AI Your Way: Where to Start with AI in Your Digital Experience

Customers now expect the kind of instant outcomes they get from AI. Many digital experiences have not caught up. Once people have used a conversational assistant, they return to a website that expects them to find the right menu, scan pages of links, and assemble the answer themselves. Patience for that is dropping, and conversational, outcome-first experiences are becoming the baseline.

The gap is that users are AI-native and most digital experiences are not. This post pulls together the main ideas from the AI Your Way webinar on how to close it without starting over: where to begin, what to look for in a CMS, and how to choose a first project that your team will actually finish.

Tools are not the same as capability

When we asked what creates sustainable digital momentum inside modern organizations, one poll result stood out. Eighty-four percent of organizations said they invest in tools, and only fourteen percent said they invest in workforce capability. It is easy to be distracted by the latest tool. The harder questions are whether the people doing the work have the right skills and whether they have been given time to build them.

It helps to sort AI initiatives into two buckets. The first is knowing something: missing metadata, broken links, stale pages, or what competitors are publishing. The second is doing something: writing help in the editor, meta descriptions and FAQs drafted on publish, or answers delivered in search. Starting in the first bucket is the lower-risk path, and it builds the trust needed to move into the second.

Start with the friction, and start by reading

Find the complaint you hear every week and work from there. The lowest-risk starting point is to read before you write. AI can audit a site for missing metadata, broken links, stale pages, and translation gaps, report what it finds, and change nothing. There is no risk to live content and the payoff is immediate.

The results can be surprising. In one live-site audit, 71 percent of pages had no meta description. A site inventory produced a spreadsheet of 585 live pages, along with 542 redirects, drafts, unpublished pages, and navigation-only entries. Neither finding required AI to change anything on the live site.

The surprise is that a small digest can turn into a strategy. A daily AI summary of what competitors publish on their websites is a good example.

Optimizing for customers who ask questions

The shift from keywords to questions changes what a website is for. People want to ask once and get the answer, not click through pages of links to assemble it. That means making content easy for AI to understand: structured data, clear headings, and FAQs. When an AI system answers on your behalf, it works from your content, so the quality of that content decides the quality of the answer.

Audience signals matter too. For an event-heavy site, AI can add local and audience context, such as who should attend, where, and when, so visitors do not have to work it out themselves.

A regional tourism organization shows what this looks like in practice. A three-person team now has an AI chat assistant on its website that gives visitors fully contextualized answers and advice. It can draw on data from other sites, such as the city's website, while the team keeps full control over the context it uses. The backend also shows which themes visitors ask about, and that visibility feeds the content strategy.

In the webinar poll, 40 percent of attendees saw the greatest opportunity in content creation and optimization, 40 percent in search and discovery, and 20 percent in understanding customer behavior and intent.

When the marketing team feels the friction

Before adding AI to content management, it is worth asking five questions of the CMS:

• Is the AI in the editor? Help that appears while content is being written is more likely to be used than a separate tool.
• Does it know your content model? Content types, taxonomies, and modules should be visible to the AI. Sitefinity's MCP server does this.
• Does it respect permissions and workflows? AI-generated work should land as drafts, not be published blind.
• Can it work in bulk? Updating many meta titles or descriptions at once matters more than polishing one.
• Is it open? Open APIs and MCP connectors keep the team from being locked into a single tool.

The same principle applies beyond the CMS. A B2B supplier with hundreds of locations treats local search as a key revenue driver, but keeping up with Google reviews at that scale overwhelmed a centralized team. The workflow now pulls reviews through Zapier, has Gemini draft a response, and sends it to a marketer to accept, reject, or edit. The human stays in the loop, and effort goes to the reviews that actually warrant attention.

Other examples followed the same pattern of draft first, review second. Writers can send Word documents to the CMS as draft posts with suggested metadata and tags, which removes a lot of copy and paste. AI can classify content against a list of approved subtopics and leave an item uncategorized when nothing fits. Translation workflows can use approved glossaries and create drafts for human review, and one example supported seven languages.

What developers and IT see

The developer side of the work uses the same open connections. A CDN connector showed that legacy redirect URLs were still being used after a rebrand. Cloud cost analysis bridged IT and marketing: traffic spikes from campaigns show up as hosting costs, and one hosting review found production database costs running about 60 percent higher from June through August. An AI agent running PageSpeed Insights, connected to the CMS through MCP, diagnosed issues in plain English and generated a fixes file for a developer to work from.

Shared AI skills and reusable components can speed this work up. Technical teams remain important for deployment, observability, and oversight of the systems the AI connects to.

Choosing the first project

A crawl, walk, run model is a useful way to think about AI maturity. Stumbling between stages is normal and expected. The first project is easier to choose with a few tests:

• It happens every week. Somebody already complains about it.
• It has an owner. Agree on what done means, and make sure AI outputs land as drafts.
• It fits where the work lives. What you already own is enough when the problem lives inside the CMS, such as SEO and metadata. Add or build something when the work crosses systems or lives in the tools your team already uses.
• It is ready. Wait when there is no owner, no governance, or the underlying process is broken. AI can make a broken process faster.

Progress outlined three paths for teams using Sitefinity. The first is the native AI features built into the CMS. The second is extending those experiences with Progress Agentic RAG, for contextual search and product discovery. The third is integrating other tools, such as Gemini or Claude, through custom MCP connections. Progress also described its acquisition of Domo, with governed data intended to help teams understand performance across channels and optimize spend.

You do not have to go all in at once. Each path can be adopted on its own, in the order that fits your team.

Start small, then dream

The most useful advice from the session was to stay curious and not worry about what the technology can or cannot do. Ask "what if?" before moving straight to solving, and let the conversation with your team and partners come first. Start small, then think about what the future could look like once the first draft-and-review workflow is working.

Many organizations are already moving. In the modernization poll, 75 percent of attendees said their initiative is already underway, and the other 25 percent said it would begin within six months.