AI in Partnering: What's Working Today, What Isn't, and What's Next

Typically, AI is presented in binary terms: either it's going to solve all your problems or it's coming to take your job. On the ground in alliance management, business development, and search and evaluation, the reality is messier and more interesting: usage is climbing fast, but most of it still only scratches the surface of what the technology can do.

Our webinar panel, "AI in Partnering: What's Working Today, What Isn't, and What's Next," dug into that gap. Louis Rinfret, PhD, founder and CEO of allianceboard, moderated, joined by Carlos Velez, PhD, Managing Partner at Lacerta Bio; Robert McBride, Director of Business Development and Alliance Management, Operations and Analytics at Exelixis, Inc.; and Martin Svorc, director of global business development, neurology, immunology & fertility at Merck KGaA, Darmstadt, Germany.

Trust is still the barrier, but there are signs it's eroding

allianceboard's 2026 survey of partnering professionals found that trust remains the biggest brake on adoption: 56 percent cite concerns about accuracy or reliability, 48 percent data privacy or security, and 40 percent uncertainty about how to best use AI.

The panel recognized the picture, with nuances. For Velez, privacy tops the list in practice: "If I receive confidential information, can I upload it into Copilot or whatever platform I'm using? ... It's the main reason why people hesitate." McBride pointed instead to the "oversale of AI": "People went through that 'Oh my God, it's the best thing ever!' phase, and then they tried it out and found that the thing they got really wasn't what they needed, or they didn't know how to work with it. Or they tried to do too much all at once, and got frustrated."

Svorc sees the trust gap closing:

"As people get more familiar with the tools and how to use them and how to properly prompt, they get more comfortable about the accuracy and reliability, on one hand, and on the other hand, the models are getting better." — Martin Svorc, Merck KGaA, Darmstadt, Germany

One less obvious barrier, per McBride, work that was never written down. When a process lives partly with "Susan over there" and partly with "John over there," no model can automate it well. "People are not writing down how they do their work, and when you don't do that, you can't expect the A.I. to read your mind."

Usage is up, but it's still "very 2025"

The adoption numbers are moving: 53 percent of partnering professionals now use AI every day in their work, up from 38 percent in 2025, and three in four use it at least weekly. Yet only 6 percent have built AI into their partnering processes. The rest is mostly individuals with chatbots: a repeat of the previous-era scenario of "one alliance, one spreadsheet," which does not scale.

What people use AI for tells the same story. Drafting emails or documents leads at 71 percent (up from 60 percent), followed by extracting or summarizing contracts and documents at 58 percent (up from 39 percent), preparing recommendations or presentations (54 percent), and preparing meeting agendas and summarizing minutes (53 percent); the latter two were newly measured this year. Use as a "thinking partner" for strategic options was the only use to decline, slipping to 36 percent (from 39 percent). A second wave of more advanced uses is only starting to gain ground: 19 percent use AI to evaluate partnership opportunities, 12 percent to forecast or predict outcomes, and just 11 percent to analyze alliance performance.

"When I see these data, it's a little bit disappointing. This is all very 2025. I mean, 2027 is coming up! Maybe it's an education issue, a training issue. Maybe it's a corporate policy issue. Maybe it's both. I'm not sure, but I hope that by the next survey, there are more people who are doing more interesting things than asking Copilot to draft an email." — Carlos Velez, Lacerta Bio

Rinfret tied the numbers to the broader debate over AI's return on investment, where massive spending has shown little measured return so far: "If you look at the uses of it, it's not surprising: a better email is great, but your company's not going to double its growth rate or innovate twice as fast because you're drafting emails faster. You need to think strategically about embedding it into your processes."

