By Jonas Deprez, COO and Co-Founder at Donna

95% of enterprise AI pilots fail. That number comes from MIT research that made the rounds in Forbes in 2025, and while the study is a year old, nothing I've seen in the field since suggests it has improved much in the B2B enterprise space.

The part most people skip past: those pilots don't fail because the technology isn't ready. It's ready. They fail because of people, process, and mindset. AI is not a tool you plug into your current way of working. It asks for a degree of transformation and most pilots are set up as if it doesn't.

Why "let's start with a pilot" is usually the wrong first move

When an enterprise wants to explore AI, the reflex is almost always the same: pick a tool, pick a team, run a pilot and evaluate after three months. It feels prudent. It's how we rolled out SaaS for twenty years.

But AI doesn't behave like SaaS, and that difference breaks the pilot playbook in two ways.

First, the interaction model is different. With a SaaS application, the interface is fixed, you get trained on it, and it works the same way every time. We've been through these shifts before: browser, cloud, mobile, touch but each of those was still a new interface for you to do the work through. With AI, part of the work is done for you. You talk to an agent, through chat, a phone call, a voice note and it executes (think about Donna). The interface we spent decades building may not even be necessary anymore. That's not a feature update. It's a different way of working, and people need to be supported in that transition.

Second, AI is not deterministic. Ask a SaaS tool the same thing twice and you get the same result. Ask an AI the same thing twice and you will get two slightly different answers. That flexibility is exactly what makes large language models useful, but it also means the classic one-off go-live is dead. You can't configure, launch, and walk away. You have to test, learn, refine your prompts and fine-tune the model. You start with a small group and scope, and widen the circle and complexity step by step.

A pilot designed like a SaaS rollout is a pilot designed to fall in the unsuccessful 95%.

So it is important to start with the foundations the pilot will stand on. Which brings me to what the successful minority actually has in common.

What companies in the successful 5% do differently

We've now taken enough enterprise customers from first conversation to production with measurable value to see the pattern clearly. Think about global customers like: ABB, AtlasCopco, Puratos, Ottobock, Döhler and many more. It's not budget, and it's not technical sophistication. It's this:

1. They have a clear process and a clear expected output

The companies that succeed can already articulate how their organization of the future will look and how they should work differently and with what output. Not the small improvements but the bigger picture: what does commercial leadership expect from their future sales organization, and where do the expect the gains? When that vision exists, even in draft form, we can plug straight into it.

The reverse is also true, and it's worth saying plainly: technology won't fix misalignment. Automating a flawed process only helps you do the wrong thing faster; at best. That’s why at Donna, we start every project with the process conversation, not the technology conversation. Before anything gets configured, we map how the sales organization works today, how it should work tomorrow, and what output leadership expects from it. If that vision only exists in draft form, we help sharpen it first.  Often a customer's CRM implementation partner already knows these processes inside out, Donna works closely with that ecosystem of partners to build on what's there. Think of partners like PwC, Deloitte, Sybit, The CRM Firm, Acorel and many more.

2. They align stakeholders before the kickoff, not after

This one costs time and gets skipped constantly. It is not unusual to hear managers  describe how a team works, and to hear a completely different story from the team members themselves. If you don't build consensus on the current way of working and the intended new one, you're not just risking confusion, you're risking quiet sabotage: people who were never consulted might derail the project later on.

C-level sets the vision, managers translate it, and the people doing the work every day need to recognize themselves in it to see and feel the value. All three, aligned, before anything goes live. Our kickoff deliberately covers all three layers, because at Donna we believe it’s important to have check-ins across management stakeholders and the reps who live it every day. We put any differences between those stories on the table before go-live, not after the field discovers them.

3. They secure real AI skills: in-house or through a partner

Implementing AI well means knowing how to evaluate it. This includes monitoring, observability, testing, iterating and fine-tuning. These skills are often underestimated and only limited available with resources internally, so often the project team works with a supporting partner. MIT confirms this and found that external partnerships that bring domain fluency reach deployment 2× more often than internal efforts (67% vs 33%).

For Donna this is the key role we take on. We bring our expertise in rolling out AI projects at enterprises, and we own the observability, testing and fine-tuning that keep an AI assistant performing in production. Our customers don't need to build that muscle from scratch before seeing value; it comes with the partnership.

4. They fix the incentives

If you expect people to work in a new way but keep rewarding them for working the old way, the old way wins. Every time. This is no different from the digital transformations of the past fifteen years, incentives for change are part of the project, not an afterthought.

For example one of our clients added an internal evaluator who scores the quality of visit reports, fully automated, and made that score part of the sales targets. Reps who work with Donna hit 100% on that KPI by default: every visit is captured, complete and in the right format. And because those visit reports feed downstream into pipeline forecasting, the incentive doesn't just reward the sales for the new way of working; it compounds into sharper forecasts for the whole business.

5. They resource the iteration, not just the launch

Successful customers assign a project owner and a team whose job includes collecting feedback from the first user group to enable finetuning based on that feedback. Organizing around that feedback loop is what makes the iteration cadence real instead of aspirational.

For projects with Donna it takes resources on both sides: we bring delivery consultants and customer success managers to the project, but there has to be someone driving dynamics on the customer side taking ownership and driving the change and the finetuning that’s needed. A project owner on the customer side keeps the feedback flowing and drives the change internally. When both sides are resourced, the iteration cadence holds and that’s where we see results.

6. They have an action plan built around early wins

Where do we start? Which team, which country? Who goes first? The pattern that works for Donna is that we begin with early adopters who give honest feedback, feel the value early, and become vocal about it. They become the internal champions who show their colleagues the new way of working. Start with a simple scope with the highest ROI. A complex use case often leads to a false start. Early wins buy you the patience the rest of the rollout needs.

What we've learned implementing Donna

Rolling out AI at scale is never plug-and-play. After taking Donna into production across hundreds of field teams, a few constants have emerged every rollout:

• Process before technology. The first conversation is about how the sales organization should work, not about features.
• Talk to the users, not only the managers. They're the ones who need to feel the value, they will be your advocates and champions.
• Staff the feedback loop on both sides. We bring delivery and customer success, we expect a project owner from the customer.
• Plan for the long term, not a launch moment. The willingness to keep fine-tuning over time is the new "configuration". The customers who embrace that end up in the 5% are set-up for success in the long run.

And the 5% is worth reaching. McKinsey found that companies investing in AI for sales and marketing see a clear revenue uplift. We see the same pattern across our own customer base: sales reps spend up to 75% less time on admin, 10× more insights and commercial intelligence get captured in the CRM, and that ultimately drives up to 20% higher sales conversion.

Rewiring how you work takes time, and the companies doing it right are in it for the long run. They start by removing the friction that gets in people's way, setting their teams up for success on both the process and the people side. That's the whole point, and it's worth doing properly.

Jonas Deprez is co-founder of Donna, the AI assistant built for field sales teams.

‍