Most teams bolt a chatbot onto the pricing page and call the funnel agentic. We have watched that movie. It ends with a bounce and a Slack screenshot.
A $14M ARR client sent us their "agent rollout" last spring. Six embeds. One job each, they said. In practice, four of them asked the same qualifying questions, one booked meetings into a calendar no one on the team checked, and one offered a demo to people already mid-trial. The stack cost more than the intern it replaced and converted worse.
That is the usual pattern. CRMs became spreadsheets. Automation became a graveyard of unopened sequences. Chatbots answered questions visitors never typed. Now every vendor is selling an "agent" and the brief is the same: drop it somewhere in the funnel and wait for pipeline.
Agents can do useful work. They can also spray your brand across a visitor's evening like a drunk SDR. The difference is placement. One job. One stage. One owner.
Short answer (2026)
Role-based agents belong at specific stages, not on every page. For most B2B SaaS teams we work with, start at Consideration with one SDR-style agent that books or qualifies. Awareness is a close second if you have real traffic. Full six-stage orchestration is a $5,000+/mo project and rarely worth it below about 10,000 qualified visitors a month.
Profound is an AI-search visibility platform (marketing agents for answer engines). It is not Drift, Qualified, or 11x. Those sit on your site and talk to visitors. Profound sits in the answer-engine layer: prompts, citations, brand presence in ChatGPT and Perplexity. Treat it as an Awareness / content-ops agent, not as a chat widget.
| Tool | Layer | First job | Do not use for |
|---|---|---|---|
| Profound | Answer engines | See which prompts cite you | On-site demo chat |
| Drift / Qualified | Your domain | Route hot pricing traffic | AI-answer measurement |
| 11x / Artisan | Outbound + inbound SDR | Volume outreach | Citation analytics |
| Custom GPT + Zapier | One stage you pick | Prove the job exists | Six-stage orchestration |
Related: Profound bake-off · Citation Index
Platforms by funnel stage
| Platform | Best stage | What it actually does | 2026 pricing (check current) | Best for |
|---|---|---|---|---|
| Profound | Awareness (answer engines) | Visibility and agents for AI-generated answers, not on-site chat | Custom. Confirm on their site. | Teams already publishing decision-stage pages who need to see how they show up in AI answers |
| Salesloft Drift | Consideration | On-site conversational routing to SDRs | Often quoted around $2,500/mo | Mid-market SaaS with an SDR desk that needs chat handoff |
| Qualified | Consideration, Decision | High-intent chat and booking on pricing / demo pages | Custom, often $1,500–4,000/mo | Pricing-page traffic that should reach a human fast |
| 11x.ai | Awareness, Consideration | Autonomous inbound and outbound SDR work | Often $3,000–6,000/mo | Teams with little SDR coverage who want both inbound and outbound covered |
| Artisan (Ava) | Awareness, Consideration | Outbound SDR automation | Often $2,000–5,000/mo | Outbound-heavy teams testing AI-written sequences at volume |
| Custom GPT on Zapier / Make | Any single stage | One narrow job on tools you already pay for | $50–300/mo | First experiment. One stage. One owner. |
Vendor list prices move. Re-check before you sign.
Section 01 / Architecture
Your funnel is no longer a leaky pipe. It is sentient.
What changed is persistence. A chatbot answers one question and dies. A useful agent keeps a job: qualify, book, chase a stalled setup, notice a usage spike. Think of a junior hire with a single KPI and no LinkedIn tab.
The question we get on calls is always the same: where do we put the first one?
Start where a missed reply costs you a deal this month. That is almost never the blog homepage.
“SaaS dynamic checklists raised average time-on-site instantly.”
Awareness: stop greeting every visitor
Most top-of-funnel traffic on B2B sites is tyre-kicking. Bounce rates that would embarrass a brochure. CTAs ignored with the warmth of a Parisian waiter on a Monday.
An Awareness agent earns its keep if it does two jobs well. Route the right visitor to a comparison or calculator. Leave the rest alone. We had a client with 50-odd feature pages. The useful version of this was an agent that offered a checklist only after someone hit two comparison URLs in one session. Time-on-site went up. So did the quality of the demo form, which is the number that mattered.
Where it lives:
- Content hubs and high-traffic landing pages, after a signal (scroll, second page, return visit)
- Inline in a long guide, as a chooser ("Need a CRM shortlist? Ask.")
- Tied to Google Ads only if you can pass UTM and landing-page context without leaking the whole CRM
If you are buying Profound, this is the stage it actually belongs to. Not as a chat bubble. As the layer that tells you which prompts mention you, which pages get cited, and which agents should push that content into more answer-engine surfaces. Mixing that job with on-site SDR chat is how you get two mediocre tools and one confused RevOps lead.
