Every firm has a number nobody reads out at the partners’ meeting, which is where WIP write-off in professional services AI can make an unglamorous but valuable difference. It is the WIP write-off: hours that were genuinely worked, recorded late or not at all, then quietly removed from the invoice before it went to the client. In professional services, AI usually gets pitched at the glamorous end, drafting documents and answering research questions. The unglamorous win is bigger. Catching unbilled time before someone decides to write it off puts money back in the business that you have already spent to earn.
This post covers where unbilled time actually disappears, why timesheet software has never fixed it, what an AI system can see that a timer cannot, and how to start without replacing your practice management platform.
Key Takeaways
- WIP write-off in professional services is rarely a pricing problem. It is a recording problem, and AI catches it while the time is still billable.
- Most unbilled time disappears in the 48 hours between doing the work and writing it down, not on billing day.
- Scope creep is the second largest source of write-offs, because the extra request arrived as a two-line email nobody logged.
- An AI brain that reads email, calendar and call notes can reconcile real activity against timesheets and flag the gap the same week.
- Write-off decisions made at 9 pm on billing day are guesses. The same decision made with evidence usually keeps more of the fee.
- A one-point lift in realisation on a seven-figure fee base is worth more than most efficiency projects a firm runs in a year.
- Start with the context layer: rates, engagement letters, scope, client history. Automation without that produces confident nonsense.
Where Unbilled Time Actually Goes
Ask a partner why they wrote off eleven hours on a job, and you will get a version of “the client would never have worn it.” That is the story told at the end. The truth is usually further upstream, and this is where an AI system wins where it falls short of traditional timesheet software.
The first leak is the gap between doing the work and recording it. Someone takes a call in the car, spends twenty minutes solving a problem, and gets to the office with three new things on their list. The timesheet entry happens on Thursday for a Monday call, if it happens at all. Reconstructed time is always conservative time. Nobody rounds up when they are guessing.
The second leak is scope that nobody logged. A client emails a two-line question that turns into two hours of work. It was never varied in writing, so when it appears on the invoice it looks like a surprise. Rather than have the conversation, the line comes off.
The third leak is the write-off decision itself, made late, under time pressure, with no evidence attached. The person signing off the invoice cannot remember what happened on the file six weeks ago. Faced with a number they cannot justify, they discount. That is not commercial judgement. That is a lack of information dressed up as commercial judgement.
Thomson Reuters Institute has tracked realisation across law firms for years, and the direction has been steadily down, even as billing systems have improved. That should tell you something. More timers did not fix it. The Thomson Reuters Institute research is worth reading if you want the sector-wide view.
Why Timesheet Software Was Never Going to Fix This
Practice management systems are excellent at storing time. They are useless at noticing that time is missing. That is the whole problem in one sentence. That is where it wins.
A timesheet is a self-report. It relies on a busy person accurately remembering and voluntarily recording something that feels like admin at the end of a day already stretched too thin. Every firm I have seen has responded to this by adding pressure: reminder emails, utilisation dashboards, a weekly nag from the practice manager. Pressure does not create memory. It creates rounder numbers and more resentment.
The same pattern shows up everywhere in a founder-led firm, which I have written about in admin overload in small business. The system captures what people put in it and knows nothing about what they actually did.
Meanwhile, the evidence of the work is already sitting in the business. Sent emails with timestamps. Calendar entries. Call logs from the phone system. Document versions saved to the file. Notes taken in a meeting. None of it is connected to the timesheet, so none of it is doing any work for you.
That is the gap. Not a shortage of data, a shortage of anything that reads the data and asks the obvious question: this person spent ninety minutes on the Henderson file on Tuesday according to their sent items, their calendar and the document history, so why is there no time entry against it?
What AI Catches That a Timer Cannot
Once the AI can see the real activity, three things become possible that were not possible before.
It reconciles activity against recorded time weekly
Every Monday, the system compares what actually happened on each matter or job against what got recorded. Not to police anyone. To produce a short list: six files where there is clear evidence of work and no corresponding entry. The person who did the work gets a prompt naming the file, the date and what the evidence was. Filling that gap takes ninety seconds while it is still fresh, instead of a shrug six weeks later.
It flags scope drift on the day it happens
The AI reads the inbound client email that starts “just a quick one.” It compares the request against the engagement letter it already holds. If the request sits outside the agreed scope, it tells the responsible person before the work starts, with a draft variation note attached. Scope conversations are easy when they happen the same morning. They are almost impossible after the fact.
It builds the case for the fee before billing day
By the time the draft invoice appears, the system has assembled the file narrative: what was requested, what was done, which unexpected issues came up, how many rounds of review it took. The partner reviewing the invoice is no longer guessing. They are reading a summary of the actual work. Some of those write-offs will still be the right call. Most of them will not survive contact with the evidence.
This is agentic work rather than a chatbot, which is worth understanding before you buy anything. I have covered the difference in agentic workflow examples.
Build the Brain Before You Build the Bots
Here is where most firms get this wrong. They buy an AI tool, point it at their data, and get back nonsense with a confident tone. The tool flags fifty items a week; half are wrong, and everyone stops reading it within a month.
The reason is always the same. The AI has no context. It does not know your charge-out rates, which clients are on fixed fee versus time and attendance, what your engagement letters actually promise, which matters are being run at a deliberate loss for a relationship reason, or that the Wednesday partners’ meeting is internal and not chargeable to anyone.
