Guide
What AI can change in gym operations
AI can take repetitive work off the gym team’s plate, from pulling reports and flagging attendance changes to handling routine membership requests. Keep people responsible for workflow design, exceptions, and member-facing coaching. AI should remove administrative legwork and improve timing, not become a substitute for judgment or relationships.
Key takeaways
- Automate repetitive reporting, monitoring, routing, and routine membership administration before handing AI consequential decisions. Source ↗
- Fix fragmented data and define each workflow before adding another layer of automation. Source ↗
- Let AI identify members who need attention, then let coaches handle the relationship. Source ↗
- Keep human approval where local context, exceptions, or consequential member decisions can change the right answer. Source ↗
- Staff adoption needs ongoing ownership and follow-up, not a one-time technology rollout. Source ↗

Which gym tasks should you automate first?
Start where the work is repetitive and the desired action is clear. Gyms have substantial opportunities across sales, cost optimization, and team capability, which makes those operational workflows better early targets than vaguely asking AI to run the business, according to Foster. Map the recurring job first: what triggers it, what data it needs, what action follows, and where a person must intervene.
Then automate the predictable pieces instead of automating an entire role. [10:19]
“There's a massive amount you can do on sales, there's a massive amount you can do on cost optimization, and there's a massive amount you can do on optimizing your team's capability.”
Karl Foster, Karl Foster - Why Most AI Projects Fail: Sport Alliance's Head of AI on Change Management vs Technology · 10:19
Reporting is a natural place to look next because operators routinely spend time assembling information before they can make a decision. Foster described the operational headache as pulling reports from finance and classes, collecting payment information, and generally running between systems. That is exactly the kind of administrative friction AI can reduce.
Centralize the information, make routine retrieval easier, and give the operator a cleaner picture of what needs attention rather than another dashboard to babysit. [47:19]
Routine membership administration also fits the pattern. An AI interface could take a request to freeze a membership and carry out that task, reducing the complexity facing reception or member experience staff, according to Foster. The important distinction is that the workflow itself still needs rules.
Operators decide what AI may change, which requests require approval, and which exceptions belong with a person. Automation handles the clicks. Management still owns the policy. [16:06]
“You could just say to Claude, member John Doe's right in front of me, wants to freeze their membership, and the AI would go away and do that.”
Karl Foster, Karl Foster (Sport Alliance Head of AI) on Running Your Gym on Autopilot with AI · 16:06
What foundation has to come before automation?
Automation gets shaky fast when member records, payment information, lead data, and attendance records live in disconnected places. Foster emphasized quality over quantity and described a single source of truth as an important foundation. For an operator, that means fixing definitions and access before chasing sophisticated AI.
Decide which system owns each critical field, remove obvious duplication, and make sure the workflow can retrieve dependable information. A fast automated process built on messy data simply produces messy work faster. [9:02]
“Once you understand that, the first thing you need to get after that is quality over quantity.”
Karl Foster, Karl Foster - Why Most AI Projects Fail: Sport Alliance's Head of AI on Change Management vs Technology · 9:02
Fragmentation also becomes harder to unwind over time. Operators can end up with custom apps, separate analytics platforms, outside data sources, and lead pipelines living elsewhere, according to Foster. That architecture matters because an automated action needs context.
Before connecting another AI product, trace where information originates, where it changes, and where the final record lives. The goal is not maximal integration for its own sake. The goal is enough reliable context for the system to perform a narrowly defined task correctly. [12:38]
Where should human oversight stay in the workflow?
Keep people at decision points where context can overturn the obvious answer. A poorly attended class may still matter because a particular group of members comes only to that class, according to Foster. An optimization system might see weak utilization and recommend removing it, while an experienced operator sees a retention risk.
Use AI to surface the pattern and quantify the tradeoff. Give the final scheduling decision to somebody who understands the local membership, programming strategy, and consequences. [33:46]
“It's like, well, sometimes you have classes on that don't necessarily make sense on the face value, but you've got a specific group of members that only come to that class and no other class.”
Karl Foster, Karl Foster (Sport Alliance Head of AI) on Running Your Gym on Autopilot with AI · 33:46
Oversight matters for a more fundamental reason too. Ford said people routinely attribute capabilities and human characteristics to AI that it does not actually possess. That is a useful operating principle, not an abstract technical debate.
A conversational interface can make a recommendation sound more considered than it is. Operators should define permissions, escalation paths, and review points based on the consequences of an error, rather than assuming a fluent system has managerial judgment. [4:36]
“People attribute more to it than it's capable of and attribute human characteristics to it.”
Ken Ford, Ken Ford: 40 Years in AI and Why Chatbots Aren’t Thinking · 4:36
Ford further said errors in current popular AI systems are a byproduct of how those systems work. That should shape which tasks a gym delegates. Retrieving information, flagging an attendance change, or preparing a draft can tolerate review.
