Sixty days ago we launched our AI products to our user base. Over the last 30 days, we read every question clients asked the market anonymously. Read on to see what they asked and what the data told them.
Almost none of those questions were about AI.
They were about the probate two streets over. The 10am valuation, and the two agents going in after them. The vendor who wants to know why nothing is happening. The instruction that went to a rival, and why?
That's the first thing 60 days has taught us. Agents don't want AI (or data for that matter). They want the next instruction, and the evidence to win it and any edge that they can get. Sprift AI is just how they ask.
If we're honest, and I've had these conversations myself, very few estate agencies can afford or justify having a data analyst in their business, nor do they want one, but the insights our clients have been able to get from the products are staggering and follow a familiar theme: more instructions, get my stock sold/let and tell me where the next MA is coming from.
Before we get into it, it's important to clear up Sprift AI vs Sprift Analyst. They do different jobs. Sprift AI works on one property or one patch, inside the Sprift platform. Ask it about a single home, its risks or its comparables, and it answers there and then. It also prepares your Morning Briefing each day.
Sprift Analyst works across the whole market: many areas, rivals and years at once. It's where you compare your offices with your competitors, track which instructions went elsewhere, and save questions to run on a schedule.
Put simply, Sprift AI supports the work on a home. Sprift Analyst supports running the business.
One in four questions was about finding the next instruction: price reductions, withdrawals, fall-throughs, probate and new EPCs.
One in seven homes slipped to a rival. In our first Sprift Analyst reports, one in seven of the homes an agency had live, and that went on to sell, completed with another named agent. And having a post-mortem on why so that they can implement process changes to address this.
Around four in ten homes for sale had passed 16 weeks on the market, for agencies and their rivals alike.
Agents test it against ChatGPT. When they do, the difference is the data. Sprift AI answers from Sprift's own property data, not the open web.
The answer isn't the win. What happens next is: a morning briefing, a call list, a scheduled report or a feed into your own systems.
About one in four questions agents asked Sprift AI and Sprift Analyst in the last 30 days had one aim: find me my next instruction.
Reductions. Withdrawals. Fall-throughs. Probate. Homes with a fresh EPC that never made it to the portals.
"Tell me which properties in [my patch] I would be most likely to win the listing on, if I went out knocking on doors this afternoon."
"Vendor is calling in and on the phone now to book the valuation and asked why they should go with us. Give me a script."
"I am going to value this house, [two rival agents] are going after me. What are they likely to pitch the vendor as a listing price?"
Read the questions back and your can see the working day:
Questions about one home, such as flood, coal mining, radon, title and planning, went to Sprift AI: one in five of everything it was asked.
Questions about the whole market went to Sprift Analyst.
And they want it to run without them. "Send me a list every Monday morning." About a third of the agencies using Sprift Analyst in their own accounts asked for a report that runs itself.
Every agent knows the feeling. A home you listed withdraws, and the second agent ends up getting the sale over the line.
The first Sprift Analyst reports put a number on it. Across the agency groups we've reported on, a median of one in seven homes an agency had live, and that went on to sell, completed with another named agent. For one group it was more than one in five.
Sprift Analyst reports, Sep to Oct 2026[CL6.1] · agency groups anonymised · live listings from Feb 2025; completions from Apr 2025 to summer 2026
Sprift Analyst gets there by following every home an agency had live on the portals, through to the completion data, then naming the agent who finished the job.
Stale stock isn't one agency's problem. It's the market's. Around four in ten homes for sale had passed 16 weeks with the agents we ran reports for, the contract length we can infer from the data. And in every group we compared, the agency's own share was lower than its rivals'.
Sep to Oct 2026 · agency groups anonymised, same letters as the chart above[CL8.1] · October 2026 snapshots
Two more things stood out. Most homes still complete with the agency that listed them: between three in four and four in five, contrary to the old adage of "you only complete on 60% of what you list".
And pricing differs by agent, not just by area. In one group's patches, the typical home sold anywhere from 2.5% to 4.1% below its first asking price, depending on who sold it.
None of this is on a portal or available with an LLM. That's the point.
Almost every agent we've shown Sprift AI to has asked the same thing. Couldn't I just do this in ChatGPT?
It's a fair and valid question. So we started answering it live.
On one call, we put an agent's buyer brief through both: period homes in his patch that matched what the buyer wanted. Sprift AI came back with five addresses. ChatGPT came back with none. On another, an agent timed the search for the weakest listings in his patch:
"It's taken 5 minutes. ChatGPT takes like half an hour to 45 minutes trying to find the weakest listing." Self-employed estate agent
The difference isn't the AI. It's what the AI can see. A general chatbot only knows what's published on the internet up until release. It doesn't have a time machine for the data we have held every day for the last ten years.
Sprift AI reads Sprift's records on more than 30 million homes in Great Britain, matched to listings from whole of market listings data and to HM Land Registry completions. That's how it knows which homes had a new EPC and never came to market.
"But yours is real life data, isn't it?" Independent estate agent
And when the data can't answer, it says so. It doesn't guess. Asked to guarantee a home wouldn't be down-valued, Sprift AI replied: "No, I can't guarantee that, and it would be wrong of anyone to."
The same standard runs through Sprift Analyst. Every figure comes from a query that can be re-run, counts only live listings, and every report is checked by a person before it reaches a client.
Sixty days in, the pattern is clear. The agents getting the most from AI aren't asking it to be clever. They're asking it where the next instruction is, then walking in with evidence the vendor can't find anywhere else.
Sprift AI can make mistakes. Verify information against the source data.