Context first, then automation. It wins where you take the time to write down the things that currently live only in a partner’s head: how jobs are scoped, what gets billed and what does not, which clients get latitude and why, what a write-off actually needs to clear. It is unglamorous, and it takes a couple of sessions. It is also the difference between a system that earns trust in the first fortnight and one that gets muted.
The pattern generalises well beyond time recording. Once the brain holds how the firm works, the same context drives client onboarding, matter reporting, and the daily summary that tells you which jobs are drifting. I have written more on sequencing this properly in AI adoption for SMEs and on measuring whether it stuck in the 90-day AI automation checkpoint.
One more thing worth being honest about: the system has to leave a trail. If it prompts someone to add time, the evidence behind that prompt needs to be visible and auditable, both for internal trust and for the client conversation that might follow. That is a design requirement, not a nice-to-have, and it is covered in AI audit trails for business.
Run Your Own Numbers Before You Spend Anything
You do not need a vendor to size this. You need twenty minutes and your own data.
Take last financial year. Pull total WIP written off across the firm. Then pull your realisation rate, the percentage of recorded time you actually collected. Most firms I talk to have never looked at the two together, and the combined figure lands somewhere between uncomfortable and alarming.
Now take the smaller of the two estimates. Assume you recover only the time that was genuinely worked and genuinely forgotten, not the discretionary discounts. In a firm of eight chargeable people, an hour a week each of recovered time is roughly four hundred billable hours a year. Put your own average rate against that. It is usually a bigger number than any software licence in the building.
Then ask the harder question. What would a one-point improvement in realisation be worth on your current fee base, and how many other projects on your list this year could deliver that? If your current approach falls short the verdict is straightforward: the revenue is already there, but traditional methods simply let it slip through the cracks.
The same logic drives every recovery project I run. When a finance broker asked me to work a database of 319 dormant leads his team had written off entirely, the AI reactivation recovered $49,000. The revenue was already there. It just needed something with the patience to go and get it. Unbilled time is the same shape of problem, sitting closer to home.
The Real Problem Underneath the Write-Off
For firms considering WIP write-off in professional services AI, the underlying issue is that WIP write-off is a symptom. The condition is that your firm has no memory of its own work outside the heads of the people who did it. Timesheets are a lossy attempt to write that memory down by hand, at the end of a long day, by people who would rather be doing anything else.
Fix the memory and the write-off shrinks on its own. Not because anyone worked harder or filled in their time more diligently, but because the business finally knows what it did. That same memory then does other work for you: briefing a new hire, telling you which matters are running hot, answering the question a client asks about a file from eighteen months ago.
That is what an AI brain plus an AI workforce actually means in a professional services firm. If your current systems fall short the verdict what you really need is simple: the brain holds how the firm operates, while the workforce does the recording, checking, and chasing that nobody in the building has ever enjoyed doing.
If you are a partner reading this at 9pm on billing day, deciding what to cut off an invoice you cannot fully account for, that is the version of the problem worth fixing.
Where to Start
If you want to work out what this is worth in your firm specifically, book a 30-minute discovery call. Bring last year’s write-off number and your realisation rate. Half an hour is enough to tell whether this is a genuine opportunity for you or a distraction, and I will tell you plainly which one it is.
Frequently Asked Questions
What is a WIP write-off in professional services?
Work in progress is time and disbursements recorded against a job but not yet invoiced. A WIP write-off is the portion of that value removed before billing, so it is never collected. It happens for good reasons, like an agreed fixed fee, and for bad ones, like time that cannot be justified because nobody remembers what happened on the file.
Why do professional services firms write off so much unbilled time?
Three causes account for most of it. Time recorded late or not at all, because the entry happens days after the work. Scope creep that was never varied in writing, so it looks like a surprise on the invoice. And write-off decisions made under time pressure with no evidence attached, where discounting feels safer than a conversation with the client.
Can AI actually recover unbilled time or does it just report on it?
Both, if it is set up properly. Reporting alone changes nothing. The useful version prompts the specific person, on the specific file, within days of the work, with the evidence attached. Recording the time is still a human decision. The AI removes the memory problem that made the decision impossible in the first place.
Do I need to replace my practice management system to do this?
No, and you should not. The AI reads from your existing systems and writes back into them. Your practice management platform stays where it is. Replacing it is a twelve-month project with a high failure rate, and it does not solve the recording gap anyway, because the new system will still be waiting for someone to type in their time.
Is it safe to let AI read client emails and file notes?
It has to be handled carefully. Access should be scoped to what the task needs, processing should sit inside systems you already trust with client data, and every prompt the system produces should be traceable to the evidence behind it. For firms with privilege or confidentiality obligations, this is a design conversation to have before anything is built, not after.
How long before it starts catching unbilled time?
The context work takes a couple of sessions. After that, the first reconciliation runs inside the first fortnight and usually surfaces obvious gaps immediately. The accuracy improves over the following month as the system learns your exceptions. Judge it at ninety days, when you can compare a full billing cycle against the same period last year.
What is a good realisation rate for a professional services firm?
It varies by discipline and by how you bill, so the useful benchmark is your own trend rather than a sector average. If your realisation has been drifting down while your recorded hours hold steady, the problem is almost certainly collection and recording rather than pricing. Track it monthly by partner and by client, not annually across the whole firm.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.