The message still has to deserve its place
I would test an AI-generated sales message with one question: would it deserve to be sent if it were the only outbound touch the company got this week?
That test makes the account choice, evidence and proposed conversation visible. A message that mentions a buyer's name and a recent announcement can still waste their time. The seller must explain why the announcement creates a problem the company is equipped to help solve, and why the recipient is the right person to ask.
AI can make research and drafting cheaper. The commercial consequence depends on what the team does with that capacity. It can investigate fewer accounts properly, or send more messages that resemble investigation. The dashboard may count both as personalised outreach. The buyer experiences them differently.
My concern is the operating model behind volume: who decides that an account deserves contact, who checks the claim, and what the company learns when the recipient objects? Those decisions belong in the sales process before a campaign expands.
One finding worth taking seriously
In Gartner's survey of 632 B2B buyers, conducted in August and September 2024 and reported in June 2025, 73% said they actively avoided suppliers that sent irrelevant outreach. That is a reported preference in a particular survey. It does not measure the revenue lost by an individual campaign, and it does not establish that buyers reject every unsolicited approach.
It does establish a reason to look beyond replies. A message can produce no visible response while changing the recipient's view of the supplier. A sales team will rarely observe that effect directly. Treating silence as a harmless cost of doing business leaves an important uncertainty outside the evaluation.
I would therefore assess an AI prospecting investment against qualified conversations, account relevance and recipient feedback together. A cheaper send is useful only when it helps the commercial outcome after research, review and follow-up costs are included.
Personalisation begins before the copy
Consider a hypothetical industrial seller approaching a contractor about a publicly announced construction project.
The weak version says the seller noticed the project and offers a meeting about innovative solutions. The stronger version checks whether the project is at a stage where the seller's contribution matters. Has the relevant package already been awarded? Is the contractor responsible for procurement? Does the project require a material or service the seller actually provides? Is the announcement still current?
Those checks can eliminate the account from outreach. That is a useful result even though it reduces activity. If the account remains relevant, the message should make a modest, verifiable connection: a known project requirement, a capability the seller can substantiate and a proportionate next step. It should make uncertainty visible rather than imply knowledge of the buyer's private plans.
An agent may retrieve project records, compare sources and prepare a draft. A seller remains accountable for the contact decision. The Pipesignal project concept explores this research workflow. It illustrates a design, rather than demonstrated pipeline improvement.
What belongs on the scorecard
Reply rate is informative, but a negative reply and a request for a technical conversation are different outcomes. Meetings also need qualification: the relevant organisation, decision-maker and identifiable problem should be clear before a booking becomes evidence of pipeline.
I would record results by the signal that justified contact. A tender notice, an existing customer introduction and an inferred technology need should not be pooled into one average. Looking at a sample of the actual messages explains differences that a campaign total hides.
Complaints and opt-outs require separate measures and prompt operational responses. An opt-out tells the company to stop that contact. A spam complaint is a different signal with implications for sending reputation. Neither becomes a universal measure of brand erosion merely because it crosses a convenient percentage.
The useful comparison is with the organisation's own baseline, the recipient population and the provider's applicable rules. Include the cost of finding evidence, correcting bad research, reviewing drafts and handling replies. AI may reduce drafting time while increasing verification effort. The investment case should show the net change.
Delivery infrastructure is part of the responsibility
Google's sender guidelines distinguish requirements for senders to personal Gmail accounts and additional requirements for bulk senders. They cover authentication, recipient feedback and unsubscribe practices. Check the current rules for the actual sending environment rather than treating a summary as a complete policy.
Separating message streams can be a legitimate technical design choice. Creating another domain does not create recipient interest or permission, and rotating infrastructure is a poor substitute for fixing unwanted messages. Sales operations and the email administrator should agree the design, monitoring and response to delivery problems.
For an executive sponsor, the important point is that prospecting shares infrastructure and reputation with other business communication. Campaign decisions deserve an owner who can see those dependencies, including customer correspondence, rather than a volume target in isolation.
Change one sequence before buying more capacity
A sales leader under quota pressure is right to ask what will replace reduced activity. The answer needs to be operational: better research, clearer exclusions, something useful to offer and enough seller expertise to respond when a buyer engages. Simply asking people to send fewer emails can leave them with the same pipeline problem.
Choose one sequence and inspect a manageable sample with the sales manager. For each message, identify the source, the account-specific consequence, the support for the claim and the size of the ask. Compare that assessment with qualified conversations, negative feedback and delivery results. Decide whether to improve the research, change the audience, revise the proposition or stop the sequence.
If the constraint is upstream, use the AI Opportunity Map to compare account research, qualification and follow-up interventions. A working conversation is more useful with the sequence and its results than with an AI-SDR feature list.
Good outreach earns the next interaction. The leadership task is to design a process that can explain why this account, this evidence and this message belong together.