Every Monday, Blackstone's investment committee convenes to decide on acquisitions. Nearly every deal presented is now accompanied by two decision memos. One document comes from the transaction team, the other from an internal AI model. This system was trained on internal memos and negotiation records spanning more than 14 years. Although the model has no voting rights on the committee, its assessment now carries established weight.
At the same time, Torsten Slok, chief economist at Apollo, cautions that outside the technology industry there are so far hardly any signs of AI-driven margin expansion. Inadequate data governance could delay anticipated returns by years. For the commercial real estate sector, these concerns hold true in many quarters. According to a survey by the Dealpath platform among 103 investment and technology specialists at institutional real estate firms, 97 per cent stated that they had integrated AI into their investment processes. However, only 51 per cent found that using it actually saved time once the verification effort was taken into account.
The verification costs of fragmented data
The return on AI is currently unevenly distributed. Around 55 per cent of respondents trust AI systems to summarise due diligence documents. However, when it comes to evaluating and prioritising deals, this trust falls to 35 per cent. This is not due to the underlying language models, which are equally available to many market participants, but to access to internal company data. While a single document is sufficient to summarise a file, transaction scoring requires the firm's institutional memory: previous acquisition decisions, rejected bids, historical comparable data and current market context.
Firms with assets under management of at least $40 billion rate their data infrastructure highest and place the greatest trust in algorithms for strategic tasks. Across the industry, 83 per cent of specialists consider their database ready for AI, yet 90 per cent concede that fragmentation or quality issues diminish its utility. As systems make assumptions when datasets are incomplete, 94 per cent of respondents almost always verify results manually. This review process consumes the very time that automation was supposed to save.
Centralisation rather than a proliferation of tools
Companies often respond to sobering AI results by introducing additional software. Around 71 per cent of survey respondents turn directly to standard assistants such as ChatGPT. However, each isolated tool sees only a fraction of the overall context, lengthening rather than shortening review cycles. Successful investors take the opposite approach: they focus on a small number of systems that are linked directly to historical deal databases and market activity.
The decisive competitive differentiator lies not in off-the-shelf licensable models, but in an organisation's consolidated wealth of experience. Firms that systematically unlock their internal data archives can integrate AI soundly into acquisition reviews and committee decisions, while other market participants are still busy reconciling inaccurate model outputs.
