Altvia partnered with UserEvidence to survey 187 alternative investment professionals in September 2026 on market pressure, LP expectations, technology, and AI. The full report publishes in November, 2026. Here is an early look at some of the findings.
AI adoption is real, but uneven
Firms are spread across the maturity curve, and most are past the pilot stage.
The other 39% are still at the pilot stage or earlier.
The barriers aren’t about the models
When we asked what stands in the way of adopting or expanding AI, the top five answers were:
Security or confidentiality concerns: 49%
Integration with existing systems: 48%
Trust in AI outputs and accuracy: 47%
Lack of in-house skills or expertise: 40%
Data quality or readiness: 37%
None of these is about how capable the models are, they’re about safety, connection, trust, and who can use the tools. Only 16% cited unclear return on investment.
The barrier changes as firms mature
The biggest obstacle is not the same at every stage. Early on, the problem is skills. At the pilot stage, it is connecting AI to existing systems. Once AI is in production, security and confidentiality take over.
Security concern jumps from 23% of firms still exploring to 55% of firms running pilots, and stays above 50% from there.
Skills concerns fade as firms gain experience, falling from 57% of early-stage firms to 32% of those in production. The same three themes appear throughout: who can use the tools, whether the guardrails hold, and whether the data is connected and current. Ryan Keough explores why in 3 Things to Learn from the Automobile When Deploying AI.
Ambition is running ahead of the foundation
91%expect to spend more on AI.53%are investing in the data it depends on.
The firms that are investing report more progress:
Investing in data infrastructure
Not investing
AI in production or embedded
69%
52%
Technology helps them keep pace
82%
66%
“Without quality data, AI can’t do its job,” says Lisa Lewis, Industry Product Manager at Altvia. “Data infrastructure is the next big Achilles’ heel, but unlike launching a pilot or establishing a policy, it’s often a multi-year commitment that nobody gets credit for.”
LP expectations are rising across the board
Respondents reported higher LP expectations over the past 12 months in every area we measured:
LP expectation
Share reporting an increase
Transparency into fund performance
87%
Speed of response to requests
86%
Frequency of reporting and updates
85%
Self-service access to data
84%
Detail and granularity in reporting
82%
What’s next
The full report goes deeper on how these patterns vary by firm size and role, and on what firms are doing about them. Subscribe to get the report.
How Investor Engagement Manager Enables Firmwide Investor Relationships
Summary: Most AI in investor relations delivers value only to the people with the expertise to operate it. Turning that into a strategic advantage the whole firm holds, and one that accelerates, takes three things in place at once. The automobile needed the same three, and took thirty years to assemble them.
By Ryan Keough
For AI to actually move an organizational objective such as investor relations, fundraising, or the strength of every LP relationship a firm holds, three elements have to be in place at the same time.
There is a version of AI success that firms are quietly settling for, and the awkward thing about it is that it works. It produces genuinely useful output, but only for the people who know the data, know how to ask, and can tell a good answer from a plausible one.
That is a real skill. It is also unevenly distributed, and today’s AI is almost entirely dependent on it. The value a firm gets is capped by how many people it has who hold that expertise, and the colleague at the next desk gets something worse without being able to say why.
The three elements
That is a set of capable individuals, not a capable firm. It does not compound, because the value lives in the operators rather than in the organization. And it is not fixed by training, because the dependency is structural: a system that only works in expert hands will only ever reach as far as your experts do.
One. It has to be accessible to the entire team. Usable by everyone who touches the relationship, not dependent on an expert user who knows how to work it.
Two. The guardrails have to be in place. Safe, reliable and governed, which is what allows a team to move fast rather than what slows it down.
Three. It has to have the essential inputs. Clean, connected, current data, arriving continuously, without anyone hand-crafting it.
Any two without the third leaves you with a pilot. All three together is a different thing entirely.
The automobile is the proof
The car is the clearest example of all three arriving, and of how little the machine alone was worth until they did.
Accessible. Early automobiles were for mechanics and enthusiasts: difficult to start, harder to maintain, and priced for the wealthy. The Model T mattered because an ordinary person could afford it and operate it1. Mass value required mass accessibility, not a better engine.
Guarded. Then the roads. Early roads had no lane markings, no stop signs, no signals, no right-of-way conventions and no requirement that the driver knew how to drive. The control layer was invented piece by piece over two decades. Center lines appeared in several states between 1911 and 1917, the first stop signs and signals around 1914 and 1915, and national uniformity only with the first Manual on Uniform Traffic Control Devices in 1935.2 Until then, driving at volume was dangerous rather than useful.
