Month One Isn't AI. It's Data Sanitation
The B2B AI industry sells "30 days to value." The real timeline of a deployment looks nothing like that — and the buyer who knows this in advance is the buyer you actually want.
Open any B2B AI vendor's website and you'll see some version of this promise: Deploy in 30 days. See ROI in 60. Transform your operations in 90.
In six deployments over 18 months, we have not had a single project where AI was producing value in the first 30 days. Not because we were slow, or our clients were unsophisticated.
Because AI deployment in non-AI-native companies is almost never about AI in the first 30 days. It's about data sanitation.
This is the part of the industry nobody wants to talk about — and the part that decides whether a deployment succeeds.
Here's what actually happens in the first 90 days of a typical B2B AI deployment, based on our six.
Month 1: Data sanitation. (35-50% of total project time.)
The client confidently tells you they have "millions of records." They do. Of those millions:
20-30% have inconsistent field naming (the same customer is "John Smith," "J. Smith," "Smith, John")
10-20% are duplicates across systems (CRM vs. ERP vs. spreadsheet exports)
15-25% have missing critical fields (timestamps, IDs, categories)
5-15% are simply wrong (legacy bad entries that nobody cleaned up)
Net usable data, before any AI touches it: somewhere between 40% and 70% of the original volume.
This isn't anyone's fault. Operational systems were never built to be AI training inputs. They were built to run a business — generate invoices, ship products, log calls. Data quality wasn't the design goal; transactional reliability was.
But before any model can do anything useful, this layer has to be cleaned, standardized, deduplicated, and version-controlled. There is no shortcut. Skip it and your AI hallucinates against bad data — which is worse than no AI at all, because the client trusts the output.
Month 1 is when we earn the right to deploy anything.
Month 2: Permission politics. (15-25% of total project time.)
Once the data is clean, you discover it lives in seven different systems, each owned by a different department, each with its own data access policy.
The buyer who signed your contract — usually a COO, VP of Operations, or business-line head — has executive authority. They don't have access. To touch the data we need to touch, we need sign-off from:
IT (custodial control of the systems)
Legal/Compliance (depending on industry — HIPAA for medical, FINRA for finance, etc.)
The data owners in each department
Often a security review (especially in healthcare, financial services, regulated industries)
Each of these stakeholders has the authority to slow the project. None of them has the authority to accelerate it. So progress in Month 2 isn't measured in code shipped — it's measured in meetings cleared.
This is the phase that kills most B2B AI projects in the wild. The founder who promised "30 days to value" reaches Week 6 with zero AI deployed, just a stack of pending security reviews. The client loses faith. The project enters death spiral.
Month 3: Infrastructure. (15-20% of total project time.)
Now you can finally build. But you can't deploy AI in isolation — it has to live inside the client's existing systems.
In our six deployments, the average tech stack we had to integrate with included:
One CRM (usually Salesforce, sometimes a custom build from 2014)
One ERP (varies wildly, often legacy)
2-4 operational systems (call routing, inventory, scheduling, etc.)
An identity provider (Okta, Azure AD)
A data warehouse (sometimes; often built ad-hoc during this phase)
Pipelines have to be built. APIs documented (and sometimes patched). Auth flows tested. Monitoring set up. Failover behavior defined.
This is unglamorous work. It's also non-negotiable. The AI is going to fail at some point — every system does — and how it fails determines whether the client trusts it for the next decade or pulls the plug in month four.
Month 4+: AI actually deploys.
This is when "the AI part" begins. Models go live in shadow mode first, running alongside the existing process. Human review of outputs. Tuning thresholds. Catching edge cases. Gradual ramp-up of which decisions the model is authorized to make autonomously.
Real, measurable value typically arrives in Month 4-6 of a deployment.
Not Month 1. Not Day 30. Not ever in the first 30 days, for any client we've worked with whose primary business wasn't already AI-native.
WHY THIS TIMELINE IS NON-NEGOTIABLE
The natural reaction here, especially for technical founders, is: can't we compress this?
Sometimes, marginally. In one of our six deployments, the client had a clean data lake and a modern Okta-based identity layer — Month 1 collapsed to two weeks, Month 2 to ten days. We were producing AI value by Week 6. That client was an outlier.
For the other five, the timeline above held remarkably consistent — give or take 20% in each phase. The reason isn't technical. It's organizational.
Data sanitation can't be parallelized past a certain point because it requires institutional knowledge — someone in the client's org has to know that "ACME Corp" in the CRM is the same entity as "Acme Inc." in the ERP. That knowledge lives in people, not systems.
Permission politics can't be skipped because the people who own the data are doing their actual jobs. Their cooperation with your project is a side quest. You wait.
Infrastructure can't be hacked together because the client will live with it for years. A duct-taped integration that works in Week 8 becomes the production system the client builds around in Year 2.
These aren't bugs in the deployment process. They're the deployment process.
WHAT THIS MEANS IF YOU'RE BUYING B2B AI
If a vendor promises you AI value in 30 days for a non-AI-native business, one of three things is true.
They've never deployed at your scale before, and they don't know what they're walking into.
They're going to skip the foundation work and ship something that breaks in Month 4, after their contract has locked in.
They're going to count "data audit complete" as "AI deployed" and quietly redefine the goalposts.
The vendor who tells you Month 1 will be data sanitation is the vendor who has actually shipped before. That conversation is awkward in a sales meeting. It's the conversation worth wanting.
Ask your prospective AI vendor: "What does your typical Month 1 actually look like?" The answer tells you everything.
WHAT THIS MEANS IF YOU'RE BUILDING B2B AI
If you're an AI services founder reading this — three things we've learned the hard way.
First, make the data sanitation phase visible to the client, not invisible. We deliver a "data quality report" at the end of Month 1, before any AI work. It includes the duplication rate, the field-completeness percentage, the systems-of-record conflicts we found. Clients are stunned by it every time. It positions our work as discovery, not delay.
Second, build the permission politics into your project plan, not around it. Schedule the security review at kickoff. Identify the four-to-six approvers in Week 1. Track their sign-offs as a deliverable, not a dependency. This is the difference between a project that ships and a project that drifts.
Third, sell the timeline you can actually deliver. We tell every prospective client: "Month 1 is data sanitation. Month 2 is permissions. Real AI value arrives in Month 4-6." About a third of prospects walk away. The two-thirds that stay are the clients we can actually serve — and they convert at a much higher rate than the prospects who needed the fiction of 30-day deployment.
ONE THING WE MIGHT BE WRONG ABOUT
This timeline applies to non-AI-native businesses — auto retail, medical devices, real estate, film, manufacturing. The kind of buyers who don't have modern data infrastructure as table stakes.
For AI-native companies — tech-forward startups, modern SaaS platforms, AI labs themselves — Month 1 collapses substantially. Data is already clean. Permissions are already centralized. APIs already exist. We've seen 30-day value delivery be entirely realistic in those contexts.
But that's a small fraction of the B2B AI buyer market. The vast majority of enterprise AI buyers are the ones for whom this timeline holds. Pretending otherwise — for them or for ourselves — is how good AI projects die in the field.
If you're selling to AI-native buyers, this timeline isn't your reality. If you're selling to anyone else: it is.
Working notes from B2B AI deployment in North America. Part of an ongoing series on what we keep noticing across wildly different industries — and what the industry isn't ready to say out loud.
