Hlib Storchak · 2026-08-23 · ~11 min readQuick answer
Gartner predicts that by 2028, AI agents will handle 90% of all B2B purchases, moving more than $15 trillion in spend through automated buying systems. That's a prediction about procurement, not a claim that outbound sellers are being replaced. The practical read for a seller: some of your target accounts will eventually run early vendor screening through a software agent before a human opens your email, and Gartner itself says this depends on "verifiable data feeds and standardized trust frameworks" that mostly don't exist yet. The fix isn't a new sales tool. It's making your pricing, comparisons, and specs the kind of content a machine, and a human, can actually parse without a call.
My take, up front
I'm Hlib Storchak. I build outbound systems for B2B founders and sales teams, and I've booked 2000+ meetings for B2B clients doing it. Every few months a headline number lands that's big enough to make a founder message me asking whether they need to rethink their whole outbound motion. "$15 trillion in B2B purchases run by AI agents" is exactly that kind of number, so this is me actually reading past the headline before reacting to it.
Short version: the prediction is real, it's sourced to a named Gartner analyst at a named event, and it comes with a real condition attached that most B2B categories haven't met yet. It's worth planning for. It is not worth panicking over, and it's definitely not a reason to rip up a cold email sequence that's still booking meetings today.
What Gartner actually predicted, and where it came from
At Gartner's IT Symposium/Xpo in October 2025, VP Distinguished Analyst Daryl Plummer told the audience that "90% of all B2B purchases will be handled by AI agents within three years, channeling more than $15 trillion in spending through automated exchanges." Gartner's own newsroom page returns a 403 to a direct fetch, so I checked the quote against two independent trade write-ups, Digital Commerce 360 and MarketScale, and both cite the same event, the same analyst, and the same 90%/$15 trillion figures by 2028.
The part that gets dropped when this number gets repeated around the internet is the condition Gartner attached to it. Per both write-ups, Gartner said this shift depends on "verifiable data feeds and standardized trust frameworks" that let a purchasing agent research, compare, negotiate, and execute a transaction with minimal human review. Those frameworks are, at best, early. That's not a reason to dismiss the prediction. It's the single most important sentence in it.
Read the caveat, not just the number. "90% of B2B purchases by 2028" is a prediction conditioned on trust infrastructure that doesn't broadly exist yet. Treat 2028 as the earliest plausible date for the categories that get the infrastructure first, not a deadline for your whole pipeline.
Buying agents and selling agents are not the same prediction
Most 2026 content about "AI and B2B sales" treats it as one blob: agents are coming, on both sides, all at once. Gartner's $15 trillion prediction is specifically about the buy side, procurement and buying-committee agents that research vendors, compare options, and in some cases execute the purchase. That is a different mechanism, a different owner, and almost certainly a different timeline than an AI SDR sending your cold outreach.
Keeping those two separate matters because they invite different reactions. A prediction about AI SDRs flooding your inbox is a prediction about your outreach getting more crowded. A prediction about buying agents is a prediction about who, or what, is reading your outreach and your website once it lands. You can't fix the second problem by writing a better subject line.
Who's actually doing the buying, and who still isn't
It's worth grounding this in where the human side of B2B selling actually stands today, because the pace of that side is a decent proxy for how fast the infrastructure Gartner's prediction depends on will actually mature. The Bridge Group's most recent SDR benchmarking report, its 10th edition, published February 2025 and based on 351 B2B companies, found the average SDR ramp time down to 3.0 months, the lowest on record, while quota attainment sits at 60%, also the lowest on record. Even on the seller's own side of the table, where AI SDR adoption is already well underway, a fully unattended, high-confidence process still isn't the norm.
That matters for reading the buy-side prediction honestly. If the sell side, with years of AI SDR tooling already in market, still runs on a mostly human process with real ramp time and real attainment gaps, it's reasonable to expect the buy side needs at least a comparable runway before agents are executing $15 trillion in purchases with minimal oversight. Gartner's own three-year window and its trust-framework caveat both point the same direction: soon, not now, and unevenly across categories.
