Delete before you automate means removing the steps a process does not need before you use software or AI to speed up the steps that remain. In sales, it is the difference between using AI to send more cold touches and using it to need fewer touches in the first place. The AI ROI data published this autumn points the same way: the tools are cheap, and the return depends on how the work is designed.
This post walks through what that data says, the order of operations it suggests, and what it means for outbound.
AI is cheap. That is not the problem
In a September 2026 deep dive on AI ROI, Chamath Palihapitiya of Social Capital cites Ramp's AI Index, which tracks AI spend at more than 70,000 US businesses. The median company spends $12.50 per employee a month. Against an average employee cost of about $8,500 a month, AI only has to make that person about 0.15% more productive to pay for itself, which he puts at about three extra productive minutes a week.
So the bill is rarely what stops AI paying off. The harder question is what happens to the time it saves.
Why the gains do not show up
The same piece argues that AI mostly speeds up production, while most of a corporate employee's week goes to coordination: meetings, alignment and sign-offs. If the minutes AI hands back flow into another meeting, the return is zero.
It also describes a 2026 experiment with 515 startups. Every firm got the same AI tools, training and support, and half also saw how other firms reorganised their work around AI. For the typical startup, revenue barely moved in either group. Almost all of the gains came from the top 10% of firms, and the group that saw the reorganised workflows pulled ahead, enough to reach 1.9x the control group's revenue overall. The tools were identical. The difference was how the work was rebuilt around them.
The order of operations
The piece closes on an old idea. Bill Gates made the point in 1995 that automation applied to an efficient operation magnifies the efficiency, and applied to an inefficient one magnifies the inefficiency. It then quotes Elon Musk's five steps, in this order:
- Question every process and requirement.
- Delete what you can.
- Simplify what is left.
- Speed it up.
- Only then automate.
Most AI projects in sales start at step five.
What this means for outbound
Cold outbound is a long chain of steps: build a list, research each person, write a sequence, send it, chase the silence, and hope it reaches someone who can decide. An AI SDR speeds up every link in that chain. It does not ask whether the chain needs to be that long.
A warm introduction asks that question. When someone the buyer already trusts makes the recommendation, most of the chain disappears. Here is the same motion, step by step:
| Step in a cold outbound motion | Automate it | Delete it |
|---|---|---|
| Build a list of target contacts | Enrichment and scoring run on their own | Start from accounts where someone already knows the buyer |
| Research each prospect | AI writes a summary per contact | The introducer already knows the context and shares it |
| Write and send a multi-step sequence | An AI SDR writes and sends every touch | One note from someone the buyer trusts replaces most touches |
| Chase non-replies | Automatic follow-ups on day 3, 6 and 9 | Follow up with the introducer, not the buyer |
| Get past a gatekeeper to the decision maker | More channels and more volume | The relationship reaches the person who decides |
The automate column makes each step cheaper. The delete column removes it. That is the distinction the AI ROI data keeps pointing at.
What it looks like in practice
Two Boomerang AI customers show what happens when relationship-led pipeline is measured against everything else:
- 1mind compared lead sources over its last 120 days of closed opportunities. Nearbound deals, the ones that started with a relationship, closed at 3x the rate of outbound and qualified 2.5x as often. Read the 1mind story.
- Storylane works former users and buyers who changed jobs. It saw a 12% reply rate on Boomerang job-change leads and more than 20 opportunities through its BDR team. Read the Storylane story.
Your numbers will differ. Before you change anything, check whether nearbound is your best pipeline source using your own closed deals.
Apply the five steps to your pipeline this quarter
- Question. Split qualification and close rates by lead source. Which source actually turns meetings into revenue?
- Delete. For your top accounts, stop sequencing people you can reach through someone they know.
- Simplify. Agree one way to ask for an intro: who asks, through whom, and how often each person can be asked.
- Speed up. Put the paths into each account where reps already work, in the CRM and in Slack.
- Automate. Only now hand the repeatable parts to software: finding the path, drafting the ask, following up and logging the result.
Where AI still belongs
None of this argues against AI in sales. It argues for pointing it at the right steps. At Boomerang AI, Rudy, our AI referral agent, does the step five work: he finds who already knows the buyer across your executives, investors and advisors, customer champions and partners, drafts the ask for the person making it, chases when a connector goes quiet, and logs every step to the CRM. The step two work, deciding that a relationship beats a sequence, is yours.
For more on where AI SDRs fit, read the AI SDR backlash and AI in B2B sales: what works and what is hype.
More questions
Where should a sales team use AI first?
After deleting steps, use AI on the work that remains and repeats: finding who already knows the buyer, drafting the ask for the person who will send it, following up when someone goes quiet, and logging the outcome to the CRM.
Frequently asked questions
What does delete before you automate mean?
It means removing steps from a process before using software or AI to speed up the steps that are left. If a step should not exist, automating it only makes the waste faster. In sales, it means asking which touches a deal actually needs before buying tools that send more of them.
Does AI pay for itself in a sales team?
On cost alone it usually does. Ramp's AI Index, as reported by Social Capital, puts median AI spend at $12.50 per employee a month, so AI only needs to make someone about 0.15% more productive to cover it. Whether the time it saves turns into revenue depends on how the work is designed.
Are AI SDRs a bad idea?
Not always. They are useful for reaching a long tail of accounts at low cost. The risk is using them to speed up a process that was not working, such as sending more cold touches to senior buyers who do not reply to strangers. Fix the process first, then decide what to automate.
.png)
