What to Expect

The short version

• The bot answers only from what you give it. It doesn’t search the web or guess. A clean FAQ or answer document is the best fuel you can provide.
• Expect a tuning period. Setup includes a few structured rounds; after launch, most bots settle into a steady rhythm.
• How you give feedback drives how fast we improve. Give the exact words to use, not “make it better.”
• A well-trained bot lands around 90% right. It’s not a mind reader — if your source is unclear, the bot will be too.
• Setup gets you live with a working bot — not a full AI transformation. Bigger builds are separate projects, and we’ll always tell you which is which.

---

How Fulcrum works

The bot answers using the sources you provide — your documents, FAQs, and approved content. It doesn’t invent answers from the open web, and it doesn’t know anything you haven’t written down somewhere it can read.

The best source is a clear, well-organized FAQ or answer document — the definitive answer key for your business. The cleaner that source, the better and more consistent the answers.

Your website can work too, but there’s a catch: most sites have grown over years and hold outdated pages and content that quietly contradicts itself. The bot reads all of it and can’t tell which version you meant — the most common cause of inconsistent answers. One document you control almost always beats a big, sprawling website.

---

The build and launch process

Once we have your content, we build the bot and give you a test link to try. This is where the real work happens — you ask the questions your real users ask and quickly see where it shines and where it needs correcting.

Most bots follow the same path:

1. Round 1 — first corrections. The initial pass of fixes after the first build.
2. Round 2 — refinements. Smaller adjustments, enough to confidently launch on your site.
3. Round 3 — real-world tuning. The questions that only show up once real customers use it — usually a two-week window after launch.

During setup we also build a test set from real questions you provide (usually 10–30) and give you a simple report to review every answer in one place before launch. We re-run it after tuning and again later whenever you make a big change.

With good training, about nine of ten answers are solid out of the gate — the tenth sticks out, and that’s what we fix. It levels off. The bot also surfaces things about your own content: gaps, unexpected questions, pages that contradict each other — useful data about where your messaging is unclear.

---

What’s included — and what’s a separate project

Setup gets you live with a working, tuned bot. It’s not a full AI transformation. These are real, worthwhile projects — just scoped and priced on their own:

• Connecting the bot to your other systems for live data
• Guided step-by-step flows or branching conversations (if the answer is no, ask something else)
• Sorting users by type and routing them accordingly
• Large content migrations or major website overhauls
• Documenting or building your internal processes
• Rewriting your company’s messaging

Our promise: when you send a request, we tell you up front whether it’s a quick change (covered by your plan) or a bigger one (we’ll estimate first). You’ll never be surprised by where a request lands.

---

How to give feedback (the important part)

The bot improves through your feedback, and specific feedback is worth ten times vague feedback. When it gets something wrong, do three things. Note: this applies to the setup process, and for managed service customers. Self-service customers are responsible for ongoing updates after launch.

1. Flag it with the word “Feedback.” That’s how we find it. You can do this mid-conversation — say “Feedback” and give the reply you’d prefer, and we can turn that into a fix.
2. Point to what the bot said (or note that it’s the last response).
3. Tell us exactly what it should have said — the specific words, facts, and links.

The rule of thumb: “Instead of X, say Y.” If you can write the “Y,” we can train it. If you can only say “be better,” the bot can’t guess it — and neither could a new hire.

The bot can only be as clear as the instructions it’s given.

---

What to realistically expect

Accuracy. If your answer document is under about five pages and genuinely clear, it should handle roughly 90% of questions well. The last 10% — the perfect link, this word not that one — is the ongoing tuning, the same last-mile work you’d do with a new employee.

It only knows what it’s trained on. The bot knows your documents, not your business the way you do — it can’t read intent you haven’t written down. If a smart new hire reading your site would be confused by gaps or contradictions, the bot will be too. The fix is almost always upstream: if two documents disagree, don’t argue with the bot — make the source say one thing.

Mistakes are rare, but not impossible. Because the bot answers from your approved sources rather than guessing, made-up answers are uncommon by design — but not zero. We include an “AI can make mistakes” note so your users have the right expectation too.

---

Voice and tone: how far we can go

We can shape how the bot sounds. It helps to know there are levels, and how much each asks of you:

1. Professional baseline. Polite, on-brand, safe. The default, and it handles about 90% of cases.
2. Recognized styles. Well-understood tones — say, a formal, compliance-minded voice.
3. Specific behaviors. Consistent patterns like “always end with a follow-up question” or “keep answers short.”
4. Personal voice. “Sound like our founder,” “be warmer,” “be more enthusiastic.” The subjective level, where expectations and reality tend to drift apart.

That last level works only with examples. “Be more kind” means different things to different people, so it’s hard to act on. What works is 10–20 “instead of X, say Y” swaps — usually gathered by talking to the bot and correcting it as you go. With those, we get remarkably close. Without them, we’re both guessing. Voice tuning is a dial, not a personality transplant.

---

What Fulcrum is — and isn’t

Fulcrum is a question-answering bot — excellent at clear facts, consistent answers, and structure. While we can build it, the basic Q&A bot is not a business-process engine, and asking it to be one tends to lead to frustration.

If you need any of the right-hand column, we can build toward it — it’s just a larger project than a basic Q&A bot, scoped separately.

---

Your ongoing plan

Once the bot has stabilized, most businesses on the managed service plans need only 1–2 small requests a month — tuning an answer, uploading an updated policy, reviewing a conversation to see why the bot answered a certain way, or a small wording or behavior tweak. You also get self-service settings: your bot’s name, logo, welcome message, alert recipients, chat history, and document uploads.

Bigger changes are separate projects, and we’ll tell you which bucket a request is in before any time is spent. One honest note: we keep the platform healthy and the bot performing, but we won’t become experts in the nuances of your business — you always know those best. That’s why your source document and feedback are what make the bot great.

What you’ll see

• Full chat transcripts — every conversation, anytime.
• A weekly report — a plain summary of conversations, or a strategy-focused Growth Report that flags patterns and opportunities.
• Alerts and exports — set up notifications and pull your chat history whenever you need it.

Is my data safe?

Yes. Your data lives on secure, US-hosted infrastructure:

• Encrypted both when stored and when sent
• Kept separate from every other customer’s data
• Sign-in required to make any changes
• Your documents are never used to train public AI models
• The bot answers only from your approved sources — never the open web

Where possible, sensitive data stays in your own systems and the bot pulls it live when needed, rather than us copying it over.

Who does what