Fueled. And the fuel. Crude oil does not run an engine; gasoline came out of a refining process, as a byproduct of making lamp kerosene, and was so plentiful that refiners burned it off.3 It was never scarce. It simply was not where the driver was. Before filling stations you bought it in cans from a pharmacy or a blacksmith, and planned long trips by telephoning ahead.4 Quality varied with whatever crude it came from, and the property that actually mattered, knock resistance, had no agreed measure until the octane scale arrived in 1930.5
A car nobody could drive, on roads nobody could survive, running on fuel nobody could find, is a hobby. All three together built an industry.
And what the three delivered together was not a better car. It was access. Ordinary people could travel routinely and arrive safely, which changed where they could live, where they could work, and how far a business could reach. Freight moved by road. Rural isolation eased. A machine that had been an enthusiast’s curiosity became infrastructure an entire society ran on, and the people who had it could do things the people who did not simply could not do.
The elements were the enabler. The advancement was the point.
The same three, in our clients’ own numbers
We asked 187 private markets professionals in September what is blocking them.6 The answers map onto those three elements almost exactly.
THE ELEMENT
IN THE CAR ERA
IN AN AI SOLUTION
BARRIER IN OUR SURVEY
Accessible to everyone
A car an ordinaryperson could afford,start and drive, not amachine for mechanicsand enthusiasts.
Value available to the whole team without an expert operator in the middle.
In-house skills: 40%
Guardrails in place
Signs, painted lines, barriers, and rules of the road, so speed was survivable.
Permissions, traceability, auditability, and human approval, so the firm can move fast.
Security 49%, trust 47%
The essential inputs
Refined fuel, available where you needed it, at consistent quality.
Clean, connected, continuously replenished data, delivered without manual effort.
Integration 48%, data readiness 37%
Security 49%, integration 48%, trust in outputs 47%, in-house skills 40%, data readiness 37%. Not one of the top five is a model problem. Every one is one of the three elements being absent.
It shows where firms are stuck: 19% informal or exploring, 20% running structured pilots, 42% in production, and only 19% with AI embedded in day-to-day work. Four in ten have not reached production at all. The middle of that curve is crowded with firms that have proven something works for somebody and cannot get it to work for everybody.
Investor Engagement Manager was built around these three elements rather than around a model. Each one is a deliberate design decision, and the combination is what turns AI from a capability into an advantage a firm actually holds.
How IEM puts the three together
ONE · ACCESSIBLE TO THE ENTIRE TEAM
IEM changes how the work gets done, and it does it for everyone.
The expertise the current generation depends on is not something IEM asks the user to supply. The intelligence is surfaced by the system and embedded in it, rather than drawn out of it by someone who knows how. There is no prompt to compose, no need to know where the data lives, no judgment call about whether an answer can be trusted, and nothing a person has to understand about how AI works in order to benefit from it. IEM surfaces what changed across every LP relationship, explains what is driving it, and puts a recommended next step in front of the person who covers it.
That design choice does something a better model cannot. Because the experience is the same for everyone, the quality of preparation is the same for everyone. The associate, the IR lead and the partner all start from the same place. Value stops being a function of who is good at the tool, which means it scales with the size of the team rather than with the number of experts on it.
TWO · GUARDRAILS THAT LET THE TEAM MOVE FAST
A system that moves first has to meet a higher standard, and IEM is built around four commitments.
Secure. IEM reads inside the firm’s existing permissions rather than building a second access model beside them. What it surfaces to any individual stays bounded by what that person could already see.
Traceable. Every alert, assessment and draft ties back to the record it came from. Nobody is asked to accept a number on faith.
Auditable. What was surfaced, what was recommended, who saw it, who approved it and what was done, all reconstructable afterward.
Human-led. Nothing consequential reaches an investor or changes a record until a person approves it.
The instinct is to read that list as friction. The roads say otherwise. Lane markings and signals did not restrain traffic; they are the reason traffic could move at volume and at speed without constant collision. Guardrails are what make speed possible, and they are the difference between a firm piloting AI in one corner and a firm running it everywhere.
THREE · THE INPUTS THAT MAKE IT WORK
No firm in private capital is short of data. It accumulates as a byproduct of the work, whether email, documents, fund reporting or notes, and like crude in the ground, none of it runs anything in the state it arrives in. It has to be refined: resolved, connected, deduplicated, put in context. Most of it never is. It sits where it landed.
Firms tend to judge their data by whether the CRM looks populated, which is the equivalent of grading fuel by specific gravity: easy to measure, and not the property that determines whether the engine runs.
