The AI Tools That Run Our Agency: A Brutally Honest Review

We run outbound strategy and paid media, with creative for B2B SaaS clients, on six AI tools plus one custom agent we built ourselves. No affiliate links, no sponsored placements, and no vendor paid to appear here.

Why we're publishing this

Search "best AI tools for marketing agencies" and you get affiliate listicles written by people who have never warmed a mailbox. Our rule on content is the same one we use on podcasts: post about the tactics we're actually running and let the work speak for itself.

These are the tools behind our allbound coordination, where paid media, outbound, and creative pull in the same direction. If you're auditing your own spend, watch where it leaks: a tool can look justified on a feature page and still fail under pricing models once you run real volume through it.

We already covered the tool-by-tool mechanics of our outbound setup in our GTM stack post, so this review focuses on cost and value judgments.

Clay

Clay is our enrichment and orchestration hub, and nothing else matches its signal precision at agency volume. Worth it above 100 prospects a week, brutal below that.

Use case

Clay takes raw lead lists and signals, runs waterfall enrichment across its marketplace of 150+ data partners, applies ICP scoring, and routes leads into sending tools.

Cost

The old $349 Explorer tier was eliminated in the March 2026 pricing overhaul, so API-dependent workflows now start at the $495/mo Growth plan (billed monthly). At real agency volume, the sticker price is only the floor: model your enrichment runs and top-ups before you scale, since top-ups carry a 30% premium and unused Actions don't roll over.

Where it delivers

Clay pulls ahead of everything else we've tested on signal-specific enrichment. We moved chunks of enrichment off Apollo because its "Growth" title filters returned real estate agents and healthcare clinics when we were hunting technical marketers. Nothing else matches that precision at our volume.

Where it fails

The learning curve is brutal. Budget several weeks before your workflows are production-ready, and expect Clay tables to hit 40 columns fast. When the one person who understands a table leaves, the workflow leaves with them. If your outbound volume is under 100 prospects a week, skip it.

Instantly

Instantly is the sending engine behind our cold email, and at agency volume nothing beats its price-per-send. Great as an infrastructure layer, but don't trust its health scores or lean on its support.

Use case

Cold email sending and deliverability infrastructure. Deliverability decides everything downstream, and Instantly is our engine for it.

Cost

We use Hypergrowth at $97/mo for 25,000 uploaded contacts and 100,000 emails a month. The Agency bundle runs $555/mo for 500,000 emails a month plus 10,000 credits, and includes the SISR (Server and IP Sharding and Rotation) system on Light Speed-tier sending.

Where it delivers

We started using Instantly for unlimited email accounts and warmup on every paid Email Outreach plan. At agency volume, nothing touches the price-per-send.

Where it fails

Ignore the health scores. We treat them as arbitrary, and ESPs can detect the warmup pattern. The done-for-you accounts also carry real risk: we've seen a single Google suspension sweep take out all 30 accounts at once. Support isn't a reason to stay either, so we enforce deliverability discipline ourselves and use Instantly only as the sending engine.

HeyReach

HeyReach runs our LinkedIn outreach across dozens of sender seats without per-user pricing punishing us. Worth the babysitting because LinkedIn stays open when email deliverability gets brutal.

Use case

LinkedIn outreach across multiple sender seats with a unified inbox.

Cost

Growth is $79/seat/mo and includes a dedicated residential proxy per sender. The Agency plan is $999/mo for 25 senders, but you bring your own proxies at that tier, which means extra setup work and another vendor to manage.

Where it delivers

Per-sender pricing instead of per-user pricing is the right model for agencies rotating client accounts. The built-in proxies on Growth also remove a whole category of setup pain that hits you the moment you scale past a couple of seats.

Where it fails

Campaigns can fail to start cleanly and take longer than expected to send connections and messages. The blocklist only applies to newly uploaded leads, so imports need scrubbing upfront, and sessions expire and need reconnecting, so budget maintenance time. LinkedIn is worth the hassle for us because email deliverability is brutal in some niches, but this channel demands babysitting.

HubSpot

HubSpot is our system of record and the database everything else feeds into. Breeze agents help on clean data, but we don't trust them for anything that needs real judgment.

Use case

HubSpot reporting and system of record. Everything Clay enriches and Instantly or HeyReach touches lands here.

Cost

Marketing Hub Professional is $800/mo with 3 core seats and 3,000 HubSpot Credits included. Extra credits run $9 per 1,000 annually, or $10 per capacity pack. Breeze agents bill against those credits at 100 credits ($1.00) per prospecting lead and 1,000 credits ($10.00) per content piece.

Where it delivers

Breeze earns its keep on drafting. In our use, Breeze Intelligence enrichment is strong enough for mid-market SaaS cases where the CRM data is clean, and it saves real time on first-pass email and content drafts.

Where it fails

The Content Agent produces drafts that read like a competent intern, so each piece still needs 20 to 40 minutes of editing before it's usable. On a dirty CRM, Breeze generates confidently wrong answers because duplicates and inconsistent stages mislead the model. That's why we use HubSpot as the database and Claude for anything requiring judgment.

Apify

Apify feeds us fresh hiring, funding, and LinkedIn signals that static databases can't match. Highest maintenance and legal risk in our stack, and the first tool we'd cut.