Still, AI is already earning its keep in partnering, according to the panelists, by:

• Scanning the news, every day. A simple check of primary news sources against the team's business development watch list, with updates to the team: new data here, a telling 10-K disclosure there.
• Navigating contract stacks. Sorting through amendments, restatements, and side letters accumulated over a 20-year alliance to determine what's current and pertinent. The best tools now point to both the section and the agreement and know when an amended-and-restated agreement replaced an earlier one.
• Supporting governance. Summarizing JSC and other governance meetings and highlighting decisions made, milestones, and upcoming inflection points.
• Producing first drafts within a legal playbook. Routine CDAs that fall within the playbook can be processed largely automatically, with anything unusual escalated to legal. Alliance managers prepare first drafts of amendments and side letters with legal AI tools, shrinking the initial workload before the lawyer's full review and sign-off.
• Onboarding new alliance managers into decades-old alliances by summarizing what has happened over the years: amendments, decisions, and context.
• Serving as a "sparring partner" for scenario planning. If Scenario A happens, how might the partner react? "The alliance manager may not have all the options in their mind," said Svorc. "This kind of brainstorming, scenario planning, is valuable."

As McBride put it, you don't need to do it all at once. "Starting small to build trust, has been kind of the key for us."

The firehose problem, and keeping humans in the loop

More output isn't automatically more capability. McBride described the "firehose effect" with a story: a seasoned researcher at one company used AI for an entire company-and-science workup, then found herself in a meeting with her boss's boss unable to recall what she had just "researched." "It didn't stick in her mind." The fix wasn't abandoning AI; it was redirecting it to point her to the sources so she could read them herself. Svorc named the underlying risk as cognitive offloading: "I don't think it's too healthy to let all the expertise and everything be done by the tool, and you just kind of pick at the end."

One thing the panelists insist on is having a human in the loop. "At the end of each of the workflows, or somewhere in the workflow, there is still a colleague who reviews the output, makes sure everything is correct, and is ultimately accountable for the output," said Svorc. "That helps with the team acceptance of AI."

Velez took it further: expertise is what makes AI powerful, not what it replaces.

"The irony of all this is that to fully leverage AI, you really need to be an expert in your discipline, because not only do you need discernment, but even the ability to write a really clear prompt. The fundamentals ... will still be there even as AI evolves, because at some point, somebody has to review it, somebody has to sign off on it. Nobody wants to stand in front of a board or a judge, saying, 'Well, Copilot told me it was a good idea, therefore I went with it.' That's never going to fly." — Carlos Velez, Lacerta Bio

One simple quality practice from the panel: run important outputs through a second model. "Have the second or third model just correct for it," said Svorc. "Use it as a judge."

Mind the legal line

For alliance professionals, the sharpest warning of the hour came from McBride, a former attorney who is no longer practicing, stressed working with your own Legal Department is key. Emphasizing, he was not giving legal advice, but giving the group things to think about with their Legal Department, including the simple fact that everything you type into an AI tool is discoverable. He pointed to a recent case in which an executive asked an AI how to avoid paying a contractual milestone; the full transcript surfaced in discovery, and the company ended up paying the milestone plus damages.

Three practical consequences for partnering teams:

• Your AI conversations carry no attorney-client privilege. "Nothing is going to replace a one-on-one conversation with an attorney when you have a question that is very sensitive."
• Be deliberate with meeting capture. For routine governance meetings, AI-generated minutes are a time-saver. For hard JSC conversations, it may be better to have the minutes taken by hand where human judgement can be applied. Always remember the "New York Times" rule applies, even where A.I. is involved.
• Use AI to navigate contracts, then verify. Read the underlying sections yourself, "but I would never, never replace your lawyer with it."

How to move up the curve

The panelists converged on a practical playbook for individuals and teams:

Start small and chunk the work. "Instead of the scary thing that the CEOs have been talking about, replacing people, we've been using it more to enhance people's jobs," said McBride. One agent, one task, one success at a time.

Pick the most painful use case and solve it together. "Try to identify the low-hanging fruits," said Svorc. "Look into the most painful day-to-day work that you have to do, and then address it, not by yourself, but with a group of colleagues who use AI. They will figure it out, life is easier, and the word spreads quickly, and you move away from summarizing or drafting an email to something that is really meaningful in your day-to-day job."

Share what works. A solved problem can be saved as a file, a prompt, or a skill and shared with the whole team in seconds, Velez noted. Everybody can use it, and everybody can improve on it: "It just compounds."

Train yourself, within clear corporate policy. The major AI vendors publish extensive free training and certifications. "The information is out there," said Velez. "You need training, whether it's self-training or corporate training, in order to advance your use of AI. But you can't separate that from the need for clear corporate policies on what you can and can't do, what's considered confidential, not confidential, within your own corporate context."

