What Most Teams Miss: Fixing the Data Foundation Before You Buy the Agent

Four stories in GTM and AI this past week look unrelated, but they are the same story. A founder asked whether his business is even ready for AI. A marketing org got handed the keys to AI adoption while other teams got cut. Governance vendors converged on the same missing layer. And Salesforce quietly admitted its own agent bet hasn't stuck. Almost nobody is failing because of the model. They're failing because of what the model was asked to sit on top of, and that gap shows up in your CAC whether you're tracking it there or not.

Table of Contents

1. Is your business actually ready for AI, or is your outbound just broken?
2. Who should own AI enablement, and why are companies cutting the teams built to lead it?
3. Why are companies pausing AI projects right now, and is that actually a bad sign?
4. Why have fewer than 10% of Salesforce customers scaled Agentforce, and what does that mean for your stack?
5. How do you actually calculate CAC for your business, step by step?
6. What do these four patterns actually cost you in CAC?
7. This week's move; starting Monday 8/10/26

Is Your Business Actually Ready for AI, or Is Your Outbound Just Broken?

Most growth-stage teams don't need AI first. They need outbound wired into something repeatable, and skipping that step is the single most common mistake I see.

I ask founders some version of this almost every week: is your data ready, and is your outbound actually broken, or just unscalable? Those are different problems with different fixes, and most teams don't know which one they have before they start shopping for AI.

Many teams are racing to add email as a "free way to get a lot of eye balls on your product", but outbound email strategy in 2026 is far more complex than it was three years ago. Deliverability rules have shifted enough that the old "spray and pray" playbook is already stale, and it damages your sender reputation more than most teams realize. Before stacking anything on top of that motion, you have to look at the layer underneath it: is it actually possible to add AI here, or do you just have a stale process that needs to be re-engineered first? Those are two different jobs, and most teams skip the second one entirely.

That's usually why the first GTM hire at a growth-stage company shouldn't be RevOps. It should be a GTM Engineer (and not a BDR or SDR), brought in to connect the tools so data flows instead of sitting in silos, turn "we send emails and hope" into something repeatable, and get the pipeline mechanics working before anyone tries to optimize them. RevOps is a natural next step once there's an engine worth optimizing, not before.

I keep watching teams rush to AI anyway, creating a new cost-of-goods line in the P&L, only to find that adding AI doesn't fix the broken process sitting underneath it. That new line item isn't neutral either. It changes the calculus on headcount at exactly the moment tech employment is already unstable, with well over 150,000 tech workers laid off this year as companies restructure budgets toward AI infrastructure. Before adding AI, it's worth answering three questions honestly:

• Do you know where deals actually stall, or are you guessing?
• Is your data clean enough to trust the output?
• And are you optimizing a process, or just dressing up a broken one?

An unscalable, unrepeatable outbound motion is already a CAC problem before AI ever enters the picture (every rep-hour spent on spray-and-pray outreach is fully-loaded cost with nothing to show for it). Adding an AI subscription on top of that motion doesn't lower CAC. It just adds a new expense to the same leak.

Learn if your business is ready for AI today!

Who Should Own AI Enablement, and Why Are Companies Cutting the Teams Built to Lead It?

The people best positioned to own AI enablement are the operators who were already doing tool orchestration long before AI gave it a name, and companies cutting those teams are eliminating their own answer.

Zapier just gave its CMO a combined role: Chief Marketing Officer and Chief AI Transformation Officer. The reasoning is straightforward once you see it: marketing teams that become a company's heaviest AI users end up setting the blueprint everyone else follows. Klaviyo made a similar bet from a different angle, acquiring an AI-powered customer success startup and installing its founder as Chief Product Officer to run its AI agents (one that builds marketing campaigns, another that handles post-sale support) in front of 200,000 businesses.

