It's Not the Stack: Why Tech and AI Rollouts Fail the Same Way

It's Not the Stack: Why Tech and AI Rollouts Fail the Same Way

Most companies already own more software than they use. Tools get bought, teams never get taught, and adoption either stalls or happens quietly outside official channels. The same pattern is now repeating itself with AI, and the money is bigger.

Beth Shero (CEO & Co-founder, Shero Commerce) sits down with Brannon Skillern (Founder & CEO, Outbloom) and Kate Sargent (Founder & Executive Coach, KJS Coaching & Consulting), who between them advise dozens of organizations on AI enablement and the people side of change, for a look at why the rollout fails more often than the tool does.

In this episode

  • The four ways companies actually make AI decisions, and what each one costs you in speed, security or sanity.
  • How to tell an AI project that looks busy from one that moves the P&L, starting with a question about the workflow underneath it.
  • The five things people need before they will change how they work, and why fear of job loss is the weakest argument in the room.
  • Formal power versus real power, and how to map who signs off, who gatekeeps and who the decision maker calls first.
  • Where AI decisions actually get made, from the budget cycle and vendor demos to the five minutes before a meeting starts.
  • The roles worth naming on any AI effort, from executive sponsor to the guardrail partners you want involved early rather than late.

Watch the conversation for the full playbook.

Episode highlights

Beth opened with a pattern familiar to anyone in eCommerce. A merchant has five tools in the stack, uses a fraction of one, and no team knows what the others are paying for. The stat she put on screen set the tone for AI: the overwhelming majority of pilots are not showing up on the P&L.

Both guests agreed the failure point is almost never the technology. It is the rollout, and the rollout is a change management problem most organizations were already bad at.

Brannon laid out the four models they see across client organizations.

  • Centralized gives you one team that owns everything, along with the red tape.
  • Decentralized lets every team pick its own tools, which creates efficiency gaps and real exposure, since nobody knows which company data is going into which model.
  • Hybrid, with guardrails set centrally and room for each function to work out its own use, is the one both guests back when it is done well.
  • The fourth is no model at all, which never means no AI. Beth named it shadow AI, and both guests confirmed it turns up even in organizations that otherwise run well.

A viewer asked how to tell an AI project that looks impressive from one that is actually making money.
Kate's test was to ask what sits underneath it: what the workflow was before, where it is documented, and what its results looked like. If nobody can answer that, the problem is bigger than AI. Brannon added the other half, which is deciding how success will be measured before anything ships. Twenty hours a week down to ten is a result. Looking cool is not.

On why people resist, the reasons were predictable.
They do not understand why they are doing it, they cannot see what is in it for them, they have no map for where to start, or they tried once, it went badly, and they felt stupid. Kate was direct about the job loss narrative: avoiding AI entirely is the fastest way to make yourself replaceable. Both landed on psychological safety as the missing piece, and on permission as something leaders have to keep giving rather than grant once.

One of the more useful reframes came when Kate argued AI is less a technical skill than a communication one. You do not need to be technical, you need to be able to say clearly what you want. Brannon added the uncomfortable flip side: anyone who has never been able to articulate a problem, a workflow or an outcome will not get good results, because there is nowhere left to hide. As Beth put it, whatever you already are, AI amplifies.

The back half was about influence, and where to find it:

  • Map the power, including who holds the title and who the decision maker calls before deciding.
  • Work the budget cycle, which is where head count, consultant money and the unloved AI projects all come from.
  • Run the vendor demos yourself and bring what you find to the table.
  • Get legal, security and IT involved at the front, not after something breaks.
  • Use the five minutes before a meeting, the break room and the walk to lunch to plant what you are experimenting with.
  • Ask to join the AI council, or just to sit in. There is no cap on seats, and as Kate noted, being on one is no guarantee anybody knows what they are doing.

Asked for one takeaway, Brannon kept it simple.
There is no established best practice yet and no methodology from Harvard Business Review to follow. In five years there will be. Right now the only wrong move is waiting, and nobody else has caught up either. Kate's addition was to stop asking AI only to build or review things and start asking what you are missing, what you should be worried about, and what you have not thought to ask. Beth closed on the human side, which was her own takeaway: there is more room right now to step up and be something more than there has been in a long time.