How does an ASI study work?

An ASI study puts your questions to hundreds of Client Proxies (synthetic personas modelled on the people who choose firms), on more than one AI model. Your team gives a briefing and two workshops. You get the findings in about four weeks for one practice area or market, six for two. It reads how clients reason, not what they’ll do.

What happens in an ASI study, step by step?

A comprehensive engagement covers two practice areas or markets and takes about six weeks. A focused one covers one and takes about four. The weeks below are for a focused engagement. A comprehensive one runs through the same steps.

1. A 30-minute call Before the study

   What happens. You tell us the decision in front of you: a pitch you keep losing, a practice group that should be growing faster, a panel review, a repositioning. You hear whether ASI can answer it. If it can’t, you’ll hear that too, and what would.

   What you give. 30 minutes, and the decision.

   What you get. A yes or a no. If it’s yes, a scope (a comprehensive engagement across two practice areas or markets, or a focused one on one) and a fixed fee set by that scope.

2. The briefing document Before the kickoff

   What happens. A short set of questions: who you compete against, the wins and losses you’ve noticed, how you describe the firm now and what you need the study to settle.

   What you give. The completed briefing (about an hour). Win–loss notes or client feedback help if you have them, with client names and anything confidential taken out. You don’t need them.

   What you get. A kickoff spent on decisions, not on bringing us up to speed.

3. The kickoff workshop Week 1 · 90 minutes

   What happens. Together we map who decides when a firm like yours is chosen: which roles, which sectors, what size of organisation, what’s changing in their business. Then we agree the hypotheses: the things it would make a difference to your business to know. The Client Proxies vary on the things that might change how a client decides, so the study can show which differences look worth testing.

   What you give. The right people in the room, usually your marketing or BD lead and a partner who knows the competition. Straight answers about where you win and where you lose.

   What you get. The study design: who the Client Proxies are, and the hypotheses they’ll be tested against.

4. Agreeing the questions Week 1, by email

   What happens. After the kickoff, the hypotheses become questions. You get them by email and work through them with us until each one asks what you need answered, in words your clients would use.

   What you give. A read of the draft questions, and your changes.

   What you get. The final questions, agreed before a single Client Proxy answers them.

5. Building the Client Proxies Weeks 1 to 2

   What happens. Hundreds of Client Proxies are built across the roles agreed at the kickoff. Each one carries a role, a sector, an organisation size and what’s changing in its business, not just a job title.

   What you give. Nothing.

   What you get. A panel built to show where your market agrees, and where it may split.

6. Asking the questions, on more than one AI model Weeks 2 to 3

   What happens. Synthetic research has known limitations that will pollute your results if they aren’t filtered out. So every question runs on more than one AI model, through many processes, to make sure what reaches you is signal, not noise.

   What you give. Nothing.

   What you get. Answers in words, with the reasoning behind each one.

7. Analysis Weeks 3 to 4

   What happens. The answers become findings: what makes a client choose a firm or rule one out, the order they rank it, the words they use, how they react to the messages you’re considering and which differences between clients look worth testing.

   What you give. Nothing.

   What you get. The report, and the materials for your partners.

8. The delivery workshop and handover Week 4 · two hours

   What happens. A working session, not a slide show. Where the findings confirm what you believed, where they contradict it and what that means for your pitches, proposals and positioning. Then everything is handed over.

   What you give. The same people as the kickoff, plus whoever will act on the findings.

   What you get. Everything in the next section. And if your partners want to hear the findings from us, Steven Lewis presents them.

What do you get at the end of an ASI study?

Four things. The study ends with the last of them.

• The findings. What makes clients in your market choose a firm or rule one out, in the order they rank it. The reasons they give. The words they use, set against the words you use. How the people you want to win react to the messages you’re considering. And the differences between clients worth testing next.
• The delivery workshop, with your team.
• Materials for your partners: the findings in a form you can put in front of the partnership. If the partners would rather hear it in person, Steven Lewis presents it to them.
• The handover: the report, the partner materials and an AI tool your team can use to question the research, so the findings shape decisions every day instead of languishing in SharePoint. All of it is yours to keep.

What happens after the study?

The study is complete at handover, and everything in it is yours. What comes next depends on what the findings tell you. Firms have gone two ways from here, and the first conversation after delivery is about which, if either, fits.

Why hundreds of Client Proxies, and where does the method break?

Ask one AI model what a general counsel thinks and you get one confident answer, from one set of built-in assumptions, with nothing to check it against. In one of our studies, two leading AI models gave opposite answers to the same central positioning question.

So an ASI study varies the client across hundreds of Client Proxies and runs every question on more than one model. You see where the market agrees, and where it may split. Where the models split, the answer says more about the model than the market, so it stays out of the report.

Running several models has a limit, and you should know it. Agreement between models isn’t proof. Models trained on similar data can share a blind spot. Running several catches one model’s bias, not a bias they share.

Three kinds of finding, and how far to trust each

1. How many Client Proxies held a view. Read it as a strength of signal: how hard the market leans. It isn’t a count of real firms, so the report gives a whole number and how strong the signal is.
2. Differences between groups. Whether, say, general counsel at private-equity-backed companies weigh fees differently from those at family-owned ones. These are hypotheses. They’re the first things to test in your own client conversations.
3. The reasoning and the words. The reasons clients give, the objections they raise, the language they use. This is the solid part. It doesn’t depend on the Client Proxies matching your market one for one, and you can use it in a pitch the week you get it.

Publishing the limits is deliberate. A vendor who can’t tell you where their method breaks hasn’t found out yet. For the questions to put to any synthetic research vendor, us included, read how professional services firms should evaluate synthetic research.

What does a finished study look like?

Client studies stay confidential, so we can’t show you one. Goliath is the study we can share: our own, published so firms can see the method’s output before they commission anything. It put questions to 334 Client Proxies modelled on Australian corporate legal clients, fielded in July 2026, about how they choose between the firm they have and a challenger. Every firm is the incumbent on some work and the challenger on other work, so the findings read both ways.

Here is the reasoning level at work. Asked why they keep high-stakes work where it is, the Client Proxies talked about their own exposure. One, a general counsel at a mid-sized organisation, put it like this:

“Some of that credibility comes from being reliable and not creating surprises. If I switch firms on a high-profile matter and anything goes even slightly wrong, I’ve created a problem for myself that didn’t need to exist.”

A firm that knows clients are protecting their own careers pitches safety, not brilliance. The report is written to the three rules above: strength of signal for counts, hypotheses for differences and the reasoning in clients’ own words. Read it before you talk to us, and judge the method by its output.

Read the Goliath report

What an ASI study does NOT need from you

• No access to your clients. No interviews to arrange, and no relationship put at risk.
• No weeks of partner time. A partner at the two workshops helps. Nothing between them.
• No existing research. Win–loss data and client feedback help. Without them, the study builds from the market.
• No long scoping process. One call sets the scope and the fee.
• No software to learn. You receive intelligence, not a platform to run.

How does ASI fit our AI policy?

ASI runs one way: intelligence comes into your firm. The study needs nothing your AI policy protects. The questions are about your market and how clients in it choose firms, not about your firm’s confidential information or your clients’ matters.

What you give us may end up in an AI model, as context for the study. That’s why the list above is all we ask for. We sign your NDA before we start. If your firm has its own requirements, raise them on the first call.