Altvia’s underlying data management and data model is what turns that raw material into an enabler rather than a constraint, and the governing principle is that refining cannot depend on the user. That is the lesson of every CRM adoption program ever run: if the value requires people to faithfully enter data first, the value does not arrive. So we handle it three ways.
1. We refine the data you already have. Existing records cleaned, connected and put right, so the firm’s history is usable from day one rather than after a migration project. Nobody starts with an empty tank.
2. We embed data consumption, so nobody hand-crafts the fuel. Integrations and imports keep other systems feeding in automatically. The picture stays current without anyone copying between tools or maintaining a second record by hand.
3. We consume unstructured email automatically. Email is the richest and least governed record of an LP relationship. It is captured, cleaned, organized and recorded into the CRM without user involvement, and made available to IEM without anyone filing anything.
The common thread is that none of the three require a person to be diligent for the system to work. That matters more than it sounds. A system reading only what people remembered to type is reading a fraction of the relationship, and the fraction is biased toward whoever is most dutiful about paperwork, which is rarely the person closest to the investor.
This is how siloed data becomes organizational intelligence: refined, pumped into the tank, and available to everyone entitled to see it.
No one person holds the whole LP relationship. With the inputs handled this way, the firm holds it, even when the individual who built it has moved on.
From changing the work to changing what the firm can do
Each element on its own is worth something and capped. Accessibility without guardrails is a fast car on an unmarked road. Guardrails without accessible value is governance over a pilot nobody uses. And either of those without clean, continuous inputs is an engine with nothing in the tank.
Together they change the work itself. Coverage stops depending on how many relationships a team has the bandwidth to work. Preparation arrives assembled rather than assembled by someone. The quiet relationship that was drifting gets surfaced while there is still time, and the re-up picture arrives months before a launch rather than during the raise.
But the change to the work is the smaller half of it.
Mass access to the automobile did not simply make individual journeys easier. It changed what was possible for the people and the businesses that had it, and it did so at the level of the whole enterprise rather than the single trip. The same thing happens here. A firm where every person works from the same quality of preparation, where coverage no longer tracks bandwidth, and where institutional memory survives the departure of the person who built the relationship, is not performing the old job more efficiently. It is operating at a level its competitors cannot reach by working harder, because working harder is precisely what the constraint always was.
That is a strategic advantage, not an operational improvement. And unlike an operational improvement, it accelerates.
Every interaction connected makes the next connection sharper. Every proactive pass improves the one after it. The firm’s accumulated history becomes an asset a competitor starting today cannot simply buy, because the only way to acquire it is to have been compounding it already. The gap does not stay the same size; it widens in favor of whoever assembled the three elements first.
That is the difference between deploying AI and gaining ground because of it.
Launching with all three
Investor Engagement Manager is available today, with the data refined and continuously replenished, the guardrails built in rather than promised, and the value accessible to every person on the team without an expert in the middle.
The automobile needed thirty years to assemble those three elements, and the assembling is what cost the time. The findings are published now. There is no reason for a firm in private capital to repeat the experiment.
In the first post of this series, we argued that every AI strategy has to answer four questions: what advantages does AI create, what are the risks, what is the integration plan, and in what sequence should that plan be executed?
The first two questions, what the advantages are and what the risks are, have answers specific to each firm. This post expands on the advantages and risks for the relationship that matters most to a GP: the one with the LP.
What AI can do for the GP/LP relationship
The obvious answers to the question of AI advantages are already familiar:
Draft an LP email
Read a year of interactions and estimate how healthy a relationship is.
Summarize a call and flag an overdue follow-up
Fill in the fields nobody filled in
These are useful, and every serious platform will offer some version of them. They are also generic AI use cases with a private markets vocabulary applied. A firm that stops there has bought a faster version of work it already does.
The strategic advantage is the firm’s capacity to hold LP relationships at depth. Today that capacity is set by headcount and memory. A team of three can know forty relationships well, and the rest get the newsletter. Institutional knowledge like who cares about what, what was promised in the last meeting, and which contact is the actual decision-maker, lives with the people who happen to cover the account. And then it’s lost when they leave the firm.
The advantage AI can deliver, if it’s built for this purpose, is that this deep relationship context, the kind of context that isn’t typically found in an account record, is housed and understood across the firm, all the time. Partner-level judgment can be applied on any of them at the moment it matters.
Why LP relationship depth matters more between fundraises
Capital is concentrating with managers LPs already know. Bain’s 2026 Global Private Equity Report finds that 53% of LPs are constrained from making new commitments because capital they already pledged has not been called, up 15 points in a year. The number of buyout funds reaching a close fell 23%. In this market, the relationship a firm carries into the fundraise decides most of the outcome before the first meeting.