Where this hits outbound first, ranked
Not every part of a B2B buying process is equally exposed to this shift, and not every part of outbound needs to react at the same speed. Here's how I'd rank it, from where I'd actually spend attention now to where I'd genuinely wait.
1. Vendor comparison and review-style content. Verdict: matters now, and it's cheap to fix. If a purchasing agent is scanning the web for who does what, an unclear or gated comparison page is invisible to it the same way it's frustrating to a human.
2. Structured pricing and plan pages. Verdict: matters now. A pricing page that requires a form fill to see a number is a page an agent can't compare, and a human buyer increasingly won't wait for either.
3. Heavily procurement-driven categories, security, compliance, dev tooling. Verdict: matters soonest. These are the categories most likely to get verifiable data feeds and standardized specs first, since they already run on structured RFPs and vendor questionnaires.
4. Cold email copy and subject lines. Verdict: mostly unaffected near-term. Nothing in this prediction changes what makes a first-touch email earn a reply from a human decision-maker today.
5. LinkedIn relationship-building and warm intros. Verdict: unaffected, arguably more valuable. Gartner's own separate research on buyer preference for human interaction in complex, high-stakes deals points the same way. The harder a deal, the less this prediction touches it soon.
6. CRM data and contract terms in machine-readable form. Verdict: matters, but it's 2027 to 2028 territory for most mid-market sellers, not a this-quarter fire drill.
Buying agent vs AI SDR vs human buyer, side by side
| Who or what | What it actually does | Timeline | What a seller should do about it |
|---|---|---|---|
| Buying agent (Gartner's prediction) | Researches, compares, and in some categories negotiates and executes a purchase for the buyer's org | Gartner: 90% of purchases, $15T in spend, by 2028, conditioned on trust frameworks maturing | Make pricing, specs, and comparisons legible without a form or a call |
| AI SDR (seller's side) | Drafts and sends outreach, qualifies replies, books meetings | Already in production across 2026, no trust-framework prerequisite | Keep a human owning strategy and the replies that don't fit a pattern |
| Human buying committee | Weighs risk, negotiates, makes the final call on anything complex or expensive | Present now, and per Gartner's own separate research, buyers want more of this, not less, once a deal gets high-stakes | Protect the human moment on your highest-value accounts regardless of what automates around it |
What actually changes in how you write for prospects
This is the five-minute check I now run before a client's site or pricing page goes live: could someone, or something, actually lift the pricing model and the honest difference-versus-competitor claim off the page without talking to a human first? If the answer is no, because the number is gated behind a form or the comparison only exists as a sales rep's talking point, that's the actual gap this prediction points at, not a reason to buy a new outbound tool.
In practice that means plain declarative sentences instead of marketing copy for anything a buyer, human or agent, needs to compare: what it costs, what it includes, what it doesn't, and how it differs from the two or three names you get compared against most. None of that requires new AI infrastructure on your end. It requires writing the honest version of your pricing page down instead of keeping it in a rep's head.
What doesn't change, and why I'm not rebuilding my funnel over this
2028 is more than two years out, and it's a prediction conditioned on infrastructure that doesn't exist broadly yet. Meanwhile the sell side, where AI SDR tooling has had years to mature, still runs on a 3.0-month ramp and 60% quota attainment at 12 months per Bridge Group's own most recent numbers. If the more mature half of this equation still leans on humans this much, betting a whole strategy on the buy side automating faster is premature.
So I'm not changing what earns a reply in a cold email, and I'm not changing how I run a LinkedIn sequence. Both still depend on the same thing they did last quarter: a sharp list and a real reason to reach out. What I am changing is smaller and cheaper, making the public-facing content a buyer or an agent hits after that first reply actually answer the questions it's there to answer.
A checklist for pressure-testing any "$X trillion" prediction
Before you act on a headline stat like this one, or repeat it to a client or a boss, run it through four questions. Does it have a named analyst and a named event attached, not just "Gartner says"? Does it carry a stated condition, and is that condition already true today or is it the whole prediction? Is the number describing a share of activity or a share of actual dollars and decisions, since those get conflated constantly? And would acting on it this week cost you more than waiting one more quarter to see whether the trust infrastructure it depends on actually shows up?
