Your AI agents are only as good as the data they can see
Mikel Lindsaar, CEO and founder of StoreConnect, speaking at Dreamforce 2026. An edited transcript of the session follows.
I usually do a lot of demos in my talks. Not this one. There’s so much to demo with AI tooling that a few minutes on stage doesn’t do it justice, and the AI agent takes 15 seconds to a minute to respond each time. So instead, let’s talk about how you create an agentic foundation on StoreConnect this week. And I do mean this week.
Context is what kills AI projects
Agents are only as good as the data they can see. That’s the whole talk in one sentence, and the problem is that the data usually lives somewhere else.
You’ve already experienced this. You open ChatGPT, ask it a question about your own business, and it can’t answer, because it doesn’t know your data. That’s context, and context is the thing that will make every one of your AI projects fail. Yours, your clients’, your staff’s. It’ll be context.
And it isn’t just context about the question being asked. Think about all the parts of your business an agent might need: help articles, blog posts, support tickets. What has this client bought from us before? What are they likely to buy next? What products pair with the ones they already own? Are they on a subscription? How much did they spend last year? What’s their name, their phone number, their state, their time zone? When are we booking their next appointment?
The volume of context you could feed an agent is almost limitless. That’s the roadblock every AI implementation hits. There is never enough context.
Between 80% and nearly 90% of organizations report that their data is disconnected. I think that’s lowballing it. One of my favorite questions to ask a new prospect is, “How many single sources of truth do you have?” They think about it for a second and say, “Oh yeah, we’ve got about seven.” That’s disconnected data.
The wrong source is worse than no source
Context and source are two different problems, and both matter. It’s all well and good to have source data flowing to your agents, but if it’s the wrong source, it’s almost worse than having nothing.
We’ve had clients pulling purchase data only from their accounting system. That data is correct. It’s genuinely what the customer bought. But it’s missing everything around the sale: what else they looked at, what products relate to the one they chose, what was in the cart.
There’s a saying I use for this. The noun should own the verb. The accounting system should own the transactions and the money flow. The CRM should own who the customer is and what they’ve done with you. The e-commerce platform should own when the transaction happened, what was bought, and what was in the cart at the time. Get the source sitting in the right place and you reduce the third problem.
Fragmentation is the third problem
Fragmentation reportedly costs enterprises between $8 million and $12 million. As a share of revenue I suspect it’s worse the smaller you go, because smaller companies have been bolting on ad hoc systems for years and the fragmentation just goes wild.
So there are three things to get right: increase the context, get the source correct, reduce the fragmentation. Do all three and your AI tooling will succeed. Fall short on any of them and your AI project will fail by roughly that much.
You can fix fragmentation the hard way, with a spaghetti mess of middleware holding everything together. It works. It’s possible, at the right price, and you’ll spend a lot of money making it work.
One of the most powerful things about Salesforce is that you don’t have to. If your Salesforce org is sitting in the middle of that mess, you have a huge opportunity to simplify the stack around it. That’s where StoreConnect comes into its own. We started as a way to connect an e-commerce website to Salesforce. We’ve become a complete content, commerce, and customer platform. We’re putting the head back onto headless Salesforce. Every Salesforce talk right now is about going headless, about interacting through Slack. That works for your internal team. It doesn’t work for a customer who has never spoken to you before. They’re going to hit a website.
What it looks like when all three go wrong
My favorite example is Chevrolet. They put an AI chatbot on their website to answer customer questions about cars. Someone asked it what the best electric car to buy was, and the Chevrolet chatbot sold them hard on a Tesla Model 3.
Maybe that’s technically correct. But look at what that agent was missing. It was missing the context that Chevrolet sells an electric car, because it never even mentioned one. It was missing the right source of data. You’re a sales agent. Your job is to help this person buy a Chevrolet. It was missing the context that this customer might have bought ten Chevrolets already, loves the brand, and is never buying a Tesla. Maybe they don’t like Elon. There’s a heap of context missing from that sales conversation.
And the data was fragmented, because the agent didn’t know that a person asking about electric cars on the Chevrolet website should be hearing about a Chevrolet.