The benefits of a deeper relationship compound in ways that show up across the private markets business:
An LP who knows the firm re-ups, and at a larger allocation
Diligence moves faster, because the DDQ confirms what is already established rather than examining it
First closes get anchored, which pulls the rest of the roster forward
Co-investment and separate account conversations get a serious hearing
A continuation vehicle has real support behind it when the firm needs one
A difficult quarter comes with room, and with people who speak well of the firm to the consultants and peers who shape the next fundraise
Every one of those outcomes is earned in the quarters when no fund is in market and where relationship capital is built through interactions like reporting delivered on time, questions answered fully, and follow-ups kept.
How LP relationships are lost between fundraises
It’s in these same quarters where relationships are lost, and they’re lost quietly. Most of the LPs a firm loses are still on the list when the fundraise opens. They took the calls, they received every report, there’s nothing in the file that indicates anything changed. Then the commitment comes back smaller, or as a polite note that this cycle is a pass.
This kicks off an internal process where the team tries to dissect the moment it went wrong. Was it a quarterly report that landed late during the LP’s reallocation? A champion who moved to a different mandate and a successor who never got the call? A follow-up from the annual meeting that stayed in a drafts folder?
Commitments recede before anyone declines. A firm can lose an LP before the fundraise begins and still has to earn every commitment when it does.
Why CRM and a disciplined IR team stopped being enough
Ten years ago, a good CRM and a disciplined IR team was a system that worked. Both still matter, what’s changed is the shape of the business around them. Firms added continuation vehicles, co-investment sleeves, separately managed accounts, and new strategies to stay relevant to the same LPs. Each addition brought its own reporting calendar, its own LP segment, and one more place for the story to diverge.
In Altvia’s interviews with IR teams this year, the recurring theme was learning about an important LP change later than they wanted and not having a system that would have surfaced it. They all knew the data was somewhere within the firm, but nobody had time to find it.
The risk question: the LP relationship answer
The generic risks are leakage and fabrication, and every vendor addresses them. The specific risks within an LP relationship define what the tools have to do.
Under the SEC Marketing Rule, a communication that offers advisory services to prospective or current private fund investors can be considered an advertisement and carries the obligations that follow. The line between an ordinary LP update and an advertisement is not always obvious. A system that drafts for IR therefore has to stop before the send and wait for a person to take the next step.
Additionally, every figure carries the firm’s reputation, so a number with no visible source is a number nobody can defend in the meeting. Partners want to know what the model can access and where the data goes. In one beta conversation this year a partner returned to that question three separate times.
And the IR and fundraising function is starting from behind: in Altvia’s forthcoming 2026 Private Markets Survey, investor relations and fundraising respondents report the lowest AI maturity of any function, a mean of 3.1 on a 0 to 5 scale, and the lowest confidence in AI accuracy, 29% against 47% of all respondents. A strategy that hands them a general-purpose assistant and a training session has misjudged both the advantage and the risk.
A Standard Emerges
Put the advantage and the risks together and a standard emerges:
AI for the LP relationship holds the full relationship in one place, with the deal and fund context that shapes it, so the firm’s knowledge of an LP no longer depends on who covers the account.
It captures that relationship with less manual effort each quarter, because a system that depends on someone remembering to update it inherits the failure it was meant to fix.
It brings the right slice of the relationship to the right person at the right moment, with the source shown.
And it proposes while a person approves.
That standard is what Altvia means by a GP Engagement Platform. Stay tuned! The next post in this series delves into the third question: what’s the plan, in practice?
Sources
Bain & Company, Global Private Equity Report 2026, Fund-raising section (published February 22, 2026). https://www.bain.com/insights/outlook-gaining-traction-global-private-equity-report-2026/
U.S. Securities and Exchange Commission, Investment Adviser Marketing, Release No. IA-5653 (the Marketing Rule, Rule 206(4)-1 under the Investment Advisers Act of 1940). https://www.sec.gov/rules-regulations/2020/12/investment-adviser-marketing
Last month I wrote about what I heard at PEI’s IR Network meeting: I saw a room full of IR professionals working out, in real time, how to use AI without letting it do the thinking for them. A few weeks later I was back in the room, this time at another PEI IR Network meeting. The same core questions were on the agenda, but this time was different.
Last month’s conversation was mostly internal: how do we use AI? How do we not get it wrong? What will our LPs think if they find out? Last week, the conversation had shifted. LPs aren’t just forming opinions about AI use anymore. They’re asking directly how GPs are using it, and that means GPs have to think differently about their AI strategy.