Set that against what's been happening since late 2024: companies scrapping marketing, scrapping demand gen, scrapping growth teams, all while claiming to go all-in on AI. Here's what those companies are missing. Marketers and RevOps operators were doing agent-style tool orchestration a decade before agents existed: wiring API connections between the CRM and the marketing automation platform, building webhook chains for lead routing, running enrichment workflows, stitching attribution across five systems that never wanted to talk to each other. Zapier's entire origin story is operators wiring tools together because nothing integrated natively. A "Zap" was an agent before agents existed, just with a human writing the logic instead of a model.

That exact muscle (mapping systems, moving data between them, defining triggers, knowing when a workflow needs a human in the loop) is what AI GTM enablement actually requires. The progression from marketing automation to AI orchestration is the same job with better tooling underneath it. So when a company cuts its marketing operators and then asks who's going to lead AI adoption, it has usually just walked its own answer out the door.

Cutting that team doesn't remove the cost of AI enablement. It just delays it and hands it to whoever inherits the mess, usually at a higher price, while the Martech stack nobody understands anymore keeps quietly degrading acquisition efficiency underneath everyone's feet.

Learn exactly how revenue leaks may be impacting your business.

Why Are Companies Pausing AI Projects Right Now, and Is That Actually a Bad Sign?

If your team paused an AI project this quarter, that's not falling behind. It's catching up to the operational questions that should have been answered before launch.

Three announcements this last week tell the same story from different angles. Red Hat launched an open-source project to automate AI governance. Zenity raised $125 million to secure AI agents, on the position that agent security is a fundamentally different problem than protecting models and prompts. SAP is expanding its AI Agent Hub toward asset discovery, governance assessments, observability, and access control built directly into the workflows themselves. Vendors don't converge on the same problem by accident. They converge because customers are stuck on it.

Here's what I think actually happened. In the race to launch, most teams deployed agents before answering the operational questions: who owns this workflow, which data can it touch, whose credentials does it run on, what gets logged, and who signs off when it acts on its own? Those questions felt like friction at the time. Now a CFO or a security team is asking them, and the honest answer is often that nobody knows yet.

Legal, RevOps, MarketingOps, and procurement teams often spotted this exact trend well before the AI race started, and they work hand-in-hand with governance and compliance to keep GTM work risk-averse. Those departments get glazed over constantly, but they're the ones providing oxygen to AI projects that actually survive contact with a CFO. Most processes go stale within about a week in 2026 given how fast the landscape is moving, which makes constant monitoring the job, not a phase you finish once.

What I take from last week's news is actually encouraging: governance is becoming an operating layer that runs alongside the work, not a policy document written after an incident. That shift is what lets a paused project get turned back on with confidence instead of getting shelved for good. The pullback isn't the failure. It's the market growing up.

Salesforce is the clearest, most expensive proof of exactly this pattern, at a scale that makes the stakes impossible to ignore.

Learn how other small businesses are deciding what to build first, and why AI projects are stalling.

Why Have Fewer Than 10% of Salesforce Customers Scaled Agentforce, and What Does That Mean for Your Stack?

Fewer than 10% of Salesforce customers have scaled Agentforce past a pilot, and the reason isn't the model. It's that the data underneath was never built for agents to trust.

Salesforce is down 30 to 40% from its 2023 stock price peak, has every structural advantage in the agent race, and still hasn't made Agentforce stick. It bought Slack, picked up Qualified, and recently acquired Fin (formerly Intercom), rolling out Fin's consumption-based pricing model across its own agentic push. All of it is building the story of an agentic ecosystem, while investors keep debating how quickly any of it actually converts to revenue.

The blockers behind that sub-10% number are the same three every time:

• Inaccessible data
• Rigid workflows
• And fragmented governance

Salesforce has a well-earned reputation as a rep's worst nightmare. Too many custom fields, too much manual data entry, too much admin work, and not enough time to actually sell. That is exactly why a wave of leaner CRM alternatives has gained traction lately.