Jo Swaddling, who runs Swan Technologies in Australia and has delivered a few StoreConnect projects with AI, put it best: having your data in separate systems makes it very hard for AI. That’s the problem we solve.
What an agentic foundation actually means
93% of retail organizations are talking about AI and how to put it to work. Everyone’s talking about it. So what do I actually mean by an agentic foundation? Four things.
One customer record, one order record, one product record. One record, not five. You don’t want customer records scattered across the universe. You want one piece of source data.
Live data, not batch. Plenty of systems will connect a website to your Salesforce org and update product data every hour, every two hours, or overnight. That won’t cut it. If a product goes out of stock, the agent cannot still be recommending it to the next customer. StoreConnect’s built-in sync means the data is there as it changes.
Governed access. Building on Salesforce ten years ago turned out to be a very smart move, because agents working on StoreConnect sites through our MCP tooling inherit that governance. They can only touch the records they’re allowed to touch and only see the data they’re allowed to see. Fine-grained access control is a Salesforce superpower that goes completely unnoticed, because it’s one of those things that only matters when it goes wrong.
Structured content in one place. Unstructured data is just another kind of fragmentation. You can point an AI agent at your Dropbox and tell it to go find the data. It’ll figure it out eventually. Might take three days. Point that same agent at a StoreConnect-powered Salesforce org and ask for a customer record with everything they’ve purchased, every recommendation, every relevant help article, and it comes back immediately. It isn’t trawling. It’s writing queries against structured data.
What this looks like in practice
BPM Group are a Salesforce customer. They had 76 websites running on 76 copies of BigCommerce, across four countries and four currencies. If you want a definition of fragmentation, that’s it. They had four people full time doing nothing but keeping product data current. On StoreConnect with Salesforce, that came down to one person for one hour a day. And now they’re in a position to put agents to work, because their data is under control. There was no way to do anything agentic across 76 BigCommerce environments. Impossible.
At the other end of the scale, Rick launched a company with three staff on Salesforce and StoreConnect. They’re doing millions a year now and still growing. His point was simply that they wanted everything in one system. They did that five years before AI showed up, and it’s set them up beautifully.
Here’s a smaller example of why it matters. A prospect called and asked me to send over some information. They run an aquarium. We have a landing page for attractions, and I could have just sent him that URL. Instead, because our own platform runs on StoreConnect, I asked our AI agent to build a landing page for Joe, using our source data on the website. It produced about twenty pages: the systems they were already running, which we’d picked up from web scraping, so we could speak directly to how we’d replace each one. Their cafe. Their restaurant. Ticket pricing. How their subscriptions work. Not a marketing page. A worked-out proposal. When I met him, I asked what he thought of it. He said, “I just couldn’t believe what I was reading. Who does that? Did you stay up all night building it?”
Why I said this week
The reason this is achievable that fast is our StoreConnect MCP Skills plugin. Install StoreConnect in a Salesforce org and every domain we publish, every store, gets its own endpoint for your AI tools to connect to, with separate authentication and separate rights. You can run different stores with different agents building out your sites.
I did exactly this with a client. I pointed Claude at our StoreConnect MCP, it used the plugin to pull in all the data, and it built out their entire website over 24 hours for the demo. When we showed them, they asked whether we’d hacked their site, because they assumed they were looking at their own.
Replatforming e-commerce has traditionally been measured in many months. StoreConnect used to take about three months to launch a store. We’re now down to days.
A manufacturing company that sells tractors and heavy equipment wanted to launch a merchandise store. Hats, stubby holders, that sort of thing. It only does $50,000 a year, so they weren’t worried if it broke, and it gave them a small project to cut their teeth on. Their IT guy got hold of StoreConnect MCP, started building, and realized he could just do the whole thing. Two weeks later he’d migrated the entire website. It took three months to sign the deal and two weeks to implement.
You can do this with your site today. Install StoreConnect for free in a sandbox, connect Claude Code or OpenAI Codex to it, and tell it to migrate your store off WordPress into Salesforce. It can be done by tomorrow. This is real. It works right now. We do it every week.
Come talk to us and we’ll help you launch. We can put a supercharger on whatever project you’re doing, and we can get you there fast.
