Cold-email reply rates ran near 7% in 2022. Three independent datasets put 2026 at roughly 3.4%. Belkins, Hunter.io and Instantly measured different corpora over different years and landed on the same trajectory: the channel's core number halved in three years. By ordinary logic, an industry built on a decaying channel should be shrinking.

The revenue line bends the other way. Cold-outreach software sells more every year, across a price band that runs from $39-a-month self-serve tools to "AI sales employees" at tens of thousands of dollars a year. In 2025, AI SDRs appeared in Bridge Group's sales-development benchmark for the first time.

There is no contradiction once you see what the business actually sells. The decay is the fuel.

The harder reaching strangers becomes, the more a tool that claims to restore meetings-per-dollar is worth, and the more buyers will pay for that claim. What follows is a teardown of the business: what it sells, where the money comes from, who captured it across twenty years, and what deserves attention through 2028.

A labor-arbitrage business

Every player in this market does the same job: automate the first-contact funnel for people who need meetings with strangers at a predictable cost. The software finds or receives a list of potential buyers, reaches out over email or LinkedIn, manages follow-ups and replies, and lately takes on a new task — judging when a conversation is ready to become business.

Three kinds of buyers pay for it: founders running their own sales, in-house SDR teams, and lead-gen agencies that resell meetings as a service.

The commercial promise has not changed in twenty years: replace part of a fully loaded $110,000–160,000 SDR seat with $500–1,800 a year of software. Bridge Group's 2025 benchmark anchors the human side of that trade. Median US SDR OTE is about $80,000; add benefits, tooling, recruiting, ramp time and management overhead, and the loaded cost lands at $110k–160k.

The meeting is the other anchor. An in-house qualified meeting costs a company roughly $850–900 all-in; agencies sell qualified meetings at $200–1,000 each. Software pricing in this category is always set against those two numbers. The ceiling was never set by competing software. It is set by the cost of a human.

Split by value chain, the category has five layers, and their state in 2026 could hardly differ more:

One pattern runs through everything: a profit pool forms at a layer, the next wave commoditizes it, and the pool migrates upward. Layers 1 through 3 were the profit pools of 2015–2023, and they are cheap or free now. Layer 4 is the pool being fought over today. The rest of this piece keeps returning to that line.

Twenty years, six waves

The category is about twenty years old and has renewed itself every two to four years. The whole timeline fits in one table:

A few patterns cut across the waves, and they explain more than any single wave's story does.

Every winner entered by giving away the previous wave's core capability. Outreach gave away email tracking, Apollo gave away data plus sequencing, Instantly gave away mailbox economics, Clay gave away data plumbing. The entry ticket is the prior wave's product, priced at zero.

Category naming runs two to three years ahead of category peak. "Sales engagement" got its G2 category in 2018 and peaked in 2021. "AI SDR" got its name in 2023, its funding peak in 2024, its bust in 2025. When a category has a name and a quadrant, the deflation is usually already underway.

Distribution innovation beats feature innovation. lemlist built in public, Instantly built an affiliate army, Clay rode the GTM-engineer movement, HeyReach posted milestone screenshots. When feature-identical rivals split the same market, the one with the newer distribution wins.

Waves that run on borrowed rails end when the rail owner changes the rules. LinkedIn closed its API in 2015. hiQ won the legal precedent in 2019 and the business died anyway. The sender rules of 2024 and 2025 ended the burn-and-churn era of sending.

A bootstrapped deflation wave follows every VC-inflated wave, and it wins on retention. Instantly, Smartlead and lemlist took almost no venture money in 2021–22, then compounded straight through the AI SDR shakeout that wiped out their funded neighbors.

Where demand comes from: harder channels, bigger wallets

Start with the decay itself. Cold-email reply rates: about 7.0% in 2022–23 (Belkins), 5.1% in 2024 (Pipeful), 4.5% in 2025 (Hunter.io, a 31-million-email sample), roughly 3.4% in 2026 (Instantly, 100M+ emails). The inflection is visible. The 2024 sender rules landed, and post-ChatGPT volume exploded: outreach volume is up about 67% while response rates fell about 23%. The average prospect now receives more than 40 sales touches a day.

The human denominator is shrinking too. SDR monthly meeting quotas are down 40% since 2018, quota attainment sits at a record-low 60%, ramp time is three months, and annual attrition runs at 40% (Bridge Group 2025, a 351-company sample).

People cost more, and they are harder to hire and harder to keep. Every point of channel decay raises the value of software that claims to restore meetings-per-dollar. That is the demand engine, and it is the counterintuitive part of this market.

Buyers stack in three tiers. At the bottom sit SDRs themselves. No census exists; the BLS buries them in a 1.4–1.5M "sales representatives" bucket, and LinkedIn-title proxies suggest 150–300k in the US and 400–700k globally. Soft numbers, and worth treating that way.