The blockers are identical at a fraction of Salesforce's size. Auditing smaller stacks, I find duplicate records, attribution fields that don't reconcile between systems, and workflows held together by manual workarounds. Nobody can say which fields an agent is allowed to touch, or who approves it when the agent gets something wrong. You can't acquire your way past that problem. Salesforce is proving it at the largest scale imaginable, and even other major vendors have had their own recent stumbles.

There's a second signal worth noticing here too: more small teams are choosing to build a CRM around their own workflows instead of renting one built for someone else's. When agents can execute rules directly, a rigid system of record starts losing its grip. My takeaway: the agent race won't be won by whoever ships the most agents. It goes to the teams that fixed their data, defined ownership, and knew the measurable outcome before they deployed anything. If the biggest CRM company in the world can't skip that step, neither can you.

Salesforce can absorb a stalled, multi-year agent bet and still call it strategy. Most GTM teams can't. Every dollar sunk into an agent license sitting on top of data nobody can trust is a dollar that inflated the cost structure without moving a single number that actually defines CAC efficiency.

Learn why teams are debating whether to build an internal tool.

How Do You Actually Calculate CAC for Your Business, Step by Step?

Start with the same formula from Edition #1 (fully-loaded sales and marketing spend divided by new customers acquired), then get specific about what belongs in each bucket before you trust the number.

Pick a consistent time period first; monthly or quarterly works for most teams. Then total every dollar spent on sales and marketing in that window: salaries and commissions, ad spend, tools and software, agency or contractor fees, and a reasonable allocation of shared overhead. Divide that total by the number of new customers you actually closed in the same window. That's your baseline, fully-loaded CAC.

Teams go wrong in the second step: deciding what counts as a "new customer." Self-serve signups, sales-assisted deals, and expansion revenue are not the same motion. Blending them into one number hides more than it reveals. If you can, calculate CAC separately by channel and by motion (paid vs. organic, PLG vs. SLG) so you can see which one is actually efficient instead of averaging a good channel and a bad one into something that looks acceptable.

This is exactly where this week's four topics loop back around. The same operational discipline that makes your data trustworthy enough for an AI agent (clean records, defined field ownership, attribution that reconciles across systems) is the same discipline that makes a CAC-by-channel breakdown trustworthy in the first place. If your lead source fields don't reconcile between platforms, your CAC math was already wrong before an agent, or anyone else, ever touched it.

Understand the formula for calculating CAC.

What Do These Four Patterns Actually Cost You in CAC?

Each pattern above hits CAC from a different angle. Lined up together, the theme repeats: the foundation gets skipped, and the bill shows up later, usually bigger.

1. GTM readiness before AI: Stacking AI on top of a broken outbound process doesn't fix the process, it just adds a new expense on top of the old inefficiency. The result: acquisition spend keeps leaking through the same broken motion, now with a subscription cost layered on top of it.
2. AI enablement ownership: The operators who already know how your systems connect are the cheapest path to AI enablement you will ever have. The result: cutting that team doesn't remove the cost, it delays it and hands the rebuild to someone else at a higher price.
3. Governance and paused projects: Skipping the ownership and access questions at launch doesn't save time, it just moves the delay to whenever a CFO or security team finally asks. The result: most retrofitted AI projects get paid for twice, once to launch and once to make them accountable.
4. Salesforce and data readiness: No amount of agent licensing fixes data that was never built for agents to trust. The result: acquisition and expansion spend riding on that data keeps compounding the same errors, now at agent speed instead of human speed.

This Week's Move; starting Monday 8/10/26

Pick the outbound or CRM workflow your team is most tempted to hand to AI next. Before you do, answer the three questions from the first section above in writing:

• Where do deals actually stall?
• Is the data clean enough to trust?
• Are you optimizing a process or just automating a broken one.

If any answer is fuzzy, that's the fix to make first, not the agent or software to buy first.

Have a workflow, metric, or mistake you want broken down in a future issue? Reply to this or send me a DM on LinkedIn.