Here’s what stood out.
Personalized LP communications are the clearest AI use case, and also the trickiest one
The most concrete idea I heard came from a conversation about turning deal memos into short, personalized notes for LPs, rather than generic updates. It’s a good use case: the raw material already exists, and AI is well-suited to reshaping it into something that reads like it was written for one person rather than a distribution list.
But the group was candid about where the line sits. There’s a real distinction between the low-hanging fruit (a quick capsule summary) and what would actually be meaningfully impactful. This almost always requires pulling data from a lot of different sources and takes real work to get right: IC summaries, LP meeting notes, and reporting all live behind different systems, often in different formats. Nobody in the room felt like they’d fully solved this. As one person put it, the real bottleneck is data quality: garbage in, garbage out. The hard part isn’t the AI, it’s getting information into a usable format in the first place. It’s a theme that came up again and again last week: technology is rarely the constraint. The underlying data infrastructure is.
Confidentiality is a harder line than it was six months ago
This came up more specifically at last week’s meeting. Sub docs and other sensitive materials can’t just get pushed into a general-purpose LLM. That’s not a compliance nicety; it’s a real operating constraint that shapes which tools a firm can actually use and how. It’s consistent with what Deloitte found in its 2025 GenAI in M&A Survey of 1,000 corporate and PE dealmakers: data security was the single most-cited barrier to GenAI adoption, ahead of every other concern. For private capital firms, where sub docs, LPA terms, and side letters are about as sensitive as documents get, that distinction isn’t theoretical.
At the same time, LPs want more from their GPs, not less. More data and more transparency. So GPs are stuck trying to do both at once: securely share more, but never make it feel like a bot did it. One comment from the room stuck with me: “Nobody wants their IR comms to read like ‘a bot.'” That’s the same reputational risk I wrote about in my last post, except now LPs are actually asking the question, not just wondering about it.
LPs are starting to ask GPs directly how AI is used in the firm
It’s no longer hypothetical that an LP might wonder about a GP’s AI adoption. LPs are now asking. The data backs this up: Private Equity International’s LP Perspectives 2026 Study found that nearly half of LPs surveyed are closely monitoring how their GPs adopt AI in investment and operational processes, and almost as many report mixed feelings about it, largely driven by risk concerns. The ILPA DDQ framework, which the majority of institutional LPs now use as a baseline for manager evaluation, continues to expand with each cycle, and it would be reasonable to expect AI governance to become a standard section, rather than an ad hoc follow-up question.
If your firm doesn’t yet have a clean, honest answer to “How do you use AI, and what guardrails are in place?” it’s time to build one. Not because someone will send it back with a red pen if you get it wrong, but because you should probably be able to answer that question clearly, regardless of who’s asking.
The fee conversation nobody wants to have, but needs to
If firms are getting more efficient because of AI, will LPs start asking why fees haven’t come down? It was framed as hypothetical, but it’s not disconnected from what’s actually happening in the market. Bain’s 2026 GP Outlook already points to real downward pressure on headline management fees, particularly at scale. A recent bfinance poll found that a large share of LPs report fee reductions for like-for-like private market strategies, with softer fundraising and disappointing distributions shifting real pricing power toward investors.
AI efficiency gains and fee pressure are two separate trends right now. But it’s not hard to see an LP putting them next to each other, and GPs should probably have a point of view before that question shows up in a meeting. The best response is to show where the efficiency actually goes: faster answers, better reporting, more time for the relationship itself. That’s a different story than “we cut costs and kept the fee the same.”
MCP servers came up, unprompted
Someone in the room made an offhand comment: if you already know what an MCP server is, you’re ahead of the curve. It’s a small signal, but it’s consistent with where the infrastructure conversation is heading. Most firms are still thinking about AI at the level of individual tools and prompts. Fewer are thinking about the underlying protocols that let those tools connect to a firm’s systems and data securely. That gap between “using an AI tool” and “having AI infrastructure” is expected to keep widening.
Where this leaves us
The first meeting focused on where and how IR teams should utilize AI. The second made it clear the conversation has moved past that. LPs are asking direct questions, fee dynamics are shifting underneath the AI conversation whether anyone planned for that or not, and the confidentiality line between what can and can’t touch a general-purpose LLM is getting firmer, not softer.
The through-line from both meetings hasn’t changed, though: the firms that feel most at ease are the ones with real infrastructure and clear answers, not the ones with the most impressive prompt. If anything, last week’s session raised the stakes on that point. An LP asking how you use AI in a DDQ isn’t looking for enthusiasm. They’re looking for governance.