The middle tier is the agencies: the global B2B lead-gen services market runs about $3.3B in 2026, with hundreds of appointment-setting firms listed on Clutch. Agencies run dozens of client campaigns and large mailbox farms, which makes them the strongest buyers in the category. Nearly every recent breakout prices per workspace rather than per seat for exactly this reason.

The top tier is the self-serve long tail: Instantly reached roughly 24,500 customers at about $130 ARPU on a $47 entry price, proof that hundreds of thousands of accounts will keep paying for affordable tooling.

What each tier pays:

Calibrate expectations on penetration. AI SDRs first appeared in Bridge Group's 2025 benchmark at 1% of respondents. The $15k–60k "AI employee" band is headroom, not the center of mass; the center of mass is $39–109 a month, and the fastest-growing wallet is the agency workspace.

The ceiling sits on the buyer's side. Gartner's numbers have rep-free preference climbing from 61% of B2B buyers in 2025 to 67% in 2026, with 45% already using AI somewhere in their buying process. This market can expand only while automation outperforms its own decay. Buyers are voting, with their feet, for fewer, better-aimed, more human-verified touches.

The supply side: many mid-sized winners, no king

By product overlap, supply splits into four clusters.

Cluster A — AI SDRs and AI sales agents (Artisan, 11x, AiSDR, Qualified's Piper). They sell "an AI agent runs your outbound," they raise big and stay loud, and retention is where the pressure shows.

11x raised about $76M in total. In March 2025, TechCrunch reported it had showcased logos that were not customers, counted total contract value as ARR, and was losing early customers at a claimed 70–80% churn; the CEO left two months later. Artisan is famous for its "Stop Hiring Humans" billboards; at its $25M Series A it carried roughly $5M ARR.

The cluster's shared lesson is uncomfortable. Autonomy currently exists only in email, which is the channel buyers are most tired of. And pricing "like a hire" raises the retention bar to human levels. The survivors converge on two shapes: managed service (AiSDR drifting toward "GTM engineers as a service"), or warm inbound (Qualified's Piper, the best-reviewed product in the category at G2 4.9, $42k–68k/yr contracts, doing inbound conversion rather than cold outreach).

Cluster B — LinkedIn automation and multichannel sequencers (Expandi, HeyReach, Dripify). Same moves, same ban risk, $30–120 per account per month, and every one of them brands on safety architecture: dedicated residential IPs, human-like pacing, weekly caps. Expandi claims about $10M ARR. HeyReach prices per sender, went agency-first, and hit $19M ARR in October 2026 — twenty-nine months after launch, having been banned by LinkedIn twice along the way without visible damage to the business.

Nobody in this cluster runs autonomous AI conversations; AI stays at first-touch copy and human-confirmed drafts.

Cluster C — email-first outreach platforms (Instantly, Smartlead, lemlist). They monetize the economics of sending at scale, plus the deliverability infrastructure underneath it. Instantly is estimated at $38–40M ARR with a 4.8 on G2 across 4,000+ reviews. Smartlead runs near $20M, its white-label and API making it the plumbing for agencies. lemlist reports $34M, bootstrapped, having publicly turned down venture money.

All three depend on metering, and credits are the number-one complaint theme in all three review sections.

Cluster D — data and platform layers (Apollo, Clay, ZoomInfo, Outreach, Salesloft). Not the same product, but the same budget. Apollo packs data, sequencing and dialing into $49–149/user/mo, holds about 40,000 paying customers, and is estimated at $150M ARR, dragging the SMB entry price toward free.

Clay sells the orchestration layer (build any outbound workflow you want), claims $100M ARR, and was valued at $7.1B in its September 2026 Series D. ZoomInfo has stalled: about $310M of revenue in Q2 2026, up 1.2% year over year, now repositioning as the data substrate for agents and buying the orchestration startup DoubleO.ai in September.

This layer presses down on everyone else, because a point tool with no exclusive capability keeps getting compressed by the bundles.

Three observations about the structure. This category is structurally not winner-take-all: it sustains multiple $10M–40M ARR independents at once, and the LinkedIn line alone supports several $10M-scale vendors.

The middle is squeezed from both ends at once: Apollo's free tier pulls entry pricing toward zero from below, suites bundling agents absorb from above, and undifferentiated point tools become M&A inventory.

Pricing power follows layer position: layers 2 and 3 fight over credit structures at $30–100/mo, layer 4 prices "like a hire" and fights over retention, and the frontier overall is sliding from per-seat toward hybrid metering — Gartner expects roughly 40% of enterprise SaaS to be priced by usage, agent or outcome by 2030, and Bain puts about 65% of vendors as already hybrid.

