Not Every Customer Is Worth Having

01 /The Concept

Law firms spend real money on intake staff whose job is partly to field calls that were never going to become cases. Dench cuts through that waste – an AI receptionist for law firms that answers every call and simultaneously evaluates whether the caller's situation is worth pursuing.

The AI receptionist is trained across a broad range of legal practice areas: personal injury, family law, criminal defense, employment disputes, immigration, estate planning, probate, and personal bankruptcy.

It picks up immediately – including outside business hours – and gets through the intake process faster than a human receptionist while still capturing every relevant detail. For a family law matter, for example, a human receptionist averages about 45 minutes on the initial call. Dench does it in 8.

The system handles conversations in over 20 languages, which is particularly valuable for immigration practices where clients often aren't comfortable in English. It also carries enough substantive knowledge to work through case-specific questions on the fly – like evaluating which visa category might apply to a given situation. That assessment takes Dench around 5 minutes, compared to roughly half an hour for a human.

A particularly interesting feature: all Dench AI receptionists across different firms are networked together. If a caller doesn't fit the profile of the firm they called, Dench can route the call to another firm in the network that might be a better match – and if that firm takes the case, the originating firm earns a referral fee of up to $500 per client. Each firm in the network gains the potential to multiply its effective inbound volume by up to 5x through this referral flow.

A human receptionist works 40 hours a week, handles one call at a time, requires training, and costs around $40,000/year in base salary before benefits and overhead. Dench works 24/7, handles unlimited simultaneous calls, requires no training, and bills at $0.99 per minute of call time.

Dench is currently in Y Combinator's accelerator program and published its launch on the YC site three days ago. Despite being early stage, it has already raised $1M, with participation from investors beyond Y Combinator.

02 /Why It Matters

What makes Dench possible is that AI has gotten genuinely good at conducting conversations – not just in text, but voice.

The clearest example is Boardy ([related review](/review/produkt-kotoryj-sam-prinosit-investorov)), which uses an AI interviewer as the front door to its community of founders and investors. A new member leaves their number; Boardy calls, runs an intake conversation to learn who they are and what they care about, then identifies and makes introductions to relevant people in the network. The startup raised $3M at launch last fall.

What happened next was remarkable: in one Friday call, Boardy spoke with a partner at a venture fund, who was so impressed that over the weekend he convinced his partners to each take a call with Boardy as well. By Monday they'd decided to invest. The founder then asked Boardy to find other community members who might co-invest, and the result was $8M raised within three months of the previous round.

Equally important: AI capable of conducting substantive intake conversations can serve as a filter – separating high-potential clients from low-potential ones before a human ever gets involved. This keeps the firm's senior time focused on cases that are actually worth taking.

A [recent review](/review/jeto-cinichno-no-pojetomu-jeffektivno) covered Justpoint, a service matching plaintiffs with attorneys for cases on contingency. The startup claims clients who find lawyers through their platform receive settlements three times higher than market average for comparable cases. The mechanism isn't magic: an AI engine evaluates incoming claims and routes only the strongest ones to attorneys – quietly declining the weaker ones under the guise of other reasons. Higher outcomes come from better case selection, not better lawyering per se.

Echo ([covered here](/review/kto-pomozhet-faunderu-prodavat-svoj-produkt)), a recent Y Combinator graduate, made a related tool for B2B founders: an AI assistant that helps them ask the right qualification questions early in a sales conversation, so they can stop investing time in prospects who were never going to buy.

The universal principle at work: garbage in, garbage out. No AI and no team can turn a fundamentally weak client relationship into a profitable one. The highest-leverage intervention is before commitment, at the filter stage.

03 /Opportunities

The direction here is building AI platforms for inbound lead handling – but with an important inversion. Most such platforms try to convert as many leads as possible. The emerging category is the opposite: filter as aggressively as possible.

The goal is to avoid handing leads to human sales or service teams unless they meet a meaningful threshold – because even a converted customer who churns quickly may cost more to serve than they generate in revenue.

This filtering logic can also be applied earlier, at the top of the funnel. Voyantis ([related review](/review/dolgo-i-mnogo-zarabatyvat-mozhno-tolko-na-teh-kto-dolgo-i-mnogo-platit)) raised $60M – $41M in a round closed two weeks ago – for a platform that predicts future customer LTV even at the ad-targeting stage, shifting spend toward prospects who will become high-value long-term customers rather than maximizing raw lead volume.

Generic filtering platforms can be horizontal, but the platforms that evaluate leads in detail almost certainly need to be vertical – built for the specific knowledge domain of the industry. Accurately assessing a personal injury case, an immigration matter, or a SaaS sales opportunity requires different subject-matter models. One generic platform can't do all of this well.

Which means there's a meaningful opportunity in each niche and even each geography, since the specifics of what makes a "good client" vary by market. The niches aren't claimed yet. That changes.