Modern B2B sales tactics work when they trade volume for specificity and timing. The tactics that stopped working did not become unfashionable — they became free. AI made generic personalisation available to everyone at zero cost, which is precisely why it now signals nothing.
This is what changed on the buyer’s side, which tactics to retire, what replaced them, and the evidence for each. For the workflow that operationalises it, see sales prospecting with sales intelligence.
What actually changed?
Buyers adopted AI faster than sellers did, and it reshaped the research phase rather than the decision phase.
- 77% of B2B buying processes used AI in 2025, and 89% of buyers now treat generative AI as a primary self-guided information source.
- Yet interactions with the winning vendor barely moved — roughly 16 per person in 2025 against 17 in 2024.
- 62% of frequent AI users fact-check AI-generated claims against primary sources.
- 69% still turn to a sales rep to validate AI-generated insights before committing.
Read together: buyers research far more than they used to, trust what they find less, and still need a human before they commit — but a different kind of human. The rep who explains the product is redundant, because the buyer already read the site and asked an LLM about it. The rep who confirms, contextualises, and navigates the internal decision is not.
Meanwhile the outbound channel got noisier. Average B2B cold email reply rates now sit between 1% and 5%, down from roughly 7% two years ago. Nothing about buyers changed to cause that. Sending capacity changed.
Which tactics stopped working?
| Tactic | Why it used to work | Why it fails now |
|---|---|---|
| High-volume undifferentiated email | Inbox competition was low | Reply rates fell from ~7% to 1–5%; sending capacity is no longer scarce |
| Merge-field personalisation | Using a first name signalled effort | AI made it free, so it signals nothing |
| Scripted cold calls with no research | Reaching someone was the hard part | Reaching them is easy; being worth the call is not |
| “Just bumping this to the top of your inbox” | Persistence read as diligence | Reads as automation, because it usually is |
| Feature-led pitching | The rep was the information source | The buyer read the site and asked an LLM first |
| Single-threading one contact | Smaller buying committees | Six to ten stakeholders now decide without you present |
The common thread is that each tactic derived its power from scarcity. When effort was expensive, spending it was a credible signal. Automation removed the cost, and with it the signal.
What replaced them?
Timing instead of frequency
The most reliable modern tactic is not a better message, it is a better moment. Outreach triggered by something that actually happened — a funding round, a leadership change, a hiring surge, a technology removal — arrives when the problem is live rather than when your cadence says so.
This is the tactic that most teams claim to run and few actually measure. The test is simple: how many days pass between a signal firing and your first touch? If the answer is weeks, you are running a calendar cadence with signal decoration.
Stacked personalisation instead of merge fields
Personalisation now works in tiers, and only the upper tiers do anything:
| Tier | What it looks like | Effect |
|---|---|---|
| None | Same message to everyone | Baseline, declining |
| Merge field | First name, company name | Indistinguishable from none |
| Single signal | One specific event plus a relevant value proposition | 15–25% reply rates reported |
| Multi-signal stacked | Two or three signals plus behavioural context | 25–40% reply rates reported |
Treat those upper figures with care. They are vendor-reported, measured on tightly targeted low-volume sends, and they are not campaign averages — the broad benchmark remains around 3.4%. What the numbers do establish is the direction and size of the gap: specificity outperforms volume by an order of magnitude, and stacking two signals beats one.
The practical rule: if your “personalised” line could be sent to a hundred other companies with one word changed, it is a merge field wearing a costume.
The phone came back
The most counterintuitive shift of the past two years: as AI-generated email saturated inboxes, direct human contact regained value.
- 57% of C-level and VP-level executives say they prefer to be contacted by phone.
- 51% of B2B leads are still attributed to cold calling.
- 82% of buyers report accepting meetings that began with proactive outreach.
This is not a case for reinstating the dialler. Unresearched, scripted, high-volume calling is as dead as it was five years ago. What works is a small number of well-informed calls into accounts that a signal already qualified — precision, not persistence, and always inside a coordinated multichannel sequence rather than as a standalone activity.
The rep as validator, not informer
With 69% of buyers using a rep to validate what AI told them, the highest-value conversation is no longer a walkthrough. It is confirming what they already believe, correcting what they got wrong, and helping them sell the decision internally to the other five to nine people in the room.
That reframes discovery. Assume the buyer arrives informed and possibly misinformed. Ask what they have already concluded before explaining anything.
How do you retire a tactic without losing pipeline?
- Run the replacement in parallel on a carved-out segment rather than switching wholesale.
- Hold volume constant in the test so you are measuring the tactic, not the effort.
- Measure meetings per 100 contacts worked, not replies. Reply rate rewards curiosity; meetings reward fit.
- Give it a full cycle. Most B2B cadences need 14–21 days before the comparison means anything.
- Cut the loser properly. Half-retired tactics consume the capacity the new one needs.
What to change this quarter
- Start measuring signal-to-touch latency. You cannot improve a number nobody owns.
- Delete every “just following up” step and replace it with a new angle or a new stakeholder.
- Require one verifiable specific in every first touch — an event, not an adjective.
- Add a second stakeholder mid-sequence by default.
- Reintroduce phone into signal-qualified accounts only.
- Cut total send volume and reinvest the time in the accounts that survive the cut.
The one thing that did not change
Every tactic above depends on knowing something true about the account — the funding round, the hire, the stack change, the right five people. Specificity is not a writing skill. It is a data problem wearing a copywriting costume.
Which is why the unglamorous work still decides the outcome: where the data comes from, whether it is verified close enough to send time to actually arrive, and whether your sales intelligence surfaces the change while it still matters.
Frequently asked questions
Is cold calling dead?
No — unresearched volume calling is. Precision calling into signal-qualified accounts performs well, and 57% of C-level and VP-level buyers say they prefer phone contact. The channel recovered partly because AI-generated email saturated the alternative.
Why did cold email reply rates drop?
Supply, not demand. Automation removed the cost of sending, volume rose, and the average reply rate fell from around 7% to between 1% and 5%. Buyers did not become less interested; they became better defended.
Does AI personalisation still work?
AI applied to real signals works well. AI applied to generic templates does not, because the recipient has seen twenty of them this week. The differentiator is the input data, not the model.
How much personalisation is enough?
One verifiable specific that could only apply to that account. Two or three stacked signals perform better still. Anything that survives a find-and-replace is not personalisation.
Should we simply send less?
Send less to more carefully chosen accounts, and reinvest the reclaimed time in research and threading. Cutting volume alone just produces a smaller version of the same problem.
What should replace “emails sent” as a team metric?
Meetings per 100 contacts worked, stakeholders engaged per account, and signal-to-touch latency. All three measure selection and timing; activity counts measure neither.
The takeaway
The tactics that stopped working were all proxies for effort. Once software could fake the proxy, buyers stopped reading it as evidence of anything.
What replaced them cannot be faked as cheaply: knowing that something specific happened at a specific company, reaching the right people about it quickly, and having something useful to say when they pick up.
Request a demo to see the signals we would surface on your target accounts, or read how the workflow fits together in how ZenBee works as a B2B prospecting tool. More in the Prospecting & List Building hub.