AI is moving faster than any technology the profession has ever adopted. New models and new ways to put them to work arrive every other week. Around the world, the adoption of AI in audit firms is moving at very different speeds, and so is the value firms are realizing from it. Some firms have one Copilot license and a few curious staff. Others are all in, employing AI agents to help with routine work like reviewing audit evidence overnight and drafting corresponding comments for clients.
Drawing on hundreds of conversations with our customer firms worldwide, we’ve classified the different ways we are seeing firms realize value from their AI investments. Identifying where your firm is today helps answer an important question many firms continually ask: Are we getting full value from the technology we've invested in?
How much value a firm gets depends on its stage. In stage 1, most of the value in the AI license remains unused. In stage 2, the value is real, but it can more than double the costs and may be constrained by a vendor's roadmap. Stage 3 is where value starts to compound, and stage 4 is where it peaks. Most firms started in the same place…
A manual loop is a process run by a human user rather than by automated programming or an algorithm.
For many, the first stage of AI value realization starts with a license and a question. A staff accountant manually downloads a PBC file, such as the general ledger or a lease agreement, uploads it to an AI chat, asks for a summary or an extraction, then copies and pastes the result, and repeats the cycle of downloading, uploading, copying, and pasting.
It's useful. AI is good at tasks like extracting information from PDFs and summarizing lease agreements. But in that moment, the model only sees the file it was given. It’s missing context. It doesn't know the engagement, last year's requests, or what the client said in yesterday's comment, so the response is only as good as the context someone pasted in the chat.
Wrapper is used to describe a software application built on top of an existing artificial intelligence model such as GPT or Claude.
The second stage has largely been defined by vendors who saw an opportunity to build AI into their platforms, such as chatbots or buttons that trigger predefined prompts. Underneath these tools are usually the same ChatGPT or Claude models that firms already license directly or through Copilot.
For some firms, the convenience felt worth paying for, since part of the manual loop was finally off their plate. But what they were buying was the convenience, not the AI. In practice, it's a fee on top of a model the firm may already license.
The convenience may come with a limit, too. The features are the ones the vendor chose to build, so if your firm's real bottleneck doesn't have a button or get a good response using their chatbot, you may be left waiting on someone else's roadmap. That trade-off is worth a second look when the same convenience can come through the AI license the firm already pays for.
Integration is the process of connecting different software applications or components so they can share data and work together as a single, unified system.
Model Context Protocol (MCP) is an open-source standard for securely connecting AI applications like Claude, ChatGPT and Microsoft Copilot to external systems: the data, tools and workflows people use every day. Think of it as a single standard connection that lets any compatible AI tool plug into any compatible system and work across several at once. For an audit firm, that means you can use an MCP to connect AI directly to your work. The model your firm already licenses extends into the audit platform, including engagements, requests, documents, and client comments/communications with predefined access rights.
Say a manager asks which requests are still outstanding on the Meridian Year-end Engagement, and which clients haven't responded in two weeks. The model instantly reads the engagement, answers the question, and drafts follow-ups for the manager to review.
The firm keeps the context introduced in stage 2 and restores the stage 1 freedom to ask anything, and does it all through the license it already owns, without the convenience fee.
Stage 4 builds on the same MCP connection as stage 3. Once the firm's AI can reach its engagements, the firm can use that access to build workflows, agents and recurring automations that run without everyone re-typing the same prompt. Three building blocks make that possible, and each one builds on the last.
Workflows. A workflow has three parts: a trigger that starts it, a skill the AI follows, and an output that lands somewhere useful. The skill is a set of written instructions that may include details such as which requests to check, what counts as incomplete, and how to word the follow-up comment to the client. Firms write skills the way they'd brief a first-year, and the results stay consistent because the instructions are. The output might be a draft client comment awaiting approval, or a summary in the manager's inbox.
A firm can write a skill for any type of repetitive, manual work that slows its team down, then decide how and when it runs.
Scheduled tasks. A scheduled task is a workflow whose trigger is a preset time, so no one has to remember to ask. At 2 a.m., for example, a task checks every document uploaded the day before against its request and flags anything incomplete, inconsistent or from the wrong period. Follow-up questions and comments are drafted and waiting when the team sits down at 9.
Agents. An agent goes a step further. It's a workflow that can check its own work: it scores its output against a standard, such as the firm's review checklist, before handing it over, and every miss becomes a fix to the skill. When an engagement rolls forward, an agent can review last year's request list, spot the requests that caused the most back-and-forth, rewrite them to be clearer, and flag the changes for the manager's approval.
Each block adds to the one before it. A workflow makes the work consistent; a scheduled task removes the need to ask; an agent handles the checks in between; and the auditor still makes the call.
AI models are improving faster than any firm can evaluate them, and today's leading model may not lead next year. A firm connected through an open standard can adopt the next one without switching platforms, retraining staff, or waiting on a vendor to decide it's worth adding.
Before paying for any AI feature, ask four questions:
Great firms won't pick the winning model in advance. They'll architect an agile tech stack that can connect to it the day it arrives.
AuditDashboard's AI Connectors take firms to stages 3 and 4 on the platform they already use for PBC requests and client collaboration. They connect the AI tools your firm already pays for, including Claude, ChatGPT, and Copilot, directly to your engagements, requests and documents. There's no bolt-on middleware, no extra third party, and no overpriced wrapper.
Every connection follows the role, access rights and permissions each user already has in AuditDashboard, so the AI can only see and do what that user's role allows. Users sign in with their normal credentials, MFA included, and can review and approve each step an agent proposes before it happens. Every action the AI takes is logged in the activity history under the user's own name.
See where your firm is today, and what stage 3 looks like on your own engagements. Request a demo.
.png)
Hundreds of conversations with firm leaders about resourcing, scheduling, and client readiness led us to build Collaborative Scheduling. Great team and client experiences rely on great planning, so we expanded our workflow to help leaders deliver a more consistent experience across all engagements.
Read moreRead more
AI is reshaping how audit teams work. None replaces an auditor's judgment, but they can remove the overhead that stands between auditors and the moments where judgment is applied.
Read moreRead more
All-in-one platforms may simplify your tech stack today, but they can create dependencies that make it harder to evolve tomorrow. This post explores why best-in-breed technology, connected through strong integrations, gives firms the agility to continuously improve, adopt better tools, and build a more resilient tech stack.
Read moreRead more
