Skip to content
InovagioInovagio

Series · Why AI is stalling

Why the knowledge build stalls

Writing down how a team really works is where most AI implementation time goes, and almost none of it is spent writing. Four stalls account for most of the delay, and the method is built around each one.

· Published · Updated · 7 min read

On this page

The usual list of AI implementation challenges covers data quality and skills gaps as well as integration and governance. The ones that cost the most time, though, are calendar problems.

The project to write down how the export review team at Tarrow Aerospace works is in its ninth week. The draft covers the routine supplier packages and few of the hard ones, because the reviewer who handles drawings with mixed classifications has been on leave since Tuesday, and nobody else knows which ones she clears and which she holds. Internal audit wants to review the draft before anyone uses it. Legal has frozen part of the shared drive for a contract dispute. The sponsor has missed three reviews to deal with a system outage. (Tarrow Aerospace · Example company)

Nothing that week involved the AI, and that is the pattern behind most AI implementation challenges. The model was ready months ago. The hard part is getting a business's rules out of its people's heads and onto paper, approved. Done by hand, that can take a year for a single team, and very little of the year goes to writing; most of it goes to waiting.

The waiting has four causes, and they show up so reliably that the method is built around them.

Four AI implementation challenges that stall a knowledge build: audit review, legal hold, the absent expert and the sponsor's bigger fire
The four stalls that cost a knowledge build the most time.

AI implementation challenges are mostly calendar problems

#

A separate piece on AI stakeholder alignment covers the setup half of a project, from the agreements and the security review to system access and the escalation rule. This one is about what happens after setup, when a team's knowledge has to be written down so a Virtual Employee can work from it.

That knowledge build is usually run as workshops. An analyst or an engineer gathers the experts in a room and captures what they say, then writes it up and circulates it for review. Every step waits on the one before it, and every review waits on a calendar. One missing person stops the whole document.

Audit review waits for a finished document

#

Internal audit, quite rightly, wants to see the rules before anything runs on them. In a workshop project, audit reviews a finished document. So audit waits until the document is finished, and the document waits for audit's comments, which arrive as a fresh set of edits to all of it.

In our method, audit review is an approval step inside the rulebook. A rule that needs audit's eye goes to audit when that rule is drafted, and audit signs that rule. A rule can require more than one approver, and nobody approves their own change. Audit sees its share of the work as it arrives, and every other rule keeps moving.

#

Legal places a hold on a set of files for a dispute or an investigation, and the project loses the records it was reading. Or legal wants to review how a rule touching customer data was written before it is used. In a document project, either one freezes everything that shares a folder or a draft with the affected material, which is usually most of it, because workshop notes and drafts tend to live together in one place.

When rules are drafted and approved one at a time, a hold touches only the rules that depend on the held records. Legal review becomes another approval step on the rules that need it, with its own named approver. The rest of the rulebook doesn't wait.

The one expert is out for two weeks

#
Building the rulebook one after another compared with building it one rule at a time, where interviews, approvals and audit review run at once
One rule at a time keeps interviews, approvals and reviews moving together.

Every team has one person who knows how a particular kind of case is really handled, and that person is out for two weeks at exactly the wrong time. In a workshop, the session is rescheduled, and the dates behind it slide.

Our agents interview people one at a time, fitting around their calendars. Rules are approved one at a time too. When the person who knows one corner of the work is away, the rules in that corner wait for them. Nothing else does. And once that person's knowledge is written down and approved under their name, the team no longer depends on their calendar, or on their staying, which is often why a department asked for the work in the first place. How to onboard an AI employee shows where these interviews sit in the whole process. What the approved rules become, together with your vocabulary and records and who may see what, is a context layer: the part of your systems an AI agent reads before it acts.

The sponsor has a bigger fire

#

The sponsor who signed the project has a bigger problem this month: an outage, a regulator's letter, a quarter-end. Reviews slip. Decisions that only the sponsor can make sit unmade, and the team waits politely.

