The short answer For a 20-person company, AI integration means wiring models into three or four specific workflows that already cost you hours every week, then measuring the hours back. It is not a chatbot on your website and it is not staff training. It is plumbing: your data, your tools, your process. What the word actually covers.
The short answer
For a 20-person company, AI integration means wiring models into three or four specific workflows that already cost you hours every week, then measuring the hours back. It is not a chatbot on your website and it is not staff training. It is plumbing: your data, your tools, your process.
What the word actually covers
“AI integration” gets sold as four different things. Only one of them is integration.
- Tool access. Buying seats for a chat assistant. Useful, cheap, and not integration. Nothing connects to your systems.
- Training. Teaching your team to prompt well. Also useful. Still not integration, because when the trainer leaves, nothing in the business has changed structurally.
- Point features. Turning on the AI toggle inside software you already own, such as your CRM or help desk. Real value, narrow scope, no custom fit.
- Integration. Connecting a model to your actual data and tools so a defined task completes without a person doing the repetitive part. This is the one that changes cost structure.
TACK works as an integrator rather than a trainer, and the distinction matters commercially. Training produces a capable team that still does the work manually. Integration removes the work.
Where small companies actually stand
The adoption data is blunt about company size. The US Census Bureau’s Business Trends and Outlook Survey found overall AI usage among American businesses hovering between 17% and 20% as of 3 May 2026, with 37% of firms of 250 or more employees using AI, 32% of firms with 100 to 249 employees, and under 20% of firms with four or fewer employees. Firms below 20 employees showed no significant change over the December 2025 to May 2026 period, while firms of 20 to 99 employees did increase.
That is the interesting threshold. Twenty people is roughly where AI adoption starts moving, because it is the point at which a company has enough repeated process to automate and enough revenue to fund the work.
Scale is where almost everyone stalls. McKinsey surveyed 1,993 participants across 105 nations and found 88% of organisations using AI in at least one business function, but only about a third scaling it across the enterprise, and 39% reporting EBIT impact at enterprise level. On agents specifically, 23% were scaling and no more than 10% were scaling agents in any individual business function. Using AI is common. Getting money out of it is not.
MIT’s NANDA initiative reached the same conclusion from a different angle in The GenAI Divide: State of AI in Business 2025, built on 150 leader interviews, a survey of 350 employees and analysis of 300 public deployments. It found roughly 95% of generative AI pilots delivering no measurable impact on profitability, and attributed that failure to flawed integration rather than model quality. Lead author Aditya Challapally noted that generic tools cannot learn from or adapt to a company’s workflows. That is the whole argument for integration over tooling, stated by people with no product to sell you.
Where a 20-person company actually gets return
Integration pays where a task is high-frequency, rules-bound and currently done by an expensive human. Four candidates cover most 20-person businesses.
| Workflow | What integration does | What you need in place first |
|---|---|---|
| Inbound lead triage | Classify, enrich, route and draft the first reply within minutes | A CRM with clean fields and a defined qualification rule |
| Proposal and quote assembly | Assemble a draft from past documents, pricing rules and the discovery notes | A structured library of prior work and a real pricing model |
| Support and FAQ deflection | Answer known questions from your documentation, escalate the rest | Documentation that is accurate and in one place |
| Reporting and reconciliation | Pull from multiple systems, produce the recurring report, flag anomalies | API access to the systems and one agreed definition per metric |
Notice that every “need in place first” is an operations problem, not an AI problem. That is the honest reason most small-company AI projects fail. The model was never the constraint.
How to tell if you are ready
Three tests, in order.
Can you name the task and count it? If you cannot say “we do this 40 times a week and it takes 25 minutes each time,” you are not ready to automate it. You are ready to document it.
Does the data live somewhere a system can reach? Information in email threads, in someone’s head, or in PDFs on a shared drive can be reached, but the cleanup is the project. Budget for it honestly.
Is there an owner who will keep it alive? Every integration drifts as tools update and processes change. Without a named owner, a working automation quietly becomes a broken one within two quarters.
If all three are yes for at least one workflow, start there. If they are no for everything, spend the first month on process documentation. That work is not wasted, because it is the same work you would need to do before hiring anyway.
What we’d do
TACK has built 50+ systems since 2009, and for a company of this size the sequence is deliberately small.
- Pick one workflow, not a strategy. One measurable task, one owner, one number that should move. A four-week build beats a six-month roadmap that never ships.
- Instrument before you automate. Measure the current hours, error rate and cycle time for two weeks first. Without a baseline you will never be able to prove the return, and the project becomes an argument about vibes.
- Keep a human at the decision point. Draft, do not send. Recommend, do not commit. The value is in removing the assembly work, and the risk sits almost entirely in the final action.
- Connect it to revenue where possible. Lead triage and follow-up speed are the two integrations that most often pay for themselves in a services business, which is why we scope them alongside paid media and conversion work rather than as a separate IT project.
- Then expand deliberately. Second workflow only after the first one has held for a quarter. Our capabilities page sets out how we sequence that.
Common mistakes
Buying licences instead of building workflows. Seat licences across 20 people produce a monthly bill and no structural change. The Census data shows plenty of firms report using AI. Far fewer report it changing anything.
Starting with the customer-facing use case. A public chatbot is the highest-risk, lowest-margin place to begin. Internal workflows fail privately and teach you the same lessons.
Confusing budget for AI with budget for integration. Gartner found CMOs allocate 15.3% of marketing budgets to AI while only 30% report mature AI capability. Money is going in ahead of the operating capacity to use it. Fund the process work and the ownership, not just the tooling.
The bottom line
For a 20-person company, AI integration is three or four automated workflows with a named owner and a measured baseline, not a platform purchase. The constraint is almost always process clarity and data access, not model capability.
If you want a straight assessment of which workflow in your business is worth automating first, book twenty minutes at calendly.com/tack-media-agency/talk-to-an-expert or call TACK at 310-620-1141. Engagements start at $5,000 per month.
Sources
Related Posts
August 23, 2026
What does Google Ads cost for a small business in 2026?
Google Ads costs two things: media and management. Here are the cited 2026…
August 23, 2026
How do you choose an SEO agency without getting burned?
Four questions expose a bad SEO agency in a single call. Here is what to ask,…
August 23, 2026
What does SEO cost for a small business in 2026?
Most small businesses pay $1,500 to $5,000 a month for SEO. Here are the cited…