Get out of the bubble. "Don't get discouraged by the hype," urged Svorc. "Go talk to your colleagues over lunch, just check in with them: 'Are you using AI? How are you using it?' You may sometimes be surprised that people have some cool ideas that you would not expect. And also, go beyond pharma, beyond this pharma bubble and this partnering bubble."

Line up an executive sponsor, and don't force it. When rolling AI out to a group, expect roughly 10 percent enthusiasts, 10 percent who will fight it tooth and nail, and a persuadable 80 percent in the middle, said McBride. Build for the 80, and revisit what didn't work: "Check in six months if that thing didn't work, because it'll probably work now."

Treat failure as tuition. "The big one is, just start getting momentum," said McBride. "Just try and build an agent. You're going to screw it up. Cool, you'll learn. You're going to mess up your first prompt. Cool, you'll learn. Accept that. Be very open to failing so often that eventually you will be good at it." And build incrementally: "You're not trying to build the whole brain at once. You're trying to build one neuron and connect it to the next neuron, connect it to the next neuron, and eventually you get this big beautiful brain. But you've got to do it organically."

What's next: climbing the automation ladder

Where is all this heading? allianceboard frames the future of partnering and alliance management with five degrees of automation, which the panel discussed:

• Level 0: None. Everything is assembled by hand; humans do everything.
• Level 1: Assist. AI summarizes, extracts, and drafts; humans review the drafts.
• Level 2: Monitor. AI monitors and flags ("three topics raised at the last JSC, no decision after 45 days"); humans read and decide.
• Level 3: Act, with approval. AI prepares the action and routes it for sign-off; humans approve or edit.
• Level 4: Work across companies. Your agent and your partner's agent reconcile milestones and obligations on both sides; humans govern the mandate.

The panel's consensus: most teams are at level 1.

"When you get to level two, that's when you're really starting to use AI well, because now you're not just reviewing documents, you're actually integrating it into your workflow. ... If you're watching this, try to get to step number two, please." — Carlos Velez, Lacerta Bio

McBride sees level 3 within reach but drew a line at level 4, at least for this generation of professionals: "I'm not comfortable with four, but my kids might have to be."

Expect automation to advance unevenly, Rinfret noted: high-volume, lower-stakes work such as the initial triage of incoming partnering requests will automate faster than high-touch, high-stakes alliance decisions.

Two developments to watch:

• Interoperability standards. The Model Context Protocol (MCP) lets AI assistants work directly with applications and data sources, and Agent2Agent (A2A) lets agents communicate with each other: the plumbing that supports moving up the AI automation ladder. Both are now governed under the Linux Foundation, with the major AI vendors participating. "A real game changer," said Rinfret.
• Data foundations. AI is only as good as the information underneath it. Svorc pointed out that workflows lose power when work is scattered across internal clouds, CRMs, and five-year-old Excel files. McBride warned that most teams face a data-cleanup phase first: "you embrace the suck for a little bit in the beginning, and then your world going forward is that much easier." Or as Rinfret put it: "When everybody has their own formats and their own files, and the data is all over the place, it's very difficult for AI to be good. ... It relies on solid, standardized information to be reliable. This is what removes the trust barrier."

Ready to move up the ladder?

The lesson of the hour: results from AI depend less on which chatbot you use than on the foundation underneath it. Documents scattered across email, SharePoint, and other systems give AI something to read. They don't give your data the structure and granular access control that secure, reliable AI needs. That's what lets AI understand the current state of your partnering and act as a useful advisor rather than a confident but untrustworthy one.

allianceboard's ISO 27001-certified platform is used by leading partnering and alliance management teams. It brings search and evaluation, business development, and alliance management together on one platform, built on a data graph that connects your alliances, deals, obligations, and decisions and keeps them current, so AI can answer reliably, monitor and flag, and automate workflows instead of just summarizing files. And it works where you already work: from Claude, Copilot, and other assistants via MCP, or in allianceboard's own AI-powered interface.

That's how partnering teams scale: level 2 and beyond, on a purpose-built AI-powered foundation. Contact us for a personalized demo.

‍

‍

‍

Note: The panelists' opinions are their own and do not represent the views of their organizations.‍