The Last Mile Problem: Why AI Assistants Still Leave You Building Deliverables Manually (5 Months of Data)
AI gets you to a good answer in seconds, then leaves you to build the deliverable by hand. Five months of usage show where the Last Mile costs the most.
Maxwell Mwangi
September 9, 2026
·
3 min read
A few months ago I wrote about a problem I kept watching smart people lose hours to. AI gets you to a good answer in seconds, then leaves you to build the actual deliverable by hand. I called it the Last Mile.
Five months of real usage later, I can tell you where it costs the most. It isn't where I expected.
What the Last Mile is
It's everything between “the AI gave me an answer” and “I sent a finished file.” Pasting text into Word and fixing what the paste broke. Rebuilding a table that arrived as plain text. Updating the slide after the model changes.
None of it is hard, which is exactly why it hides. It never shows up as a line item, so teams absorb it as the cost of doing business.
What five months of usage shows
Most teams assume the Last Mile lives in formatting. The usage data says otherwise: legal review and financial modelling top the list.
What those two have in common is that many files must agree and a mismatch is expensive. A contract and a closing model quoting different earn-out figures. A deck still showing last week's exit multiple. Formatting is a one-time annoyance per document. Keeping a whole set of files consistent after something changes is a recurring cost.
It's also the cost single-file assistants can't see. The error doesn't live in the memo or the model. It lives between them, and catching it means holding both in view at once. A careful reviewer does that when there's time, and it's the first thing dropped when deadlines tighten.
Where the tools stand at Month 5
- ChatGPT is excellent at drafting and reasoning, but the output still lands in a chat window. Assembly is on you.
- Microsoft Copilot is a real advantage if your team lives in Microsoft apps. It's AI added to an existing suite, and it generally works within whichever app is open.
- Google Gemini is strong inside Docs, Sheets, and Slides, with the cross-file work left to you.
- Overten treats the finished file as the output across Word, Excel, PowerPoint, and PDF, and reasons across them together.
The Last Mile in four functions
Finance. The exit multiple changes in the model on Tuesday; the LP deck still shows the old one two weeks later. Nobody was careless. Nothing was watching both files.
Legal. The purchase agreement states one earn-out figure and the closing model another. Cross-checking every document at 11 PM isn't a job a person should do by hand.
Consulting. Model in Excel, deliverable in Word, client deck in PowerPoint, an AI chat open beside all three. Twenty minutes of thinking, forty of assembly.
HR. A policy change has to land in the handbook, the onboarding deck and the PDF employees download. Miss one and people get conflicting answers from official documents.
Test it on your own work
Pick one real deliverable that involves at least two files. Time the work after the AI's first answer. Then change one number upstream and count how many other places you have to fix by hand. If that count is higher than you'd like, you've found your Last Mile.
I built Overten because I was tired of watching good work slowed down by the part nobody planned for. Not an AI assistant. An AI office.
Keep reading
See Overten AI in action
See how the Overten AI Workspace Agent helps your team cut the document busywork, catch conflicts early, and free up leaders for the decisions that actually matter.
Maxwell Mwangi
Co-Founder & CTO
Overten AI is an AI document intelligence workspace that reads, cross-references, and reasons across every file in a project, so teams catch conflicts, answer multi-document questions, and generate documents in minutes.