If you run a lower mid-market ($10M to $50M) or core mid-market ($50M to $500M) company, you probably think that the lessons around AI implementation that enterprise-level companies face don’t apply to you. But you would be wrong. The gap between the enterprise and the mid-market is shrinking fast, and the same patterns of wasted spend and stalled projects are showing up in companies that are a fraction of the size.
As Juan Sequeda observed after attending the Gartner Data & Analytics 2026 event in London, enterprises face some sobering realities around AI implementation:
- 1 in 5 AI initiatives achieve ROI: You are being sold AI tools, copilots, and agents weekly. Gartner reports that even companies who spend 10x the amount on these tools still failed 80% of the time.
- 90% of architectures need an overhaul: In most cases, a complete architecture overhaul is required to make things work optimally.
- Most companies can’t afford a 24-month foundation project: That is two years of potentially-wasted effort, often with very little to show for it.
For mid-market CEOs, the financial math is even less forgiving than it is for the Fortune 500. You do not get to write off a failed two-year platform bet as a learning experience.
You Can’t Buy Context
AI models run on context. But here is the rub: what makes context valuable is that it comes from your organization. You can’t outsource it. And the context you do outsource can only serve to support the truly valuable context, that which is sourced from your organization.
As Juan puts it, context isn’t missing inside your company. It is fragmented and poorly managed. It lives in your operators’ heads, your CRM notes, your spreadsheets, your customer service tickets, and your finance system. The job is not to go buy it. The job is to surface it, structure it, and make it usable.
(To find out how to build a knowledge graph on your own specific niche, check out Topic Modeler.)
The Way Forward: Pick One KPI-Bearing Use Case and Go
The way to start is to begin with a specific business outcome in mind. Don’t go for cool-sounding or impressive projects. Find something measurable that your organization cares about. Build the minimum data foundation to support it. Ship and repeat.
Juan calls this the iron-thread or pay-as-you-go approach: you start from the outcome, identify the minimum foundation to support that outcome, execute end-to-end, and build the foundation simultaneously with the value delivery. You are not doing foundations first and value later. You are doing them together, use case by use case. For a mid-market company, this is the only practical path.
Most Mid-Market Companies Lack a Chief Data Officer
Of course, it would be excellent to throw this task to your chief data officer, or CDO. But most mid-market companies don’t have a CDO. In fact, as CEO, chances are YOU are your own data leader by default.
This is where JLytics comes in. We act as your external CDO to implement these kinds of projects. We fill that gap, without the cost or hiring lead time of a full-time executive.
For Mid-Market Companies, Governance Is “Who Owns the Single Source of Truth?”
If you don’t have a CDO in your organization, you probably don’t have an official data governance policy in place either. That’s okay. You just need to find out:
- Who owns the data
- What is the source of truth for revenue
- What can the AI tool actually access without leaking Personally-Identifiable Information (PII)
Governance for the mid-market doesn’t have to look like a 40-page policy document. It looks like a handful of clear answers and a documented handshake between the teams that create the data and the teams that use it.
Day in the Life of a Piece of Data Exercise
A simple exercise to try: follow one piece of data through your organization from capture to close, fulfilment, and retention. This pressure test will shine a light on where AI can and cannot be trusted today. You will find the unowned hand-offs, the silent re-keying, and the places where the definition of customer or revenue quietly changes between systems. Those are exactly the seams where AI projects fail.
The Question You Should Be Asking Yourself
Do I understand how work gets done well enough that an AI agent could do part of it reliably? Looked at another way: if your own processes are a black box to you before AI, you should think twice about applying AI to the process.
To find out more, take our data assessment or reach out to get the conversation started today.
