You're probably spending your AI budget in the wrong place

Let me tell you what I see most often when I start working with a founder on their AI strategy.

They've invested in a new AI content tool. Maybe a smarter outbound email sequence. There's a chatbot being trialled on the website. The sales deck mentions AI three times. The board is happy because the metrics are visible and the narrative sounds right.

And yet, every week, that same founder is still chasing someone for meeting notes. Waiting until month end for the finance team to flag missing expenses. Asking the ops lead to pull together a report that takes half a day because it lives across four different spreadsheets.

AI is running. The friction is still there.

I've been thinking about this a lot, both in how I've built Project Bonsai and in the work I do helping clients get real value from projects such as AI adoption. A recent piece of research from MIT has put some hard numbers around what I've been seeing on the ground.

What the Data Actually Shows

Source: State_of_AI_in_Business_2025_Report

MIT NANDA's State of AI in Business 2025 analysed over 300 AI implementations across organisations of all sizes. The headline finding is stark: 95% of AI pilots generate zero return on investment. As Fortune reported when the research was published, most enterprises are stuck in a cycle of experimentation without transformation.

But buried inside that report is a finding that I think every founder should read twice.

Despite 70% of AI budgets flowing toward sales and marketing functions, the organisations generating the highest and most measurable ROI are doing it in the back office. Document processing, contract management, expense workflows, and internal reporting. The unglamorous stuff.

The best performers are eliminating outsourced BPO contracts and reducing external agency spend by up to 30%. Not by replacing salespeople or writing better ads. By removing the operational drag that quietly bleeds time and money every single month.

Why Founders Get This Wrong

It's not a failure of intelligence. It's a failure of attribution.

Sales and marketing AI wins are easy to show. Demo volume is up. Email response rates improved. There's a graph you can put in a board update. The link between the tool and the number is direct and clean.

Back-office wins are harder to quantify in a slide. How do you value the hour you didn't spend chasing a missing receipt? How do you measure the decision you made faster because the data was already in front of you, rather than buried in someone's inbox?

The MIT report makes this point directly. Legal, procurement and finance functions offer real efficiencies, but they're subtle and harder to surface in investor conversations. So budgets flow toward what's visible, not what's valuable.

For large enterprises, this is a structural problem. For founders and growing SMEs, it's actually an opportunity, because you can move on this quickly and without the politics.

What It Looks Like in Practice

Here's what I've been building, and helping clients build, over the past year.

Meeting intelligence that removes the silo. Google Gemini captures meeting notes automatically and drops them into a shared folder. Slack and Notion point to the same source. No more chasing someone for the recap after a client call. No more different versions of what was agreed living in different tools. The information exists once, and everyone can find it.

Expense prompts that don't wait for month end. An accounting connector monitors for missing submissions and nudges the individual directly, before the crunch. Not a reminder blast to the whole team. A quiet, specific prompt to the right person at the right time. Finance stays clean, and nobody's scrambling at month close.

Dashboards built for your business, not borrowed from a template. Instead of waiting on a developer or paying for another SaaS tool that almost fits, we use Claude Code to build simple reporting dashboards tailored to the specific metrics that matter. It takes hours, not sprints, and it doesn't require reallocating any dev resource.

None of this replaces strategy. None of it is the thing that wins the deal or builds the brand. But all of it creates the conditions where you can focus on those things without the operational noise pulling you back.

The Real Point

The phrase I keep coming back to is this: let the ops handle themselves.

Not because operations don't matter. They absolutely do. But because a founder's attention is a finite and genuinely scarce resource. Every hour you're rebuilding a report or chasing a process that should have been automatic is an hour you're not spending on the work that actually moves the needle.

The MIT research shows that mid-market companies who've got this right are implementing in 90 days. Large enterprises take nine months to get out of the pilot phase. That gap isn't about budget or talent. It's about the ability to make a decision and act on it.

As a founder, that's your advantage. You don't need a committee. You don't need a business case that takes three months to approve. You can look at where your time actually goes, identify the workflows that are eating it, and start automating them this week.

Where to Start

If you're not sure where to look first, I'd suggest starting with the question: what do I do every week that feels like it should already be automatic?

That friction point, the thing that makes you sigh slightly every time it comes up, is usually where the value is hiding.

That's the work I do with founders at Project Bonsai. Not helping businesses chase the flashiest AI application, but helping them find the specific places where removing operational drag creates genuine space to grow.

If that sounds like a conversation worth having, I'd love to hear what your version of that friction looks like.

James Ryall is the Founder of Project Bonsai, working with founders and SMEs to build practical AI systems that create space for the work that matters.