Two things help here. The sponsor approves the scope and the escalation rule up front, so most decisions don't need them week to week. And because progress is counted in rules approved, with the median days from draft to approval, the sponsor can see in one line whether the build is moving, without sitting through a meeting to find out.

Agents do the legwork, so experts only approve

#

Whoever does the drafting, the knowledge has to stay with your team. Where a vendor's engineers build it on site, it often leaves with them, and nobody inside can keep it current; what you should own when they leave lists what to ask for. "Gartner predicts 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering by 2028" (Gartner, September 29, 2026).

Gartner's prediction that 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering by 2028
A Gartner prediction. Source: Gartner, September 29, 2026.

Our agents start from a role that is already trained: software does the interviewing and drafting, and keeps track of every rule. The right people on your side do the approving, and the rulebook that results is yours, in open formats. Synesis is the knowledge layer. It runs the interviews and keeps each rule tied to who said it, and the guide to how a rule is approved shows what each rule contains.

One rule at a time: agents ask, a rule is drafted, the right person approves or corrects it, and the rule goes live for every task
Synesis runs the interviews and keeps each rule tied to who said it.

Counting delays by cause

#

The method doesn't make any of the four go away. It limits each one to the rules it touches.

The stallHow the method is built around it
Audit reviewAn approval step on the rules that need audit, signed as each is drafted
Legal holdOnly the rules that depend on the held records wait
The absent expertInterviews and approvals run one rule at a time, around calendars
The sponsor's bigger fireScope and escalation agreed up front; progress shown as rules approved

We also count the delays that remain, by cause: days lost to audit review, legal, access, competing priorities, absent experts and our own defects. Those categories are on the scoreboard, beside time to first correct task and cost to first correct task, because a buyer should be able to see where the time went.

Tip

Before you sign for any AI implementation, ask the vendor how the rules get out of your people's heads, who approves them, and what happens to the schedule when one of those people is away.

How it works is our answer in full, and how Virtual Employees learn your rules shows the method from interview to work. Shorter answers are in Virtual Employees: questions and answers, and why AI projects fail covers what happens when this step is skipped.

Share

Questions

What people ask.

What are the biggest AI implementation challenges?

The biggest AI implementation challenges are usually not technical. They are getting a team's rules out of people's heads and approved, and keeping that work moving through audit review, legal holds, absent experts and sponsors pulled onto other problems. The model is often ready long before the business knowledge it needs is written down.

Why do AI projects take so long to implement?

AI projects take so long because the knowledge build usually runs as workshops and documents, where every review waits on the one before it. A single absent expert or a pending audit review can stop the whole document. Drafting and approving rules one at a time limits each delay to the rules it actually touches.

How do you capture expert knowledge before people retire?

You capture expert knowledge by interviewing the people who do the work, writing their rules down one at a time, and having them approve each rule under their name. Inovagio's agents do the interviewing around people's calendars and tie each rule to who said it, so the knowledge outlasts any one person.

What is a forward-deployed engineer in AI?

A forward-deployed engineer is a vendor engineer placed at a customer to build and connect an AI system, often including the business knowledge it relies on. Whoever builds it, the rules and who approved each one should stay with you when the project ends.

How does Inovagio handle audit and legal review of AI rules?

Audit, legal and compliance review are approval steps inside the rulebook, not meetings outside it. Each rule that needs a reviewer goes to that reviewer when it is drafted, and they approve or correct that rule. You decide who approves which kind of rule, a rule can require more than one approver, and the agents that draft rules never approve them. Every other rule keeps moving, and every approval is recorded with the reviewer's name.

See it on your work

See how one of your tasks runs, start to finish.

A 30-minute walkthrough of one of your workflows, the approval step and the audit record, with the people who build the platform.

We use these details to answer your request. See our privacy policy.

What happens next

  1. 1We reply within one business day to find a time.
  2. 2A 30-minute walkthrough of one of your tasks, start to finish.
  3. 3You leave with a short write-up of what we covered.