On market size, the defensible numbers are these. Sales engagement, narrow scope: $4.2–4.5B in 2024–25, headed toward ~$9.8B by 2033 at roughly 10% CAGR. Sales intelligence: $4.5–5B in 2025, $7–9B by 2030. The "$15B AI SDR market" figure that circulates has weak provenance. I would not cite it.

Three taxes every player pays

The LinkedIn platform tax. Section 8.2 of LinkedIn's user agreement bans automated access, connection requests and messaging outright, and third-party tools have no licensed lane. Enforcement reality in 2025–26: weekly invite caps around 100 on free accounts and 150–200 on paid tiers; hitting the cap triggers a temporary restriction of about a week; and one small-sample agency test (n=50, a single source) measured a 23% account-restriction rate within 90 days.

The vendor playbook — dedicated residential IPs, account warm-up ramps, human-like pacing, weekly caps, stop-on-reply — is probabilistic risk reduction, not permission.

Three consequences follow. The risk lands on the customer's own account, which is why "safety architecture" became a marketing requirement. The channel's effective cost keeps rising, and scarcity protects LinkedIn's reply premium over email. And there is a hard ceiling on how agentic LinkedIn automation can visibly be.

The deliverability tax. Google and Yahoo's February 2024 bulk-sender rules (SPF, DKIM and DMARC alignment, spam-complaint rates under 0.3%, one-click unsubscribe) plus Microsoft's May 2025 follow-up turned deliverability into a permanent compliance layer, and then into a product line of its own. The large warm-up networks process 80–120M emails a day; deliverability testing tools sell at $49–599/mo.

The industry's answer was mailbox fragmentation: 3–5 mailboxes per domain, rotating secondary domains. The net effect is that scale sending consolidates into the infrastructure platforms while everyone else gets pushed toward low-volume, high-authenticity postures.

The saturation tax. Reply rates sit at 3–4.5% and keep drifting down, and AI makes the problem worse before it makes it better: generic outreach at zero marginal cost accelerates channel fatigue. Every layer-4 product is, at bottom, trying to buy reply rate back with relevance — sharper targeting, personalization that is actually personal, or actual conversation.

The decay curve, not AI itself, is the deepest demand driver of the current wave.

Who gets in, and who can't get out

The 2020–2026 entrant record is unusually legible: six breakouts and two instructive failures.

The two failures deserve close reading. 11x lost on the mismatch between price and retention: it sold "replacement of a human" on annual contracts before the product could reliably book meetings, and the $76M it raised bought narrative, not retention. Kennected was executed by the platform: the leading LinkedIn automation tool of 2020–21 died from a cease-and-desist that removed its only substrate, sold cheap in 2024, and rebranded.

Behind both, the cheap-LinkedIn-tool die-off ran quietly. After enforcement escalated in 2023–25, tools whose only asset was volume on a rented surface kept dying without obituaries.

Eight entry patterns fall out of the cases:

1. Winners enter through a pricing-unit inversion, not a feature story. Per-mailbox becomes unlimited, per-seat becomes white-label, per-contact becomes per-sender, enterprise becomes freemium. "The incumbent's pricing unit is wrong for you" is this category's proven opening pitch.
2. The wedge buyer is whoever the incumbent over-charges, which almost always means agencies. They buy in volume, feel per-seat pricing hardest, resell what they buy, and evangelize in public.
3. Distribution means product-embedded affiliates, an agency ecosystem, and a founder audience. Paid ads come last. The one entrant that inverted this order, Artisan, bought awareness and did not buy retention.
4. Speed from launch to the first $1M is a selection filter. Bootstrappers manufacture it with lifetime deals and prepaid annuals; funded entrants manufacture it with ad spend. Both routes punish slow starts.
5. Platform risk decides which shapes survive: owned infrastructure beats agency plumbing, agency plumbing beats diversified substrate, and all of them beat single-platform volume tools.
6. AI-native entry after 2023 is a different game with a different failure mode: retention, not awareness.
7. Bootstrapping is the dominant winning shape here, not the scrappy exception. Four flagship bootstraps — Instantly, lemlist, Smartlead, HeyReach — hold around $110M of combined ARR at near-zero venture money. The VC cohort (11x plus Artisan, $100M+ raised) reached comparable or smaller revenue with worse retention.
8. Reported metrics are a market of their own, and milestone transparency is cheap distribution. What separates a HeyReach from an 11x is whether the dashboard would survive an audit.

The one-sentence version: enter at the layer where the incumbent's pricing unit is weakest, sell to the buyer it over-charges, arm that buyer to resell you, compound on ground you own rather than rails you rent, and let retention rather than awareness set the pace of the story.

