Financial Due Diligence
Financial services · Financial due diligence
Thread AI Editorial
September 28, 2026
A Worker that reviews the books of potential M&A targets, and drafts a first-pass quality-of-earnings report, so deal teams start with targeted and flagged adjustments instead of raw ledgers.
Fewer analyst hours
Per engagement, as the team starts by reviewing flagged first-pass reports instead of combing through ledgers
A faster QoE draft
Earnings issues surface while there's still time to act on them before the deal is signed
More engagements
From the same team, as deal teams spend less time on data prep and more on judgment calls
The challenge
When a private equity firm or corporation buys a company, it typically hires an external firm to run a quality-of-earnings review, which shows how much of the company's reported profit is real and will last.
The review starts with the company's full transaction record, often spread across several accounting systems that each categorize expenses differently. Analysts must first merge everything into one format, then comb line by line for costs that distort the picture, like an owner's personal expenses or a one-time legal settlement.
Under deal deadlines, the team can end up spending most of its time finding these issues by hand instead of analyzing a company’s integrity. Costs pile up, and effectiveness can decrease.
The solution
The team built a Worker on Lemma that pulls transaction records from each of the company's accounting systems and merges them into one format. The Worker then flags costs that may need adjusting, like personal expenses and one-time items, and links each flag to the transactions behind it so analysts can check it quickly. The team adds a human-in-the-loop step so that nothing goes into the review until someone approves it.
Analysts start with a draft of flagged adjustments instead of raw data, and the firm can take on a company built from many acquisitions without needing to staff up the engagement.
Lemma Capabilities
1
Clean data from messy systems.
Code, not AI, converts each accounting system's export into one format, then checks the result against the company's own totals.
2
The AI never supplies a number.
The model identifies which transactions belong to each flagged item and explains why, citing the evidence. Every dollar figure is calculated in code from the company's actual records, so no total can be invented.
3
Built-in data checks.
At each step, checks confirm that no transaction was dropped, double-counted, or made up, and that everything the draft cites actually exists. If any check fails, the run stops rather than producing a flawed report.
4
Required sign-off.
A specialist reviews every flagged item before the final report exists, correcting pre-filled entries rather than starting from scratch. Every decision is logged, giving the team a record to defend each adjustment in negotiation.