An irreversible action affecting a member may deserve approval first. The practical model is controlled delegation: give software specific tools and boundaries, then keep humans accountable for policies and consequential exceptions. [10:49]
What operators are doing
Karl Foster outlined systems that use member behavior to decide when outreach may be useful. His example follows a member whose visit frequency starts drifting, allowing the system to recognize the change and activate communication. That is a strong model for back-office automation because the machine handles continuous monitoring that staff cannot realistically perform member by member.
Operators can define the signal and response while coaches spend their attention on conversations that actually need a person. [19:44]
“So we understand when and then we can activate.”
Karl Foster, Karl Foster - Why Most AI Projects Fail: Sport Alliance's Head of AI on Change Management vs Technology · 19:44
Foster also explained another approach: let AI identify members who need care, then hand that list to personal trainers rather than automating the interaction itself. This cleanly separates intelligence from coaching. Software does the scanning, sorting, and personalization work behind the scenes.
A trainer gets a useful prompt and decides how to connect. That design preserves something gyms are unusually well positioned to provide: a real person who knows the member and can respond to context the database does not capture. [32:13]
Anaelle Oiknine offered a parallel lesson from health technology, where data only becomes useful when it supports decisions between human interactions. Oiknine noted that she wants continuous data to help close information gaps between medical visits. Gym operators face a different setting, but the workflow principle transfers: collect and organize signals in the background, then make the relevant information available when a professional needs it.
More data is not the goal. Better-informed human attention is. [20:57]
What should stay with coaches and staff?
Member-facing coaching should remain human when the job involves motivation, interpretation, trust, or an emotionally loaded conversation. Foster noted that technology should not get in the way of personal human interaction. AI can prepare the coach by identifying declining attendance, organizing member history, or suggesting that somebody needs attention.
The coach can then ask what is happening, read the room, and choose the right response. That division uses automation for preparation without confusing preparation with coaching. [46:05]
“The only thing I would say, in my personal opinion, I've said we've had we've had this discussion before, as long as the data and technology doesn't get in the way of the personal human interactions in the gender.”
Karl Foster, Karl Foster (Sport Alliance Head of AI) on Running Your Gym on Autopilot with AI · 46:05
That distinction also helps operators think about jobs more clearly. Foster observed that he expects roles to shift toward programming, creative thinking, and connecting the dots while AI takes more of the legwork. The useful question is therefore not which employee AI replaces.
Ask which pieces of that employee's day require judgment and relationships, then remove the repetitive work surrounding them. A coach freed from manual list building is still a coach. The software should make that person better prepared to coach. [27:43]
How do you make automation stick?
A technically sound workflow still fails if staff route around it. Foster outlined cases where people returned to familiar spreadsheets after rollout because there was no follow-up mechanism, even when newer technology and processes were available. Treat implementation as an operating discipline.
Assign ownership, train around the actual workflow, check whether employees use it, and revisit friction after launch. The first version is not the finish line. Automation only saves work when it becomes the normal way the team works. [40:07]
Start small enough that the team can see what changed. Foster advised operators to work with intent, build the culture around the technology, and address the surrounding process before expanding its use. That sequence matters more than having the longest AI feature list.
Pick a contained administrative workflow, define the human owner and escalation rule, establish reliable data, then see whether it actually reduces work. Once that loop functions reliably, move to the next back-office bottleneck. [3:43]
“Do it with intent, start small, build the culture, do all the things around the technology first, and then look at how you can use the technology.”
Karl Foster, Karl Foster - Why Most AI Projects Fail: Sport Alliance's Head of AI on Change Management vs Technology · 3:43
Explore the Tech and AI topic hub for more source-backed conversations.
Frequently asked questions
What should a gym automate first?
Start with repetitive administrative work that has clear inputs and outcomes, such as reporting, member monitoring, and routine workflow actions. Foster observed that sales, cost optimization, and improving team capability all offer substantial opportunity. Pick a narrow process first, define its rules and owner, then automate the predictable steps. Hear the source ↗
Can AI handle gym member retention?
AI can support retention by watching behavioral signals and telling staff when attention may be useful. Foster added that systems can detect changes in visit frequency and activate an appropriate workflow. The stronger division is to automate monitoring and prioritization while letting staff handle sensitive or nuanced retention conversations. Hear the source ↗
Should AI replace gym coaches in member communication?
Not as a default operating model. Foster added that automation can identify members who need care without taking over the interaction. A gym can route that information to trainers and preserve the human touch. That puts AI behind the coach, where it reduces guesswork, instead of asking software to become the relationship. Hear the source ↗
How can gyms reduce AI mistakes?
Give AI narrow tasks, controlled permissions, and clear escalation points rather than treating fluent output as judgment. Ford detailed that current AI errors arise from how the systems are designed. Human review should therefore increase with the consequence of the action, especially when a decision directly affects a member. Hear the source ↗
Quotes are lightly edited for clarity.
