TL;DR
- Organizations are struggling to see return on investment (ROI) from AI because they prioritize speed of generation over proper workflow integration, with 95% of organizations reporting zero return on their enterprise AI investments.
- Unpredictable costs, driven by "probabilistic execution," make AI projects difficult to budget for, as every AI retry, hallucination, and review loop consumes additional, costly tokens.
- AI has effectively shifted the productivity bottleneck; while document creation is now faster, organizations lack the necessary infrastructure for critical routing, approvals, and audit trails required for compliance.
- To bridge this gap, organizations must implement "deterministic governance," which enforces consistent, rule-based execution for every document, ensuring audit trails are captured and CRM systems are correctly updated.
- Platforms like altaFlow act as the necessary orchestration layer, allowing AI to generate content while the platform manages the governed lifecycle of approvals, routing, and system-of-record synchronization.
The AI experience is evolving rapidly, yet looks similar across users.
Six months into a mandatory AI agent pilot, you get two emails. One is the monthly usage invoice. It’s nearly three times higher than forecast. Every new document request, retry, and review loop added to the bill.
The second email is an internal audit request for a batch of AI-drafted contracts before an upcoming compliance review. That’s the moment where the real cost of using AI becomes clear. Without governance over what AI has produced, you’re left trying to stitch together an approval chain where one never existed to begin with. You can’t prove who approved what, when, or why.
The pilot fails, not because the AI was bad, but because there was no governance over what AI produced.
This scenario is becoming increasingly common as organizations race to adopt AI workflow automation. The conversation has focused almost entirely on generating work faster. Far fewer organizations have asked the more important question: what happens after the document is created?
The promise versus the P&L
We were all sold on a vision of how AI would revolutionize our lives. But that vision only told part of the story. After months of market-wide adoption of AI, the gaps are now quite evident.
A report by MIT's NANDA initiative found that despite $30-$40 billion in enterprise investment into AI, 95% of organizations are getting zero return. The vast majority of organizations are seeing no measurable P&L impact, and the reason isn’t that the AI models are failing. It’s that they’re failing to properly integrate with workflows.
In 2024, Gartner predicted that 30% of generative AI projects would end by 2025, thanks to inadequate risk controls, escalating costs, and unclear business value. High investment requirements and unpredictable costs were cited as a major challenge.
Gartner extended their prediction in 2025, saying that agentic AI projects would also come to an end, largely due to escalating costs.
What we’re seeing in the market is that AI models themselves are not the problem. Rather, cost, risk, and value are repeatedly cited as barriers to AI, all of which can be mitigated with governance controls.
Without governance, AI can get us started, but can’t help us reach the finish line.
The invoice surprise that reveals what AI truly costs
One of the biggest reasons why AI is not delivering on its promised vision is that the model is built on probabilistic execution. Probabilistic execution refers to systems where outcomes and paths are determined by statistical likelihoods instead of fixed results. In other words, retries, human loops, and per-token costs can vary AI outcomes versus deterministic execution which behaves exactly the same way regardless of user or run.
Probabilistic execution results in highly unpredictable unit economics. The per-token model is tricky. It’s a pay-to-play situation where every player does a little something different.
Every retry consumes additional tokens. Every hallucination introduces another review cycle. Every exception requires human intervention. Every model upgrade changes behavior. Someone in legal may need 1,000 tokens for a single run, whereas someone in sales may only need 100. As usage grows, costs scale alongside it.
Gartner believes that at least half of AI projects will overrun their budgeted costs. That means that when the AI invoice hits at the end of the month, it’s likely to be way higher than was forecast. And it’s difficult to make an organization-wide token distribution plan, especially in midmarket and enterprise organizations that have many users.
You likely bought into AI because you thought scalable automation would translate into massive productivity gains. In reality, you purchased a variable cost that behaves differently every time you run it. When you have a surprise bill month after month, it’s hard to budget for, forecast, and justify the need for AI.
The bottleneck moved
For years, document creation is what slowed teams down. Sales teams waited days for proposals. Legal departments spent hours drafting contracts. HR teams manually assembled onboarding packets.
AI dramatically reduced that bottleneck.
McKinsey estimates that AI can automate 60-70% of employee time. The capacity for acceleration is there, and it has certainly already changed the way we work. However, acceleration without governance opens us up to a greater risk of exposure. And that’s where the bottleneck now is.
When it comes to AI-generated contracts, creation cost has collapsed towards zero. The cost and capacity to create is not the constraint. What remains unsolved is the routing, approval, signature, auditing, and system-of-record syncing that are crucial to contract lifecycle management.
We’ve started to solve the document execution puzzle. The missing piece is the document workflow orchestration layer that must exist between AI generation and business execution.
What deterministic governance looks like
It’s really important to make the distinction between what AI can and can’t do. AI can augment steps inside a document workflow. It cannot replace the entire workflow.
Though we can’t control the probabilistic nature of LLMs, we can introduce deterministic governance to control document execution. In other words, we still have a say over what happens after a document or contract is generated.
Deterministic governance as it relates to contract lifecycle management means setting controls so that every single time a certain condition is met, then a specific action executes. This ensures:
- Routing gets enforced
- Audit trails are completed
- CRM to document to CRM data loops are closed every time
- Documents are secure and compliant every time
Let’s take a look at this in action.
A mortgage lender operating under CFPB requirements can use AI to draft loan documents at an incredible pace. What happens when those documents move through the business without a governance layer?
Loan officers make edits. Underwriters request changes. Compliance teams add required disclosures. Managers approve exceptions. Borrowers sign the final documents.
Months later, an auditor requests evidence that every loan followed the institution's required approval process.
The lender can produce the final documents, but it cannot consistently prove who approved each version, whether the required reviewers signed off in the proper sequence, or whether policy exceptions received the necessary authorization. Approval records are scattered across emails, chat messages, and disconnected workflow tools. In some cases, they are missing altogether.
The issue has nothing to do with AI's ability to generate the document. The institution generated the paperwork faster than ever. What it lacked was a governed process that enforced routing rules, captured every approval, maintained a complete audit trail, and connected those records back to the system of record.
For regulated organizations, creating documents is only one step in the process. Demonstrating that every document followed the correct workflow is what stands up to an audit.
altaFlow: The governance layer your document workflow needs
Instead of stitching together separate tools for document generation, approvals, PDF editing, signatures, and CRM updates, organizations can execute governed document workflows inside Salesforce with consistent business logic.
Rather than asking employees to manage every exception manually, governed workflows ensure that every approval, routing decision, signature, and system update occurs according to predefined business rules.
Simply put: AI generates. altaFlow governs.
altaFlow is a no code, governed document infrastructure platform that connects and tracks every step of a document’s journey, from the moment a Salesforce record triggers document generation to the moment a signed document writes its data back to that record. altaFlow closes the Salesforce to document to Salesforce loop every time, ensuring that records always reflect the most up-to-date data.
With altaFlow:
- 7.3 million workflows were executed in Q1 of 2026 alone, at a 99.66% success rate
- Contract cycle times are reduced by 40%
- No IT dependency means 80% faster automation build time
- ROI is doubled
- Multiple, disconnected tools are consolidated into one single platform
- Admin hours are reduced by 70%+
Governance is the layer that makes AI truly powerful, especially when it comes to workflow automation.
If your organization is evaluating AI workflow automation, start by understanding what happens after the document is created. Read What Is Workflow Automation? How to Eliminate Bottlenecks and Save Time for a deeper look at governed execution, and watch for next week's article exploring the Salesforce-to-document-to-Salesforce loop that closes the automation gap.