PwC, Deloitte and BDO India put Model ML's AI on live client work, inside a $10 billion consultancy AI bet

The go-live is the clearest usage test yet of consultancies' $10 billion-plus AI gamble: whether software can take over the junior-analyst hours behind advisory margins. Model ML is valued above $1 billion on exactly that bet.

Vincent JiangVincent Jiangthree-minute read
Share
Screenshot of Model ML's Modules workspace, listing AI modules for industry research reports, list building rankings, public comps and deal logging by number of steps.
1 / 6Slide 1 of 6
Model ML's Modules workspace: the platform's research, comps and deal-logging agents that PwC, Deloitte and BDO India run on client work.

Live deals, not a pilot

BDO India's deal teams are running live client due-diligence work on an AI platform from London startup Model ML, announced November 5, with the system configured around the firm's existing processes rather than run as a trial 1. BDO is the third major advisory name to convert in under four months. PwC began a global rollout of agents built alongside Model ML in July 15, and Deloitte signed a multi-year agreement covering document review, due diligence and transaction analysis weeks later 15.

A $1 billion valuation riding on junior hours

Model ML, founded in 2024 by brothers Chaz and Arnie Englander, automates the document-heavy work of finance: pitch books, investment memos, diligence and valuation models 14. The platform routes each task across models from OpenAI$1.18T — OpenAI, private, latest valuation $1.18T, Anthropic$350B — Anthropic, private, latest valuation $350B, Google and open-source providers, scoring them on accuracy, latency and cost 1.

It has raised more than $100 million in 18 months, including a $75 million Series A last November 56, at a reported valuation above $1 billion, or £748 million 12. The pool it sells into is deep: the largest accounting and consulting firms have committed more than $10 billion to AI since 2023 1, much of it aimed at research, decks and spreadsheet work that junior staff used to do 2.

Usage is not a return

The sector's open question, put by City A.M. on November 9, is where the real return on that spend lands 3. BDO is the first usage ledger. In eight weeks its teams asked more than 10,000 questions, analyzed more than 25,000 documents and generated more than 14,000 AI review comments 2. Eight weeks of logs show the platform is being used; nothing published shows it pays.

BDO's first eight weeks on Model ML: 25,000 documents, 14,000 review notes, 10,000 questions

05,00010,00015,00020,00025,000Documents analyzed25,000+AI review comments14,000+Questions asked10,000+
Data
Value
Documents analyzed25,000+
AI review comments14,000+
Questions asked10,000+
Counts as reported from Model ML's BDO case study, covering the first eight weeks of deployment. Source: Infercline, November 10, 2026.2

Reputational risk keeps the checking costs in place 2, and the account list keeps compounding anyway: three of the Big Four live by the company's own count 4, HSBC and Santander as customers 1, a global HSBC rollout in September 5, an equity stake from HSBC Asset Management in August 6, and a Dubai office opened in late October to court sovereign funds 4.

The same budgets have other suitors

Model ML is not alone in converting analyst hours into software. Rogo's~$1B — Rogo, private, latest valuation ~$1B annual recurring revenue passed $50 million on clients including Lazard and JPMorgan, and Hebbia$670.7M — Hebbia, private, latest valuation $670.7M is approaching $50 million on a contract covering about 5,000 Morgan Stanley−0.14% — Morgan Stanley, down 0.14 percent today bankers 7. The labs sell the same work: OpenAI shipped ChatGPT for Financial Services with Morgan Stanley and Evercore−0.03% — Evercore, down 0.03 percent today as design partners, and Anthropic debuted ten finance agents in May 7.

Deloitte senior partner Udit Gandhi says his firm picked Model ML because it is "more than just a generalist tool with a finance skin on it" 1. The Financial Stability Board, the global standard setter, adds its own line item: final guidance on October 15 demands more human oversight of agents, which it warns "do not reason like humans" 8.

The next hard reading is the graduate intake

If these deployments hold, the cost base they eat is the pyramid of junior analysts under consultancy margins 2. The next hard readings are whether the fourth Big Four firm signs and what the Big Four's next graduate intake looks like. Until one of these firms publishes the return, the usage counters are doing the accounting.

Get Applications by email, free

How AI is changing the software people and companies use. In our daily email, every weekday.

Loading MS

Reader comments

0 comments

    More in Applications

    All in this topic →

    Applications

    Spotify sells human narration at 15 hours a month as it plans a $4-to-$5 AI audio tier for 2027

    Spotify now sells 15 hours of human narration at ₹199 a month in India and bundles the same hours inside $12.99 and $21.99 US Premium plans, while a planned $4-to-$5 flat-rate AI remix tier is due in early 2027. Voice AI is pulling down the price of an hour of speech, and the listener may be left holding the risk.

    Applications

    HubSpot brings back co-founder Brian Halligan as CEO with the stock down nearly half in 2026

    Yamini Rangan, chief executive since September 2021, leaves the board the same day, and the filing announcing her exit also reaffirmed guidance for fourth-quarter growth of about 12 percent. A stock down nearly half in 2026 gets a new strategy story while its numbers stand.

    Applications

    Adobe opens Photoshop and Premiere files to ChatGPT, Gemini and Claude agents as net-new ARR falls 37% to 38%

    Coworker is free, and the Google deals carry no price, date or term. With Adobe's new subscription sales falling sharply, December's report will show whether open files and a free agent defend Adobe's licenses or hand the agent layer to its rivals.

    Applications

    Priceline's Claude-powered travel bot Penny goes live in Snapchat chat ads

    Snap's newest ad format is a rented conversation running on Priceline's bot and Anthropic's models, while subscriptions, not advertising, supplied about three quarters of Snap's growth last quarter. That split is what SNAP holders are actually buying into the holiday quarter.

    Applications

    EU plan would let users switch off the streaks Duolingo's paywall now runs on

    The EU's Digital Fairness Act would let consumers switch off streaks days after cheap AI models destroyed Duolingo's ability to charge for its best feature, leaving gamified friction to carry the paywall that Brussels now targets and shareholders are suing over.

    Applications

    ADP's AI ran more than 4,400 interviews as its own hiring gauge cooled from 26,000 to 19,750 jobs a week

    ADP's AI rollout is a cost story inside a raised margin guide, and the weekly US hiring gauge ADP itself publishes has cooled from 26,000 jobs a week in late September to 19,750. That split, automated cost and cooling volume, is what shareholders now hold.

    Get the newsletter

    The day's AI and market news, in your inbox every weekday.