Use case

Apify collects hiring and funding signals for Clay workflows, and Clay supports running Apify Actors natively. We also use it for LinkedIn signal capture, which carries the enforcement risk covered below.

Cost

Scale is $199/mo at $0.16 per compute unit (1 CU = 1 GB of RAM for 1 hour), and unused credits expire at the end of each billing cycle with no rollover. Residential proxy on Scale runs $7.50/GB on top of compute.

Where it delivers

We tolerate the maintenance for the fresh signals. In our workflows, hiring-intent data works at 24-hour latency and funding at 72 hours. That freshness is the whole point of running Apify instead of buying static data.

Where it fails

Marketplace Actors break when target sites change, and debugging from execution logs is tedious. We also treat LinkedIn scraping as a Terms-of-Service and enforcement risk: we don't assume the hiQ ruling shields us from breach-of-contract claims, and we model enforcement risk from account restrictions up to federal lawsuits. We'd rather tell you that upfront than pretend the signal layer is risk-free.

Claude

Claude is our cheapest tool with the widest surface area, powering content, analysis, and the internal tooling non-developers can actually ship. Budget for the API price hike and the shared usage pool, but this is the last tool we'd cut.

Use case

Content strategy, analysis, and the engineering muscle behind our internal tooling.

Cost

Pro is $20/mo and Max plans run $100 to $200/mo, both sharing one usage pool across chat, Claude Code, and Desktop. On the API, Sonnet 5 pricing is $2/$10 per million input/output tokens until August 31, 2026, then rises 50% to $3/$15. Budget for that jump now.

Where it delivers

The ceiling is high even for non-engineers. For us, it earns its keep on content, analytical work, and internal tooling, without turning every workflow into a full engineering project. Claude Code also lets our non-developers ship real automations we'd otherwise wait weeks for.

Where it fails

Chat and coding work share one usage pool on rolling five-hour windows, and long conversations suffer context rot. API usage in Claude Code has no hard ceiling, so we manage spend ourselves, and we model newer-tokenizer runs as roughly 30% more expensive for the same text. Even with those constraints, we'd keep it.

The custom internal agent: automations and limits

Our agent build post documents the full build. Today, it automates four things for us:

• Answers natural-language questions against LinkedIn Ads and Instantly performance data.
• Drafts replies to positive outbound responses using client knowledge bases.
• Turns Fireflies transcripts into structured follow-ups, roughly 60× faster than manual.
• Triggers Instantly and HeyReach sends off funding and hiring signals via Clay.

The automations only save time with supervision. Agent drift needs daily human management, and every workflow carries explicit escalation logic because trust breaks the moment an agent misses a handoff. Human-in-the-loop review before anything client-facing sends is our default, and it remains a permanent operational cost.

Our AI agency stack at a glance

What we'd cut if we had to pick five

Apify goes first. The signals are useful, but Actors break often, debugging from logs is slow, and it carries the most legal risk from the LinkedIn enforcement issues above. Some of what it produces can be bought through Clay's marketplace at a higher per-lead price, which makes Apify the only tool in our stack with a ready-made replacement.

After Apify, we would keep the rest for these reasons:

• Clay stays because nothing else matches its enrichment depth at our volume, even with the credit costs.
• Instantly and HeyReach each own a channel. Cut Instantly and email stops until we rebuild deliverability from scratch, which takes weeks of warmup. Cut HeyReach and LinkedIn goes dark.
• HubSpot is our system of record, and replacing a CRM is a three-month project nobody wins. Moving the data is the easy part; rewiring every connection to Clay, Instantly, HeyReach, and the agent is harder, and we'd still have to check that nothing quietly broke in a client's live pipeline.
• Claude is our cheapest tool and does the most jobs, so it would be the last to go.

That leaves Apify as the only cut that saves money without breaking the system. The wider industry data backs up the audit habit: Gartner's 2025 Marketing Technology Survey found 81% of martech leaders are testing or using AI agents, but 45% say those agents fall short of expectations, and stack utilization sits at 49%. Half of most stacks goes unused, so auditing yours every year beats piling up tools you might need someday.

Run this stack without running it yourself

Most SaaS growth leaders spend more time coordinating vendors than improving campaigns. Four specialists, four dashboards, four invoices, and prospects still get disconnected experiences that hurt pipeline.

Understory runs the exact stack above for B2B SaaS teams: Clay for enrichment and orchestration, Instantly for email, HeyReach for LinkedIn, HubSpot as the record layer, Apify for signals, and Claude behind our internal agent. Paid media, signal-based outbound, and creative sit under one roof, feeding one another instead of running in parallel silos.

Book an intro call and we'll map this stack to your ICP, motion, and current tooling.

FAQs

What AI tools do marketing agencies use?

Signal sources (Apify, RB2B) feed Clay for enrichment and orchestration. Clay routes into Instantly or Smartlead for email and HeyReach for LinkedIn, with HubSpot or Attio as the CRM layer.

Is Clay worth it for B2B outbound?

Yes, if you send 100+ prospects a week and have a full-time owner, especially for signals Apollo can't filter (ad spend, hiring specifics). No, if you're solo or low-volume without technical ownership; the ramp and credit costs will eat you.

What are the best AI tools for GTM engineering?

Apify for sourcing, Clay for enrichment and orchestration, n8n for high-volume conditional logic, and Claude or direct AI APIs for classification and personalization. Pre-process leads in n8n before they touch Clay credits.