This one clears the first two bars honestly, a named analyst, a named event, and a disclosed condition, which is more than most of the recycled stats I end up tracing back to an unsourced blog post. That's exactly why it's worth planning for. It's also exactly why the plan is "clean up your pricing page this quarter," not "rebuild your whole GTM stack this month."
What it costs to make your content agent-legible: a simple model
Here's a rough model, built on stated assumptions you should swap for your own. Assume a freelance writer or ops hire at €60 to €90 an hour, five pages that actually matter for comparison, a pricing page, two competitor-comparison pages, and two spec or feature pages, and four to six hours per page for a plain-language rewrite plus basic schema markup. That's 20 to 30 hours total, or roughly €1,200 to €2,700 as a one-time project, plus a quarterly two-hour audit going forward to keep the numbers current. If your team already owns the copy and just needs the schema and structure added, cut the estimate by roughly half.
That range is intentionally wide because your real cost depends on how many pages actually get compared before a prospect talks to a human, and how far your current pricing page is from a plain, gated-free answer. Run the actual page count and hourly rate you have, not mine, before you budget it.
What I'd actually do this quarter
Audit your five highest-traffic, most-compared pages and ask honestly whether a stranger could read a real answer to "what does this cost and how is it different" without filling out a form. Fix those five before anything else. Keep your cold email and LinkedIn motion exactly as sharp as it needs to be on its own merits, since nothing in this prediction changes what earns a human reply today. And put a date on your calendar, one or two quarters out, to revisit this specific prediction and check whether the trust-framework condition Gartner attached to it has actually started showing up in your industry, instead of assuming the countdown started the day the headline ran.
Key takeaways
- Gartner analyst Daryl Plummer predicts AI agents will handle 90% of B2B purchases, moving over $15 trillion in spend, by 2028, a prediction conditioned on "verifiable data feeds and standardized trust frameworks" that mostly don't exist yet.
- This is a buy-side prediction about procurement agents, a different mechanism from AI SDRs on the sell side, and the two shouldn't be treated as one trend.
- The Bridge Group's most recent SDR data (3.0-month ramp, 60% quota attainment at 12 months, both lowest on record) shows even the more mature sell-side automation still leans heavily on humans, a reasonable proxy for how far off full buy-side automation likely is.
- The practical fix now is cheap: make pricing, comparisons, and specs legible without a form or a call, not a new sales tool.
- Cold email and LinkedIn outreach quality is unaffected near-term. Protect the human moment on your highest-stakes accounts regardless of what automates around it.
FAQ
Does Gartner's $15 trillion prediction mean AI will replace B2B sellers?
No. It's a prediction about the buy side, procurement and buying-committee agents that research and compare vendors, not a claim that human sellers or outbound teams are being replaced. It's also conditioned on trust infrastructure Gartner itself says mostly doesn't exist yet.
What's the difference between an AI SDR and the "buying agent" Gartner describes?
An AI SDR is deployed by the seller to send outreach and book meetings, and it's already in production today. A buying agent is deployed by the buyer's own organization to research, compare, and sometimes execute a purchase. They're on opposite sides of the table and moving on different timelines.
Should I change my cold email or LinkedIn copy because of this prediction?
Not because of this specifically. Nothing in Gartner's prediction changes what earns a reply from a human decision-maker today. What's worth changing is your public-facing pricing and comparison content, since that's the part a future buying agent, and a present-day human buyer, actually has to parse.
How much does it cost to make a website "agent legible"?
Using stated assumptions of €60 to €90 an hour and five key pages needing four to six hours each for a plain-language rewrite plus schema markup, expect roughly 20 to 30 hours, or €1,200 to €2,700, as a one-time project. Swap in your own page count and rate before budgeting it.
When should I actually revisit this prediction?
Put a check on the calendar one or two quarters out and look for whether the "verifiable data feeds and standardized trust frameworks" Gartner named as a prerequisite have actually started appearing in your industry, rather than assuming the 2028 countdown applies evenly to every category today.