, regular check-ins and office hours. OpenAI's description of the role says strong champions begin by asking where the team's work is getting stuck, before they bring up a new feature.

Neither page says how many hours the role takes, and in our view that is where the role succeeds or fails. A champion who is expected to do this on top of a full client load has no hours to build anything, and client work will always come first. The three options compare as follows.

Option	Works when	Fails when
Volunteer champions	Each has protected hours and one workflow to own for one group	The role is an honorary title added to a full workload
One senior lead	A partner or senior manager is given the hours and the authority to change how work is done	The lead can recommend but cannot change anyone's process
Embedded engineer	The firm wants the building done between sessions and its people's time kept for review and correction	Nobody inside the firm owns the workflows after they are built

A firm can do much of this without outside help if someone senior is given the time. The embedded model is how King & Company does it: we work inside the client's own AI workspace, hold one working session a week with the person who owns the task, do most of the building between sessions, and leave the client owning every workflow and document.

Build a filter for what is worth your attention

A team with running workflows has a simple filter for AI news. When a new model, feature or product appears, ask whether it would make one of the existing workflows better, cheaper or safer, or whether it makes possible the next item on the list. If the answer to both is no, it can wait.

One person should hold that filter and report to the group once a month in a few sentences. The fuller method is in AI noise versus what matters.

Put a person on review where it matters

Every workflow needs a named reviewer at the point where its output reaches a client, a deal, or a filing. Name the person, list what they check against the source document, and decide what happens when they find an error. A lease abstract goes to a broker who checks dates, rent steps and options against the lease. A draft proposal goes to the partner who will sign it.

This step is what lets a firm rely on the output, and it is also where staff learn what the tool gets wrong. Which checks apply to regulated work is a question for your own counsel or compliance lead.

Measure whether the work changed

Seat counts and login numbers tell you who opened the tool. The useful questions are about the work: how many of this month's abstracts, proposals or reports went through the workflow, how often the reviewer had to correct them, and whether the owner still runs it without being reminded. We go through the measures in how to measure AI adoption on a team.

A twelve week sequence a firm can run

This is an example schedule, built around one working session a week. Adjust the pace to your calendar, and keep the order.

Weeks	What happens	What exists at the end
1	List recurring work, pick the first task, confirm approved tools and data rules	A ranked list and a one-page foundation everyone has read
2 to 4	Build the first workflow against real files with its owner, name the reviewer, write down how it runs	One workflow in use, with a written procedure
5 to 8	Add a second and third workflow with different owners, hold the first monthly news review	Three workflows, three owners, a working filter
9 to 11	Add one more, count how much real work went through each, fix what reviewers keep correcting	Usage and correction numbers for each workflow
12	Review the list, retire what is unused, choose the next quarter's work	A plan the firm can run without outside help

Writing each workflow down matters as much as building it. The procedure, the instructions and the review checklist should sit in the firm's own workspace, so the workflow survives when its first owner changes roles.

What to do after the first workflows are in use

Keep the weekly session and move it to the next item on the list. Have the owners of the first workflows teach them to colleagues, since Google's guidance is right that "the best influencer is a coworker people know and trust". Revisit each workflow when a major model release arrives, using the same real documents to see whether the output improved.

At that point staying current is a by-product of the work. If you would like a senior engineer in the room for the first twelve weeks, get in touch.

Common questions

How do I keep my team from falling behind on AI?

Pick one recurring piece of work that takes senior time, build an AI workflow for it against the firm's real files with the person who owns that work, and keep a weekly working session that moves to the next task once that one is in use. A team that runs a few workflows it understands has a practical way to judge every new tool, which matters more than general knowledge about AI.

How much time should a team spend on AI each week?

In our engagements the shared time is one working session of about an hour a week, with most of the building done between sessions. A firm running this itself should plan protected time for the person leading it on top of that session, because someone has to prepare the draft the session reviews.

Do we need an AI champion or an AI committee?

A champion helps when the role comes with protected hours and a narrow mandate, such as owning one workflow for one practice group. A committee is useful for policy decisions like which tools are approved and what data may go into them, and it is a poor vehicle for building workflows.

Should we train everyone or start with a few people?

Give everyone the same short foundation on approved tools, data rules and review, then start the workflow building with the few people who own the work you picked first. Their finished workflows become the training material for the rest of the firm.

Can a small firm do this without hiring an AI specialist?

Yes, if someone senior is given the hours and the authority to change how a piece of work gets done. Outside help mainly shortens the building time between sessions and brings experience with the review and documentation steps.