Where Profound actually fits
Profound's own site positions it as marketing agents for visibility in AI search: prompt volumes, answer-engine insights, agents across channels. That is Awareness work in the 2026 sense. Buyers ask ChatGPT and Perplexity before they hit your pricing page. Profound is built for that layer.
Do not buy it as a Drift replacement. Drift and Qualified sit on high-intent pages and route humans. Profound measures and pushes presence in generated answers. Complementary, if you have budget for both. Substitutes, if your brief was "we need a chatbot that books meetings."
A first Profound-shaped deployment we would actually recommend:
- See the prompts. Export which buyer questions already mention you or your category. Pair that with the pages those answers cite.
- Fix the cited pages first. Comparison, alternatives, pricing. If the agent cannot point at a clean source, you are paying to amplify a muddle.
- Then add an on-site agent at Consideration, using Drift, Qualified, 11x, or a $100 Zapier bot, depending on traffic.
We would not put Profound on the pricing widget. We would not wait to build a standalone "Profound vs X" page until the existing funnel pillar has been given 60 days to absorb those queries. If share has not moved, then consider the extra URL.
Is Profound a chatbot?
No. Chatbots wait for a typed question on your site. Profound is sold as agents plus analytics for AI-generated answers.
How is it different from Drift or Qualified?
Those two live on your domain and hand hot visitors to SDRs. Profound lives in the answer-engine layer. Use both only if you have traffic and a content engine worth measuring.
What traffic volume justifies an on-site SDR agent?
Below about 5,000 qualified visitors a month, a custom GPT on Zapier is the cheaper test. Above 10,000, a specialist platform usually pays for itself inside a quarter if you give it one job.
Does any of this talk to HubSpot?
Most of the on-site tools do. Profound lists CRM-style workflows; confirm the current HubSpot CRM connector on their site before you promise RevOps a native sync.
Isolate Minimum Viable Curiosity.
Reject unqualified leads directly. Genuine interest stems from honest boundaries, not deceptive form walls.
Interest: preheat, do not spray
Here people are squinting. Overpriced vapourware, or actually useful. You do not need more MQLs. You need a reason to continue.
An Interest agent should notice a signal and offer one next step. "Is this for your team or just you?" A 90-second clip. A comparison they have not opened. Gate nothing with a form if you can trade a useful answer instead.
Where it lives:
- Mid-scroll on case studies and feature comparisons, after they have read something
- In reply to a cold email click that never converted. One nudge. Then stop.
- As an overlay only if it can shut up. Rage-inducing popups are not agents. They are popups.
Train it to decline bad fit. A serious buyer relaxes when the system says you might be the wrong vendor. We have seen that line do more for trust than another case-study carousel.
Consideration: this is where most funnels go to die
This is the stage teams abandon to prayer and a Calendly link. It is also the only stage I would start with if you can only fund one agent.
Give it the SDR job and nothing else. Book the call. Pull a one-page business case that names the competitor they already mentioned. Answer pricing, security, and integration objections from a maintained pack, not from last quarter's deck.
Where it lives:
- Pricing pages and integration lists
- Inside the "Get a demo" click, qualifying before an AE's Tuesday disappears
- In the inbox as a copilot, if your reps will actually read the summary
After the chat, have it drop a shared doc: what was asked, what was promised, what happens next. Buyers hate taking notes. Reps hate rewriting them. One of our mid-market clients started doing this by hand in 2024. The agent version is the same habit with less typing.
Decision: timing beats another discount
You are at the cliff. Most funnels shove. A Decision agent should remove the last stupid blocker.
Proposal summary from CRM fields. SLA and security pack on the second ask, not the seventh. A ping when legal is stuck, aimed at the person who can unstick it, sometimes the buyer's boss.
Where it lives:
- Deal-room pages and proposal portals
- Next to DocuSign or PandaDoc, answering questions before signature theatre starts
- A Slack or Teams ping to a human when the deal actually needs a human
One startup we advised bolted a GPT sidekick into proposal PDFs. Close rate moved 22 percent, mostly because legal questions died inside the file instead of a seven-email thread. That number is from a single account. Treat it as a hint, not a benchmark.
Convert fresh buyers into true believers.
Activation: buyers still get lost
Post-sale agents belong here more than vendors admit. Customers wander through onboarding like tourists without a map. Your CS team is in a Zoom about a renewal.
An Activation agent should walk setup by role and stack, offer help when an integration stalls, and collect the first week of friction while a human can still fix the product.
Where it lives:
- In the product, on the setup checklist
- As a Slack or Teams bot for admins
- In email only if it sounds like a person with a name and a stop button
If setup dies, book a 15-minute call. No form. We have watched more logos churn on an unfinished Salesforce sync than on price.
Expansion: this is CS work, not a growth hack
You have the account. Now you want the expansion, the referral, the renewal. An Expansion agent watches usage, spots a drop or a ceiling, and asks a specific question. "Your team hit 85 percent of the seat cap. Want a 20-minute look at the next tier?" beats a newsletter called Enablement Insights.