What Product Hunt says about the next generation

Product Hunt is more useful than G2 for reading where the new generation is heading. The incumbents' home turf is not PH: Instantly, Dripify, Expandi and HeyReach never mounted real launches there. HeyReach and Instantly were born on AppSumo lifetime deals; the rest grew on SEO and G2. PH activity itself, then, signals the new generation. The 2026 PH data gives three pieces of evidence.

First, the finding layer is changing its story. The top of PH's daily boards in 2026 belonged to "real-time web search replaces the static database." Lev8 (No. 1 product of the day on July 21, 578 upvotes) uses parallel agents to search the live web for people plus buying signals. Lessie (No. 2 on April 7, 414 upvotes) takes a natural-language description of your target audience and finds them directly, and open-sourced its core. Jesse (No. 3 on June 18) puts it bluntly: stop building Apollo and Clay lists, search the live internet.

All three attack the static-database model of Apollo and ZoomInfo, and none of them does execution.

That makes them an upstream variable for the current structure, not a direct replacement.

Second, autonomous conversation is spreading. SalesRobot's AI Appointment Setter already claims to take over conversations and book meetings. AiSDR's new flagship Ami (No. 1 product of the week on September 16, 543 upvotes) runs LinkedIn DMs, reply handling and booking, and backs its AI judgments with 17,700 historical campaigns and nearly 20,000 booked meetings.

Its founder used the launch post to concede that the AI SDR industry "over-promised, under-delivered, and users hate it." A second sweep added four more data points — B2B Rocket, Coldreach, Unify, Sellinger — and Sellinger's launch copy calls out HeyReach by name for stopping at the first message.

"Chats on its own" became a standard claim in 2026. What has not become standard is "when it must stop": stop on human reply, never auto-send on AI failure, gate forbidden phrases, hand off only when qualified. No incumbent and no newcomer has made that safety vocabulary a product principle yet. The window is narrowing, but it is open.

Third, prices are polarizing. Self-serve sits at $29–100/mo: Sparkle $29, Autobound $39, Overloop $69–99, Breakcold $99, Skylead $100. Fully managed AI SDRs sit at $500–3,000/mo: Reply.io's Jason at $500, Unify around $700, B2B Rocket $799–1,799, Coldreach $899–3,000, Nooks around $30k/yr. The middle of that range is empty.

The new generation all acquires with free credits — Jesse gives 1,000, Autobound gives 5 a day, Coldreach gives 250 — which turns "affordable and unmetered" into the scarce position inside the PH ecosystem. One more commoditization signal: beyond Lessie, AgentSDR open-sourced its entire site. The second curve, where tools get invoked by agents instead of humans, has started.

One reading caveat. PH ratings inflate badly; perfect 5.0 scores with 32 or 105 reviews are artifacts of launch upvote culture. Treat PH as a window onto narrative and tactics, and keep horizontal reputation checks on G2 and Trustpilot.

The next two or three years

Compress the outlook to three things, each backed by an evidence chain from earlier in this piece.

Where the judgment layer lands. Data and send infrastructure are commodities. Sequencing is a $39–109 checkbox. The open profit pool is layer 4. Platforms are absorbing it from above (Agentforce, Omni, DoubleO), infrastructure tools are bolting it on from below (Instantly's Reply Agent), and LinkedIn's native conversational surface has no owner yet. Whoever turns conversation plus qualification into retention first takes the next wave.

The AI SDR shakeout. Qualified went into Salesforce, DoubleO into ZoomInfo, Salesloft into Clari, 11x's retention collapsed, Artisan carries an overhang, AiSDR went managed-service. The market has priced all of this in early. Expect the AI-SDR version of the 2021 sequencer die-off in 2026–27, aimed at whoever prices like a hire and retains like a trial.

Meanwhile the bootstrapped infrastructure cohort keeps compounding. Instantly, Smartlead and HeyReach crossed the AI bust without venture money, and their retention-through-resale design is the most durable moat in the category right now. Once the platforms finish absorbing agents and need execution plumbing, those companies are the natural acquisitions.

Buyer avoidance and trust. 67% of buyers want rep-free purchasing and 45% already use AI in their buying process, so the category grows only while automation beats its own decay. After 11x, auditable metrics, published behavior contracts spelling out what the AI will never do, and safety vocabularies are converting skepticism into conversion. The category's next brand asset is verifiability.

The decay will continue. That is about the only thing in this industry that is safe to predict. It is bad news for your prospects' inboxes, and it is not bad news for this business. It has always grown on decay.

Data notes: private-company ARR figures are company claims or third-party tracker estimates and none are audited. The reply-rate series mixes corpora; the direction (halving since 2022) is consistent across independent datasets, so treat any single year's decimal as soft. SDR headcount is a LinkedIn-title proxy, not a census. LinkedIn enforcement figures come from small-sample tests and should be read directionally. Data as of October 2026.