In 2025, the Wall Street Journal reported that Johnson & Johnson had run close to 900 generative AI pilots, then deliberately cut back. The reason was not that the technology disappointed. It was that roughly 10 to 15% of those pilots accounted for about 80% of the value, and the company decided to stop spreading resources across the rest. Its CIO called the shift a maturation process.
That is the most useful number in enterprise AI right now, and almost nobody in life sciences is drawing the obvious conclusion from it. The pilots that survived were not the ones with the cleverest models. They were the ones sitting on data somebody had already done the unglamorous work of fixing.
We have spent 18 years building CRM systems for medical device companies, CROs, biotech firms and healthcare nonprofits. Here is what we see now: teams are being asked to evaluate AI agents on top of a CRM that cannot yet answer “which campaign produced this closed deal” without a spreadsheet and an afternoon. An agent will not fix that. It will inherit it, at speed, and with more confidence than the data deserves.
An AI-ready CRM is one where every record has a single trusted version, every lead carries its source through to revenue, and the compliance rules are enforced by the system rather than by whoever remembers them. That is the whole definition. It has nothing to do with which AI product you buy.
The reason it matters is mechanical rather than philosophical. An AI agent that summarises an account reads whatever records it finds. If a hospital system exists three times in your CRM under three spellings, the agent will confidently summarise one third of the relationship. If lead source is blank on 40% of your records, an agent asked to recommend where to spend next quarter is guessing with a straight face.
In our experience they stall for four reasons, and none of them are about the AI.
Regulated industries generate duplicates faster than most, because the same account arrives through conference badge scans, distributor lists, clinical contacts and inbound forms. When GammaTile came to us, conference leads were being uploaded manually after every event, in a different format each time. Automating the path from iCapture into HubSpot and Salesforce produced a 43% reduction in duplicate conference lead records. That number is not an AI result. It is the precondition for one.
Marketing lives in one system, sales lives in another, and the join between them is a weekly export. Allucent, a global CRO, could not follow a lead through the funnel at all before we built bi-directional sync between HubSpot and Salesforce. Once attribution surfaced across the full funnel, the team recognised 2.1x marketing ROI. Not because they spent more, but because they could finally see what the spend had already done.
HIPAA and consent rules that live in a policy document instead of in field-level permissions do not survive contact with an autonomous system. Anything that can read your CRM on behalf of a user needs the boundaries encoded, not documented.
A CRM designed for the reporting layer rather than the rep produces workarounds, and workarounds produce exactly the gaps an AI agent will later trip over. We wrote more about these failure patterns in why healthcare CRM integration fails.
In order, and before you evaluate a single AI product:
The Connecticut Children's Foundation engagement is the clearest illustration of the payoff. Unifying sales and marketing on an integrated CRM produced 1.5x fundraising growth in year one. No AI was involved. The gain came entirely from the organisation being able to see its own funnel.
We tell clients to wait more often than a consultancy is supposed to. If your CRM data is in the state described above, an AI agent pilot in the next quarter will most likely join the 85% that J&J deprioritised, and it will burn executive patience that you will want later, when the foundation is actually ready.
Six months of data and integration work is not a delay to the AI programme. Based on what J&J's own numbers show, it is the difference between being in the 15% that produces the value and the 85% that gets cut.
Yes. AI agents read your existing records; they do not repair them. Duplicates, missing lead source and unenforced consent rules become faster, more confident errors rather than fixed ones.
For a mid-sized life sciences organisation with two connected systems, the integration and data quality work is typically a matter of months, not weeks. The Allucent HubSpot and Salesforce integration ran roughly six months end to end, including discovery and dashboard design.
Neither, categorically. We hold deep certification in both and a bias toward neither. The answer depends on where your regulated processes live and where your revenue team actually works. That is what discovery is for. We covered the evaluation criteria in how to choose CRM consulting services.
That the same question, asked of the CRM today, requires a human to reconcile two systems before answering it.
If you cannot currently trace a closed deal back to the campaign that produced it without manual work, that is your first project, not an AI pilot. We scope that work honestly, fixed-price when it is well defined, and we will tell you if the answer is to wait six months.
Moblize.it has built CRM systems for healthcare and life sciences organisations from Series-A biotech to the F100 for 18 years, including 50+ integrations across tools like iWave, MailChimp, Apsona, Klaviyo and Epic. If you want a candid read on whether your CRM data would survive an AI agent, get in touch.