The AI tools an audit team reaches for today are good, and getting better. Ask Claude or ChatGPT a hard technical question and the answer often holds up. You can hand one a great deal in a single go, the amount these tools can take in has grown enormously, and the conversation is still waiting when you come back to it tomorrow.
These tools are capable, but reading a file and knowing it are two different things. Hand one the whole engagement and it won't tell you which numbers carry the risk, which policy governs which account, or why last year's team concluded what they did. Pile more in and the answers often get worse rather than better. And it doesn't grow sharper about the client as the weeks go on, so the work of pointing it at the right thing stays your job.
An audit is months of work that builds on itself. What it calls for is less like a tool you keep re-briefing and more like the colleague who already knows the client: something that carries the whole engagement, has the right part on hand for every task, and understands it better the longer the job runs.
Think of the manager who's been on a client for years. Ask them about a revenue figure and they don't reach for a file. They already know the recognition policy behind it, the risks flagged in planning, the name of the client's revenue clerk, and the three customers who make up most of the balance. The engagement lives in their head, and it grows sharper every season they're on it. That knowledge is what makes them fast, and it's most of what makes them good.
Relay is built to be that manager. Here's what it knows, how it keeps the right part to hand, and what that does for the work.
Why more isn't the answer
It's tempting to think the answer is simply more room: hand the model the entire engagement and let it sort things out. It doesn't work that way. A model reads everything you give it in one pass, and the more you pour in, the thinner its attention spreads, the way a junior handed forty lever-arch files and told to keep all of them in mind at once will keep none of them well. Past a point, more information makes the work worse rather than better.
So the job was never to feed the model everything. The job is to hand it the right slice at the right moment: when the question is about revenue, the revenue policy and last year's workpaper and the sample, and none of the payroll reconciliation or the fixed-asset register. Working out which slice matters for each task, and keeping that current as the engagement moves, is the hard part. The field calls it context engineering, and it's where most of the real work in a system like Relay lives.
A map, not a pile
You give Relay the engagement: prior-year workpapers and AFS, the trial balance, the client's accounting policies, the supporting evidence, the email trail. Most tools would file that as a stack of documents and search it one at a time. Relay reads all of it and records how it fits together: which policy governs which account, which piece of support backs which number, who at the client provided it, and what last year's team concluded and why.
Those connections become a knowledge graph: a single, linked map of the engagement.
Reading the engagement
Relay reads the whole engagement into one connected map — last year's file, the ledgers, the policies.
High Level Audit Knowledge Base Example
The map is what lets Relay know the engagement the way that manager does. Ask about revenue and it follows the links to the policy, the planning risks, the sample, and the support. It pulls exactly that slice without dragging the rest of the file along. It's the same map Relay works from to perform testing, build workbooks, and answer a query.
How your firm does things
Two firms can run the same test and document it nothing alike: different templates, different sample sizes, different thresholds, different bars for what counts as sufficient support. Both satisfy the ISAs; each simply has its own way of working. A good auditor carries their firm's way in their head and applies it without being told. Relay does the same.
It works to your templates, your sampling rules, your standard for valid support, your house style for a workpaper that's ready to review, applied the same way on every engagement and by every person on the team. Where your methodology is silent, it falls back on the ISAs, which it holds in full.
The map that keeps up
An engagement is never fixed, and neither is the map. As new support arrives and figures change, Relay folds each update into the graph, so the picture it works from is always the current one.
This is where knowing the whole engagement earns its place. The inevitable updates to the trial balance are made. Materiality gets revised. Relay sees exactly which figures moved, and because it holds the map, it knows which accounts they touch and which completed work they unsettle. The version-control scramble, the afternoon spent hunting down every schedule that referenced the old number, goes away.
What it looks like on a job
Take revenue testing: weeks of work on every audit, and the kind of job that trips over its own early decisions.
Relay starts from the engagement it already knows: last year's revenue workpapers, the client's recognition policy, this year's trial balance, the sales ledger, the relevant client correspondence. Before any testing, it reaches for your firm's methodology: your template, your sampling rule, your standard for support. It builds the workbook, states each procedure, and selects a sample. The sample goes to Relay's Open Items, ready to request from the client. Support comes back, gets inspected, and drops into the map, ticked in and referenced in the workbook.
Then reality arrives. The support isn't quite what was asked for. A revised trial balance comes in. An error pushes you into extended sampling. Relay adjusts the workbook and the documentation as it goes, and keeps a clean record of every turn. No one had to re-explain the engagement at any step, because Relay already knew it.
What this gives a team
- Less explaining, more auditing. Relay doesn't need catching up on the engagement each time you sit down. Your people spend their hours on judgment.
- Consistency. Your firm's methods, applied the same way, on every job and by every person.
- A file a reviewer can follow. Every decision carries its reason and where it was applied: traceable, checkable, and quick to check.
That last point marks the line we won't cross. Relay does the groundwork and the drafting; the judgment, the sign-off, and the opinion stay with you. When the evidence is thin or won't reconcile, it tells you, rather than papering over it.
An audit only grows as it runs. Because Relay knows the whole engagement, and knows it better the longer the work goes on, none of what's already done has to be rebuilt. It works less like a tool you keep re-briefing and more like a colleague who has been on the client from day one. The manager who already knows.