Where it lives:
- Admin dashboards
- Account Slack channels
- CS email footers that actually do something
A PLG company we know claimed a 32 percent lift in upsell clicks after usage-triggered nudges. Again: one company, their number. The useful lesson was narrower. The message named a limit the user could see in the UI. Vague "you might like this add-on" notes were ignored.
The map, for people who think in tables
| Funnel stage | Agent role | Typical actions | Where it lives |
|---|---|---|---|
| Awareness | Discovery / answer-engine | Route traffic, or measure and improve AI citations | Blog, ads, landing pages, Profound-class tools |
| Interest | Warm-up | One contextual question, one next asset | Case studies, mid-funnel email |
| Consideration | SDR | Book, handle objections, write the recap | Pricing, demo requests |
| Decision | Deal | Unblock legal, surface the pack, ping a human | Deal room, proposal docs |
| Activation | Onboarding | Guide setup, chase the stall, alert CS | In-product, Slack, email |
| Expansion | CS | Watch usage, prompt a specific upgrade, book the QBR | Dashboards, account threads |
Why most of these experiments flop
Agents fail for the same reasons sales interns fail.
- No access. If it cannot see CRM, product, or the FAQ, it is guessing.
- Too many jobs. Lead gen plus scheduling plus close is not a role. It is a mess.
- Wrong voice. Talking to a CFO like a developer. Watch the pipeline go quiet.
- No owner. An agent without a human accountable for its replies is a brand risk with an API key.
One agent, one job, one stage. Specialists. If you need a Swiss Army knife, hire a person.
What this actually costs
You already pay for most of the ingredients: CRM, a chatbot licence you forgot about, Zapier, model tokens. The missing piece is orchestration and someone who will read the transcripts.
- $50–100/mo: one GPT on ChatGPT, Zapier, or Make
- $500–1,500/mo: an embedded agent wired to CRM and analytics
- $5,000+/mo: multi-agent setup with retrieval, custom APIs, and weekly tuning
Against a loaded SDR or CS hire in the US, the lower two tiers are pocket change. You are not firing the team. You are stopping them drowning in work a script should have handled in 2019.
Funnels were a diagram. Agents ignore the diagram
The old picture was a cone. The useful picture is a set of loops. Agents re-enter, escalate, go quiet, come back at 11pm on a Thursday while your AE is on PTO. That is the point, and the risk.
Start with one stage. Give the thing a job you can score next month. Watch the transcripts. Then copy what worked.
This is not a 2028 slide. It is this quarter's missed meetings if you keep adding widgets and never pick a stage.
If you want a second pair of eyes on where the first agent should sit, the B2B SaaS work is the door. Bring traffic numbers. Leave the six-agent architecture at home.
FAQ
1. What is an agentic funnel in B2B marketing?
A funnel where a named agent has a single job at a named stage: qualify, book, onboard, expand. The rest of the funnel stays human.
2. How is that different from a chatbot?
A chatbot answers the last sentence. An agent is allowed to act: book, route, chase, stop.
3. Where should the first agent sit?
Consideration, on pricing or demo, unless your problem is answer-engine invisibility. Then start with measurement, not a widget.
4. Can an agent qualify before a sales call?
Yes, if it can see fit, budget signals, and urgency, and if a human still owns the weird cases.
5. Does this work for product-led growth?
Activation and expansion are the natural seats. Usage is the signal. A chatbot on the homepage is not a PLG motion.
6. Do I need a custom build?
Not for the first test. ChatGPT plus Zapier plus the CRM you already have will tell you if the job is real.
7. How do agents personalise anything?
They read CRM fields, page path, and prior replies. If those are empty, the "personalisation" is a Mad Lib.
8. What does ROI look like?
Fewer no-shows, fewer "just checking in" emails, faster setup. If you cannot name the number before you buy, you are buying a demo.
9. What goes wrong if you start too early?
Cold traffic gets a confident wrong answer. Qualified traffic gets bounced to a form. Both are self-inflicted.
10. Can several agents share a lead?
They can pass context. They will also duplicate questions unless one owner maps the handoff.
11. What is Profound, next to Drift or Qualified?
Profound is for AI-search visibility. Drift and Qualified are on-site conversation tools. Different layer. Do not swap them.
12. Where should a Profound-class tool go first?
On the prompts and cited pages you already have. Add on-site chat only after that picture is clear.
13. How much does Profound cost?
Custom. We will not invent a rate card. Ask them. For on-site SDR chat, the published ranges above are a starting point only.
14. Is Profound a chatbot?
No. Chatbots wait on your page. Profound is sold around agents and analytics for generated answers.
15. Should we publish a standalone Profound page?
Give the existing funnel pillar 60 days with the name in the copy. If citation share has not moved, then consider a dedicated URL.
