Half a revenue team does not use the AI it was given. That sentence gets read as a culture problem, so the fix people buy is another training session. Then the same half of the team keeps ignoring the tool.
The cause is narrower. The assistant can draft a follow up email. It cannot see the CRM record, the call recording and the invoice that make the email correct. So a careful person checks the old way, once, and goes back to it. Training teaches someone to use a tool. Access decides whether the tool is faster than not using it.
Anthropic makes this argument in Building an AI-native revenue organization, and the line worth taking seriously is that how much a sales organization gets out of an assistant depends on what it has built around it, meaning which systems the assistant can reach. That is a vendor making the connector first case in public. It is also the part most rollouts get wrong.
Phase one: set up before the pilot
Five decisions belong before anyone opens the tool.
One owner. A rollout with three owners has none. Name the person who reports the verdict, and give them the authority to say the pilot failed.
The connector list. Write down the systems a rep actually opens during account prep. Usually that is the CRM, email, the call recorder, a chat tool, and the finance system that says whether the account is even worth the call. If the assistant reaches only the CRM, the rep still crosses four systems to be sure, and the assistant is a fifth tab rather than a replacement for four.
IT and security. Decide what the assistant may reach and what it may do with it. CorpusIQ retrieves from external sources without writing back, so the answer is auditable and the source record stays the one that owns the number. State that boundary before the pilot, not after the first question about it.
Two or three metrics. Not twelve. Prep time per call, share of the team asking at least once a week, and one accuracy measure such as whether two systems report the same revenue for the same month.
Spend visibility. Someone must be able to see usage and cost per team before the pilot scales. A rollout with no spend view gets shut down by finance in month three regardless of results.
Phase two: pilot on one workflow
Pick the workflow with the most prep time per unit of outcome. For most revenue teams that is call or meeting prep: thirty minutes of gathering for fifteen minutes of conversation.
Run thirty days, one team, one workflow. Ask the same three questions at the start and at the end:
- How long did prep take for your last call?
- How many systems did you open to be sure?
- When two of them disagreed, which did you trust, and why?
The third question is the one that predicts adoption. A rep who cannot say which system wins will not hand the decision to an assistant. Give the assistant a stored rule for that, and the answer becomes stable: same question, same number, source attached.
Hold the pilot to the two or three metrics from phase one. A pilot judged on enthusiasm produces a rollout that dies quietly in month two.
Phase three: scale the rules, not the licences
Scaling is not buying more seats. It is making what one team learned available to the next team without rebuilding it.
The durable asset is the written rule: which source wins when two disagree, which definition of revenue the company means, which workflow is approved for this question. Save those once, and every team that connects inherits them. A new team then starts at the standard instead of at zero, and the same question returns the same answer across the assistants your people already use.
That is the difference between a rollout and a purchase. A purchase adds a tool. A rollout adds a standard the tool has to satisfy.
How to measure the return
Group the measurement three ways.
Efficiency. Time and cost removed from work you already do. Prep time per call, reporting hours, the number of exports someone builds by hand each month.
Expansion. Work you now do that you previously could not afford. Every account reviewed before a call instead of the top twenty. Every renewal checked against usage data instead of the loudest customer.
New capabilities. Questions that had no owner before, such as whether two systems agree on this month's revenue, answered in one place with the evidence attached.
Any rollout that reports only the first group is measuring the cheapest part of the return.
What stalls a rollout
Training instead of access. The tool is taught, the records stay out of reach, and the team correctly chooses the faster path.
A pilot with no verdict. No owner, no metric, no date. It ends by fade rather than decision.
Definitions left to the assistant. If the same question can produce three defensible answers, the team learns to distrust all three.
One connector, called done. Access to a single system makes the assistant a convenience beside the work, not part of it.
No spend view. The program is cancelled in finance before it reaches the teams that would have used it.
A short checklist
- Name the owner and the date of the verdict.
- List the systems a rep opens during prep.
- Agree what the assistant may reach and confirm it does not write back.
- Choose two or three metrics, one of them about agreement between systems.
- Run thirty days on one workflow with one team.
- Write down the rules the pilot produced, and hand them to the next team.
The tool was never the hard part. Access is. Decide it before the pilot, and the rest of the rollout is arithmetic.
Explore the connectors CorpusIQ supports, how the MCP endpoint works, and what a plan includes. Related reading: client reporting tools and ad budget pacing.
FAQ
Why does half a revenue team ignore the AI it already has?
Usually because the assistant cannot reach the records the work depends on. If the CRM note, the call recording and the invoice are outside its reach, checking the old way stays faster than asking. That is an access problem, not a training problem, and it does not fix itself with more enablement sessions.
What should be decided before an AI pilot starts?
Five things. Who owns the outcome. Which systems the assistant may read. What IT and security require. Which two or three metrics decide success. Who can see usage and spend. Deciding these after the pilot starts is how pilots end without a verdict.
How long should a revenue AI pilot run?
Thirty days on one workflow with one team is enough to produce a verdict. Shorter measures enthusiasm. Longer hides a weak workflow inside a long evaluation, and the team quietly returns to its old habits before you have a decision.
How do you measure the return on a revenue AI rollout?
In three groups. Efficiency, meaning time and cost removed from work you already do. Expansion, meaning work you now do that you previously could not afford. New capabilities, meaning questions answered that had no owner before.